aid stringlengths 9 15 | mid stringlengths 7 10 | abstract stringlengths 78 2.56k | related_work stringlengths 92 1.77k | ref_abstract dict |
|---|---|---|---|---|
1905.11353 | 2947080931 | How to optimally dispatch orders to vehicles and how to trade off between immediate and future returns are fundamental questions for a typical ride-hailing platform. We model ride-hailing as a large-scale parallel ranking problem and study the joint decision-making the task of order dispatching and fleet management in online ride-hailing platforms. This task brings unique challenges in the four aspects. First, to facilitate a huge number of vehicles to act and learn efficiently and robustly, we treat each region cell as an agent and build a multi-agent reinforcement learning framework. Second, to coordinate the agents to achieve long-term benefits, we leverage the geographical hierarchy of the region grids to perform hierarchical reinforcement learning. Third, to deal with the heterogeneous and variant action space for joint order dispatching and fleet management, we design the action as the ranking weight vector to rank and select the specific order or the fleet management destination in a unified formulation. Fourth, to achieve the multi-scale ride-hailing platform, we conduct the decision-making process in a hierarchical way where multi-head attention mechanism is utilized to incorporate the impacts of neighbor agents and capture the key agent in each scale. The whole novel framework is named as CoRide. Extensive experiments based on multiple cities real-world data as well as analytic synthetic data demonstrate that CoRide provides superior performance in terms of platform revenue and user experience in the task of city-wide hybrid order dispatching and fleet management over strong baselines. This work provides not only a solution for current online ride-hailing platforms, but also an advanced artificial intelligent technique for future life especially when large scale unmanned ground vehicles going into service. | The options framework @cite_42 @cite_9 @cite_21 formulates the problem with a two-level hierarchy, where the low-level - option - is a sub-policy with a termination condition. Since traditional options framework suffers from prior knowledge on designing options, jointly learning high-level policy with low-level policy has been proposed by @cite_13 . However, this actor-critic HRL approach needs to either learning sub-policies for each time step or one sub-policy for the whole episode. Therefore, the performance of the whole module often prone to learning useful sub-policies. To guarantee gaining effective sub-policies, one alternative direction is to provide auxiliary rewards for low-level policy: hand-designed rewards based on prior domain knowledge @cite_29 @cite_31 or mutual information @cite_16 @cite_2 @cite_23 . Given having access to one well-designed and suitable reward is often a luxury, proposed FeUdal Networks (FuN), where a generic reward is utilized for low-level policy learning, thus avoid designing hand-craft rewards. Several works extend and improve FuN with off-policy training @cite_3 , form of hindsight @cite_39 and representation learning @cite_25 . | {
"cite_N": [
"@cite_29",
"@cite_9",
"@cite_42",
"@cite_21",
"@cite_3",
"@cite_39",
"@cite_23",
"@cite_2",
"@cite_31",
"@cite_16",
"@cite_13",
"@cite_25"
],
"mid": [
"2963262099",
"1585861384",
"1556824961",
"2109910161",
"2950614095",
"2920215304",
"1585855589",
"2111967991",
"2964118262",
"2606433045",
"",
"2894605519"
],
"abstract": [
"Learning goal-directed behavior in environments with sparse feedback is a major challenge for reinforcement learning algorithms. One of the key difficulties is insufficient exploration, resulting in an agent being unable to learn robust policies. Intrinsically motivated agents can explore new behavior for their own sake rather than to directly solve external goals. Such intrinsic behaviors could eventually help the agent solve tasks posed by the environment. We present hierarchical-DQN (h-DQN), a framework to integrate hierarchical action-value functions, operating at different temporal scales, with goal-driven intrinsically motivated deep reinforcement learning. A top-level q-value function learns a policy over intrinsic goals, while a lower-level function learns a policy over atomic actions to satisfy the given goals. h-DQN allows for flexible goal specifications, such as functions over entities and relations. This provides an efficient space for exploration in complicated environments. We demonstrate the strength of our approach on two problems with very sparse and delayed feedback: (1) a complex discrete stochastic decision process with stochastic transitions, and (2) the classic ATARI game -'Montezuma's Revenge'.",
"Decision making usually involves choosing among different courses of action over a broad range of time scales. For instance, a person planning a trip to a distant location makes high-level decisions regarding what means of transportation to use, but also chooses low-level actions, such as the movements for getting into a car. The problem of picking an appropriate time scale for reasoning and learning has been explored in artificial intelligence, control theory and robotics. In this dissertation we develop a framework that allows novel solutions to this problem, in the context of Markov Decision Processes (MDPs) and reinforcement learning. In this dissertation, we present a general framework for prediction, control and learning at multiple temporal scales. In this framework, temporally extended actions are represented by a way of behaving (a policy) together with a termination condition. An action represented in this way is called an option. Options can be easily incorporated in MDPs, allowing an agent to use existing controllers, heuristics for picking actions, or learned courses of action. The effects of behaving according to an option can be predicted using multi-time models, learned by interacting with the environment. In this dissertation we develop multi-time models, and we illustrate the way in which they can be used to produce plans of behavior very quickly, using classical dynamic programming or reinforcement learning techniques. The most interesting feature of our framework is that it allows an agent to work simultaneously with high-level and low-level temporal representations. The interplay of these levels can be exploited in order to learn and plan more efficiently and more accurately. We develop new algorithms that take advantage of this structure to improve the quality of plans, and to learn in parallel about the effects of many different options.",
"Temporally extended actions (e.g., macro actions) have proven very useful for speeding up learning, ensuring robustness and building prior knowledge into AI systems. The options framework (Precup, 2000; Sutton, Precup & Singh, 1999) provides a natural way of incorporating such actions into reinforcement learning systems, but leaves open the issue of howgood options might be identified. In this paper, we empirically explore a simple approach to creating options. The underlying assumption is that the agent will be asked to perform different goalachievement tasks in an environment that is otherwise the same over time. Our approach is based on the intuition that states that are frequently visited on system trajectories, could prove to be useful subgoals (e.g., McGovern & Barto, 2001; Iba, 1989).We propose a greedy algorithm for identifying subgoals based on state visitation counts. We present empirical studies of this approach in two gridworld navigation tasks. One of the environments we explored contains bottleneck states, and the algorithm indeed finds these states, as expected. The second environment is an empty gridworld with no obstacles. Although the environment does not contain any obvious subgoals, our approach still finds useful options, which essentially allow the agent to explore the environment more quickly.",
"Learning, planning, and representing knowledge at multiple levels of temporal ab- straction are key, longstanding challenges for AI. In this paper we consider how these challenges can be addressed within the mathematical framework of reinforce- ment learning and Markov decision processes (MDPs). We extend the usual notion of action in this framework to include options—closed-loop policies for taking ac- tion over a period of time. Examples of options include picking up an object, going to lunch, and traveling to a distant city, as well as primitive actions such as mus- cle twitches and joint torques. Overall, we show that options enable temporally abstract knowledge and action to be included in the reinforcement learning frame- work in a natural and general way. In particular, we show that options may be used interchangeably with primitive actions in planning methods such as dynamic pro- gramming and in learning methods such as Q-learning. Formally, a set of options defined over an MDP constitutes a semi-Markov decision process (SMDP), and the theory of SMDPs provides the foundation for the theory of options. However, the most interesting issues concern the interplay between the underlying MDP and the SMDP and are thus beyond SMDP theory. We present results for three such cases: 1) we show that the results of planning with options can be used during execution to interrupt options and thereby perform even better than planned, 2) we introduce new intra-option methods that are able to learn about an option from fragments of its execution, and 3) we propose a notion of subgoal that can be used to improve the options themselves. All of these results have precursors in the existing literature; the contribution of this paper is to establish them in a simpler and more general setting with fewer changes to the existing reinforcement learning framework. In particular, we show that these results can be obtained without committing to (or ruling out) any particular approach to state abstraction, hierarchy, function approximation, or the macro-utility problem.",
"Hierarchical reinforcement learning (HRL) is a promising approach to extend traditional reinforcement learning (RL) methods to solve more complex tasks. Yet, the majority of current HRL methods require careful task-specific design and on-policy training, making them difficult to apply in real-world scenarios. In this paper, we study how we can develop HRL algorithms that are general, in that they do not make onerous additional assumptions beyond standard RL algorithms, and efficient, in the sense that they can be used with modest numbers of interaction samples, making them suitable for real-world problems such as robotic control. For generality, we develop a scheme where lower-level controllers are supervised with goals that are learned and proposed automatically by the higher-level controllers. To address efficiency, we propose to use off-policy experience for both higher and lower-level training. This poses a considerable challenge, since changes to the lower-level behaviors change the action space for the higher-level policy, and we introduce an off-policy correction to remedy this challenge. This allows us to take advantage of recent advances in off-policy model-free RL to learn both higher- and lower-level policies using substantially fewer environment interactions than on-policy algorithms. We term the resulting HRL agent HIRO and find that it is generally applicable and highly sample-efficient. Our experiments show that HIRO can be used to learn highly complex behaviors for simulated robots, such as pushing objects and utilizing them to reach target locations, learning from only a few million samples, equivalent to a few days of real-time interaction. In comparisons with a number of prior HRL methods, we find that our approach substantially outperforms previous state-of-the-art techniques.",
"",
"This paper designs and implements a taxi telematics service system, aiming at providing an efficient framework by means of a Linux cluster to host emerging telematics services that need intensive computing. Combined with global positioning system and radio communication technology, the taxi telematics service system traces the position of taxis, finds a time saving route between start and destination points, dispatches the nearest taxi to the service call point based on the latest traffic information, and finally decides an efficient route for multiple destinations. The performance measurement result demonstrates that the implemented system can process up to 200 map matches for every minute, keeping average response time for other requests below 1.5 seconds.",
"Many reinforcement learning (RL) tasks, especially in robotics, consist of multiple sub-tasks that are strongly structured. Such task structures can be exploited by incorporating hierarchical policies that consist of gating networks and sub-policies. However, this concept has only been partially explored for real world settings and complete methods, derived from first principles, are needed. Real world settings are challenging due to large and continuous state-action spaces that are prohibitive for exhaustive sampling methods. We define the problem of learning sub-policies in continuous state action spaces as finding a hierarchical policy that is composed of a high-level gating policy to select the low-level sub-policies for execution by the agent. In order to efficiently share experience with all sub-policies, also called inter-policy learning, we treat these sub-policies as latent variables which allows for distribution of the update information between the sub-policies. We present three different variants of our algorithm, designed to be suitable for a wide variety of real world robot learning tasks and evaluate our algorithms in two real robot learning scenarios as well as several simulations and comparisons.",
"",
"Deep reinforcement learning has achieved many impressive results in recent years. However, tasks with sparse rewards or long horizons continue to pose significant challenges. To tackle these important problems, we propose a general framework that first learns useful skills in a pre-training environment, and then leverages the acquired skills for learning faster in downstream tasks. Our approach brings together some of the strengths of intrinsic motivation and hierarchical methods: the learning of useful skill is guided by a single proxy reward, the design of which requires very minimal domain knowledge about the downstream tasks. Then a high-level policy is trained on top of these skills, providing a significant improvement of the exploration and allowing to tackle sparse rewards in the downstream tasks. To efficiently pre-train a large span of skills, we use Stochastic Neural Networks combined with an information-theoretic regularizer. Our experiments show that this combination is effective in learning a wide span of interpretable skills in a sample-efficient way, and can significantly boost the learning performance uniformly across a wide range of downstream tasks.",
"",
"We study the problem of representation learning in goal-conditioned hierarchical reinforcement learning. In such hierarchical structures, a higher-level controller solves tasks by iteratively communicating goals which a lower-level policy is trained to reach. Accordingly, the choice of representation -- the mapping of observation space to goal space -- is crucial. To study this problem, we develop a notion of sub-optimality of a representation, defined in terms of expected reward of the optimal hierarchical policy using this representation. We derive expressions which bound the sub-optimality and show how these expressions can be translated to representation learning objectives which may be optimized in practice. Results on a number of difficult continuous-control tasks show that our approach to representation learning yields qualitatively better representations as well as quantitatively better hierarchical policies, compared to existing methods (see videos at this https URL)."
]
} |
1905.11559 | 2947253871 | Robust road segmentation is a key challenge in self-driving research. Though many image-based methods have been studied and high performances in dataset evaluations have been reported, developing robust and reliable road segmentation is still a major challenge. Data fusion across different sensors to improve the performance of road segmentation is widely considered an important and irreplaceable solution. In this paper, we propose a novel structure to fuse image and LiDAR point cloud in an end-to-end semantic segmentation network, in which the fusion is performed at decoder stage instead of at, more commonly, encoder stage. During fusion, we improve the multi-scale LiDAR map generation to increase the precision of the multi-scale LiDAR map by introducing pyramid projection method. Additionally, we adapted the multi-path refinement network with our fusion strategy and improve the road prediction compared with transpose convolution with skip layers. Our approach has been tested on KITTI ROAD dataset and has competitive performance. | Computer vision and pattern recognition research cover a lot of fields, and the a an important topic is feature extraction and analysis @cite_26 @cite_4 @cite_38 @cite_42 @cite_0 @cite_17 @cite_27 . In the past decade, deep learning with CNN becomes a very important feature extractor for classification and segmentation problems @cite_14 @cite_20 @cite_18 . Since road segmentation is a semantic segmentation task, the following subsection is the traditional ideas on road segmentation. Then the subsequent subsection reviews deep neural network based semantic segmentation as our work are inspired by many recent progress in new architecture in CNN classifier and semantic segmentation methods. Finally, data fusion methods in previous works is listed. | {
"cite_N": [
"@cite_38",
"@cite_18",
"@cite_26",
"@cite_4",
"@cite_14",
"@cite_42",
"@cite_0",
"@cite_27",
"@cite_20",
"@cite_17"
],
"mid": [
"2464929676",
"2524029660",
"2779475425",
"1992738091",
"2552684688",
"2095176685",
"2402601457",
"2791528017",
"2805903690",
"2762668439"
],
"abstract": [
"In this paper, a novel group sparse canonical correlation analysis (GSCCA) method is proposed for simultaneous electroencephalogram (EEG) channel selection and emotion recognition. GSCCA is a group sparse extension of the conventional CCA method to model the linear correlationship between emotional EEG class label vectors and the corresponding EEG feature vectors. In contrast to conventional CCA method or previous GSCCA methods, a major advantage of our GSCCA method is the ability of handling the group feature selection problem from raw EEG features, which makes it very suitable for simultaneously coping with both EEG emotion recognition and automatic channel selection issues where each EEG channel is associated with a group of raw EEG features. To deal with EEG emotion recognition problem, we adopt the popularly used frequency feature to describe the EEG signal by dividing the full EEG frequency band into five parts, i.e., @math , @math , @math , @math , and @math frequency bands, and then extract the frequency band features from each band for GSCCA model learning and emotion recognition. Finally, we conduct extensive experiments on EEG-based emotion recognition based on the SJTU emotion EEG dataset and experimental results demonstrate that the proposed GSCCA method would outperform the state-of-the-art EEG-based emotion recognition approaches.",
"There have been increasing research interests in automatically constructing image dataset by collecting images from the Internet. However, existing methods tend to have a weak domain adaptation ability, known as the \"dataset bias problem\". To address this issue, in this work, we propose a novel image dataset construction framework which can generalize well to unseen target domains. In specific, the given queries are first expanded by searching in the Google Books Ngrams Corpora (GBNC) to obtain a richer semantic description, from which the noisy query expansions are then filtered out. By treating each expansion as a \"bag\" and the retrieved images therein as \"instances\", we formulate image filtering as a multi-instance learning (MIL) problem with constrained positive bags. By this approach, images from different data distributions will be kept while with noisy images filtered out. Comprehensive experiments on two challenging tasks demonstrate the effectiveness of our proposed approach.",
"In the traditional graph embedding framework, the graph is usually built by k-NN or r-ball. Since it is difficult to manually set the parameters k and r in the high-dimensional space, sparse representation-based methods are usually introduced to automatically build the graphs. In recent years, nuclear norm-based matrix regression (NMR) has been proposed for face recognition using the low rank structural information (i.e., the image matrix-based error model). Inspired by NMR, we give a NMR-based projections (NMRP) method for feature extraction and recognition. The experiments on FERET and extended Yale B face databases show that NMR can be used to build the graph while NMRP is an effective feature extraction method.",
"In graph embedding based methods, we usually need to manually choose the nearest neighbors and then compute the edge weights using the nearest neighbors via L2 norm (e.g. LLE). It is difficult and unstable to manually choose the nearest neighbors in high dimensional space. So how to automatically construct a graph is very important. In this paper, first, we give a L2-graph like L1-graph. L2-graph calculates the edge weights using the total samples, avoiding manually choosing the nearest neighbors; second, a L2-graph based feature extraction method is presented, called collaborative representation based projections (CRP). Like SPP, CRP aims to preserve the collaborative representation based reconstruction relationship of data. CRP utilizes a L2 norm graph to characterize the local compactness information. CRP maximizes the ratio between the total separability information and the local compactness information to seek the optimal projection matrix. CRP is much faster than SPP since CRP calculates the objective function with L2 norm while SPP calculate the objective function with L1 norm. Experimental results on FERET, AR, Yale face databases and the PolyU finger-knuckle-print database demonstrate that CRP works well in feature extraction and leads to a good recognition performance. We give a L2 norm graph based on collaborative representation.We propose a collaborative representation based projections (CRP) for feature extraction.CRP is a Rayleigh quotient form and can be calculated via generalized eigenvalue decomposition.",
"The goal of this work is to automatically collect a large number of highly relevant natural images from Internet for given queries. A novel automatic image dataset construction framework is proposed by employing multiple query expansions. In specific, the given queries are first expanded by searching in the Google Books Ngrams Corpora to obtain a richer semantic descriptions, from which the visually non-salient and less relevant expansions are then filtered. After retrieving images from the Internet with filtered expansions, we further filter noisy images by clustering and progressively Convolutional Neural Networks (CNN) based methods. To evaluate the performance of our proposed method for image dataset construction, we build an image dataset with 10 categories. We then run object detections on our image dataset with three other image datasets which were constructed by weak supervised, web supervised and full supervised learning, the experimental results indicated the effectiveness of our method is superior to weak supervised and web supervised state-of-the-art methods. In addition, we do a cross-dataset classification to evaluate the performance of our dataset with two publically available manual labelled dataset STL-10 and CIFAR-10.",
"Sparse representation based classification (SRC) has received much attention in computer vision and pattern recognition. SRC codes a testing sample by sparse linear combination of all the training samples and classifies the testing sample into the class with the minimum representation error. Recently, Zhang analyzes the working mechanism of SRC and points out that it is the collaborative representation but not the L1-norm sparsity that makes SRC powerful. Based on the analysis, they propose a very simple and much more efficient classification scheme, called collaborative representation based classification with regularized least square (CRC_RLS). CRC_RLS is a linear method in nature. Here we propose a kernel collaborative representation based classification with regularized least square (Kernel CRC_RLS, KCRC_RLS) by implicitly mapping the sample into high-dimensional space via kernel tricks. Our approach is highly motivated by the kernel methods which can capture the nonlinear similarity among samples and have been successfully applied in pattern recognition and machine learning. The experimental results on the CENPAMI handwritten digital database, ETH80 database, FERET face database, ORL database, AR face database, demonstrate that Kernel CRC_RLS is effective in classification, leading to promising performance.",
"Recent successes in visual recognition can be primarily attributed to feature representation, learning algorithms, and the ever-increasing size of labeled training data. Extensive research has been devoted to the first two, but much less attention has been paid to the third. Due to the high cost of manual data labeling, the size of recent efforts such as ImageNet is still relatively small in respect to daily applications. In this work, we mainly focus on how to automatically generate identifying image data for a given visual concept on a vast scale. With the generated image data, we can train a robust recognition model for the given concept. We evaluate the proposed webly supervised approach on the benchmark Pascal VOC 2007 dataset and the results demonstrates the superiority of our method over many other state-of-the-art methods in image data collection.",
"Abstract This paper introduces a novel dimensionality reduction algorithm, called collaborative representation based local discriminant projection (CRLDP), for feature extraction. CRLDP utilizes collaborative representation relationships among samples to construct adjacency graphs. Different from most graph-based algorithms which manually construct the adjacency graphs, CRLDP is able to automatically construct the graphs and avoid manually choosing nearest neighbors. In CRLDP, two graphs (the within-class graph and the between-class graph) are constructed. Based on the two constructed graphs, the within-class scatter and the between-class scatter are computed to characterize the compactness and separability of samples, respectively. Then CRLDP seeks to find an optimal projection matrix to maximize the ratio of the between-class scatter to the within-class scatter. Experimental results on ORL, AR and CMU PIE face databases validate the superiority of CRLDP over other state-of-the-art algorithms.",
"Abstract Studies show that refining real-world categories into semantic subcategories contributes to better image modeling and classification. Previous image sub-categorization work relying on labeled images and WordNet’s hierarchy is labor-intensive. To tackle this problem, in this work, we extract textual and visual features to automatically select and subsequently classify web images into semantic rich categories. The following two major challenges are well studied: (1) noise in the labels of subcategories derived from the general corpus; (2) noise in the labels of images retrieved from the web. Specifically, we first obtain the semantic refinement subcategories from the text perspective and remove the noise by using the relevance-based approach. To suppress the search error induced noisy images, we then formulate image selection and classifier learning as a multi-instance learning problem and propose to solve the employed problem by the cutting-plane algorithm. The experiments show significant performance gains by using the generated data of our approach on image categorization tasks. The proposed approach also consistently outperforms existing weakly supervised and web-supervised approaches.",
"Due to the variations of viewpoint, pose, and illumination, a given individual may appear considerably different across different camera views. Tracking individuals across camera networks with no overlapping fields is still a challenging problem. Previous works mainly focus on feature representation and metric learning individually which tend to have a suboptimal solution. To address this issue, in this work, we propose a novel framework to do the feature representation learning and metric learning jointly. Different from previous works, we represent the pairs of pedestrian images as new resized input and use linear Support Vector Machine to replace softmax activation function for similarity learning. Particularly, dropout and data augmentation techniques are also employed in this model to prevent the network from overfitting. Extensive experiments on two publically available datasets VIPeR and CUHK01 demonstrate the effectiveness of our proposed approach."
]
} |
1905.11559 | 2947253871 | Robust road segmentation is a key challenge in self-driving research. Though many image-based methods have been studied and high performances in dataset evaluations have been reported, developing robust and reliable road segmentation is still a major challenge. Data fusion across different sensors to improve the performance of road segmentation is widely considered an important and irreplaceable solution. In this paper, we propose a novel structure to fuse image and LiDAR point cloud in an end-to-end semantic segmentation network, in which the fusion is performed at decoder stage instead of at, more commonly, encoder stage. During fusion, we improve the multi-scale LiDAR map generation to increase the precision of the multi-scale LiDAR map by introducing pyramid projection method. Additionally, we adapted the multi-path refinement network with our fusion strategy and improve the road prediction compared with transpose convolution with skip layers. Our approach has been tested on KITTI ROAD dataset and has competitive performance. | . FCN is the first well-known method for end-to-end deep semantic segmentation @cite_13 . FCN's designation follows the encoder-decoder pattern with transposed convolutions and skip layers, this architecture laid the foundation for segmentations. In the meanwhile, SegNet @cite_28 use max-pooling indices in the decoders to perform upsampling of low resolution feature maps which retains high frequency details in the segmented images as well as reduces the total number of trainable parameters in the decoders. To make strong use of data augmentation of available annotated samples, UNet @cite_50 develops architecture consists of a contracting path to capture context and a symmetric expanding path that enables precise localization. DeepLab series now evolve to DeepLabV3+ @cite_6 , it extends DeepLabv3 by adding a simple yet effective decoder module to refine the segmentation results especially along object boundaries. Some work achieve very good result by combining FCN with Conditional Random Field(CRF) using recurrent net @cite_23 @cite_10 . | {
"cite_N": [
"@cite_28",
"@cite_10",
"@cite_6",
"@cite_23",
"@cite_50",
"@cite_13"
],
"mid": [
"2963881378",
"2124592697",
"2964309882",
"2964288706",
"1901129140",
"1903029394"
],
"abstract": [
"We present a novel and practical deep fully convolutional neural network architecture for semantic pixel-wise segmentation termed SegNet. This core trainable segmentation engine consists of an encoder network, a corresponding decoder network followed by a pixel-wise classification layer. The architecture of the encoder network is topologically identical to the 13 convolutional layers in the VGG16 network [1] . The role of the decoder network is to map the low resolution encoder feature maps to full input resolution feature maps for pixel-wise classification. The novelty of SegNet lies is in the manner in which the decoder upsamples its lower resolution input feature map(s). Specifically, the decoder uses pooling indices computed in the max-pooling step of the corresponding encoder to perform non-linear upsampling. This eliminates the need for learning to upsample. The upsampled maps are sparse and are then convolved with trainable filters to produce dense feature maps. We compare our proposed architecture with the widely adopted FCN [2] and also with the well known DeepLab-LargeFOV [3] , DeconvNet [4] architectures. This comparison reveals the memory versus accuracy trade-off involved in achieving good segmentation performance. SegNet was primarily motivated by scene understanding applications. Hence, it is designed to be efficient both in terms of memory and computational time during inference. It is also significantly smaller in the number of trainable parameters than other competing architectures and can be trained end-to-end using stochastic gradient descent. We also performed a controlled benchmark of SegNet and other architectures on both road scenes and SUN RGB-D indoor scene segmentation tasks. These quantitative assessments show that SegNet provides good performance with competitive inference time and most efficient inference memory-wise as compared to other architectures. We also provide a Caffe implementation of SegNet and a web demo at http: mi.eng.cam.ac.uk projects segnet .",
"Pixel-level labelling tasks, such as semantic segmentation, play a central role in image understanding. Recent approaches have attempted to harness the capabilities of deep learning techniques for image recognition to tackle pixel-level labelling tasks. One central issue in this methodology is the limited capacity of deep learning techniques to delineate visual objects. To solve this problem, we introduce a new form of convolutional neural network that combines the strengths of Convolutional Neural Networks (CNNs) and Conditional Random Fields (CRFs)-based probabilistic graphical modelling. To this end, we formulate Conditional Random Fields with Gaussian pairwise potentials and mean-field approximate inference as Recurrent Neural Networks. This network, called CRF-RNN, is then plugged in as a part of a CNN to obtain a deep network that has desirable properties of both CNNs and CRFs. Importantly, our system fully integrates CRF modelling with CNNs, making it possible to train the whole deep network end-to-end with the usual back-propagation algorithm, avoiding offline post-processing methods for object delineation. We apply the proposed method to the problem of semantic image segmentation, obtaining top results on the challenging Pascal VOC 2012 segmentation benchmark.",
"Spatial pyramid pooling module or encode-decoder structure are used in deep neural networks for semantic segmentation task. The former networks are able to encode multi-scale contextual information by probing the incoming features with filters or pooling operations at multiple rates and multiple effective fields-of-view, while the latter networks can capture sharper object boundaries by gradually recovering the spatial information. In this work, we propose to combine the advantages from both methods. Specifically, our proposed model, DeepLabv3+, extends DeepLabv3 by adding a simple yet effective decoder module to refine the segmentation results especially along object boundaries. We further explore the Xception model and apply the depthwise separable convolution to both Atrous Spatial Pyramid Pooling and decoder modules, resulting in a faster and stronger encoder-decoder network. We demonstrate the effectiveness of the proposed model on PASCAL VOC 2012 and Cityscapes datasets, achieving the test set performance of 89 and 82.1 without any post-processing. Our paper is accompanied with a publicly available reference implementation of the proposed models in Tensorflow at https: github.com tensorflow models tree master research deeplab.",
"Abstract: Deep Convolutional Neural Networks (DCNNs) have recently shown state of the art performance in high level vision tasks, such as image classification and object detection. This work brings together methods from DCNNs and probabilistic graphical models for addressing the task of pixel-level classification (also called \"semantic image segmentation\"). We show that responses at the final layer of DCNNs are not sufficiently localized for accurate object segmentation. This is due to the very invariance properties that make DCNNs good for high level tasks. We overcome this poor localization property of deep networks by combining the responses at the final DCNN layer with a fully connected Conditional Random Field (CRF). Qualitatively, our \"DeepLab\" system is able to localize segment boundaries at a level of accuracy which is beyond previous methods. Quantitatively, our method sets the new state-of-art at the PASCAL VOC-2012 semantic image segmentation task, reaching 71.6 IOU accuracy in the test set. We show how these results can be obtained efficiently: Careful network re-purposing and a novel application of the 'hole' algorithm from the wavelet community allow dense computation of neural net responses at 8 frames per second on a modern GPU.",
"There is large consent that successful training of deep networks requires many thousand annotated training samples. In this paper, we present a network and training strategy that relies on the strong use of data augmentation to use the available annotated samples more efficiently. The architecture consists of a contracting path to capture context and a symmetric expanding path that enables precise localization. We show that such a network can be trained end-to-end from very few images and outperforms the prior best method (a sliding-window convolutional network) on the ISBI challenge for segmentation of neuronal structures in electron microscopic stacks. Using the same network trained on transmitted light microscopy images (phase contrast and DIC) we won the ISBI cell tracking challenge 2015 in these categories by a large margin. Moreover, the network is fast. Segmentation of a 512x512 image takes less than a second on a recent GPU. The full implementation (based on Caffe) and the trained networks are available at http: lmb.informatik.uni-freiburg.de people ronneber u-net .",
"Convolutional networks are powerful visual models that yield hierarchies of features. We show that convolutional networks by themselves, trained end-to-end, pixels-to-pixels, exceed the state-of-the-art in semantic segmentation. Our key insight is to build “fully convolutional” networks that take input of arbitrary size and produce correspondingly-sized output with efficient inference and learning. We define and detail the space of fully convolutional networks, explain their application to spatially dense prediction tasks, and draw connections to prior models. We adapt contemporary classification networks (AlexNet [20], the VGG net [31], and GoogLeNet [32]) into fully convolutional networks and transfer their learned representations by fine-tuning [3] to the segmentation task. We then define a skip architecture that combines semantic information from a deep, coarse layer with appearance information from a shallow, fine layer to produce accurate and detailed segmentations. Our fully convolutional network achieves state-of-the-art segmentation of PASCAL VOC (20 relative improvement to 62.2 mean IU on 2012), NYUDv2, and SIFT Flow, while inference takes less than one fifth of a second for a typical image."
]
} |
1905.11559 | 2947253871 | Robust road segmentation is a key challenge in self-driving research. Though many image-based methods have been studied and high performances in dataset evaluations have been reported, developing robust and reliable road segmentation is still a major challenge. Data fusion across different sensors to improve the performance of road segmentation is widely considered an important and irreplaceable solution. In this paper, we propose a novel structure to fuse image and LiDAR point cloud in an end-to-end semantic segmentation network, in which the fusion is performed at decoder stage instead of at, more commonly, encoder stage. During fusion, we improve the multi-scale LiDAR map generation to increase the precision of the multi-scale LiDAR map by introducing pyramid projection method. Additionally, we adapted the multi-path refinement network with our fusion strategy and improve the road prediction compared with transpose convolution with skip layers. Our approach has been tested on KITTI ROAD dataset and has competitive performance. | . Guosheng Lin present a generic multi-path refinement network called RefineNet. RefineNet is a generic multi-path refinement network in which the information available along the down-sampling process are explicitly exploited to enable high-resolution prediction using long range residual connections @cite_16 . The most important component in RefineNet is Chained residual pooling(CRP). The key idea of RefineNet @cite_16 is deeper layers that capture high-level semantic features can be directly refined using fine-grained features from earlier convolutions. The experiments have proved that it achieved state-of-the-art performance in many semantic segmentation datasets. This network is elastic since there exists many configuration after encoder. Our work is based on the principle of RefineNet and a fusion module is added to CRP module. Figure. is an illustration of RefineNet. | {
"cite_N": [
"@cite_16"
],
"mid": [
"2563705555"
],
"abstract": [
"Recently, very deep convolutional neural networks (CNNs) have shown outstanding performance in object recognition and have also been the first choice for dense classification problems such as semantic segmentation. However, repeated subsampling operations like pooling or convolution striding in deep CNNs lead to a significant decrease in the initial image resolution. Here, we present RefineNet, a generic multi-path refinement network that explicitly exploits all the information available along the down-sampling process to enable high-resolution prediction using long-range residual connections. In this way, the deeper layers that capture high-level semantic features can be directly refined using fine-grained features from earlier convolutions. The individual components of RefineNet employ residual connections following the identity mapping mindset, which allows for effective end-to-end training. Further, we introduce chained residual pooling, which captures rich background context in an efficient manner. We carry out comprehensive experiments and set new state-of-the-art results on seven public datasets. In particular, we achieve an intersection-over-union score of 83.4 on the challenging PASCAL VOC 2012 dataset, which is the best reported result to date."
]
} |
1905.11559 | 2947253871 | Robust road segmentation is a key challenge in self-driving research. Though many image-based methods have been studied and high performances in dataset evaluations have been reported, developing robust and reliable road segmentation is still a major challenge. Data fusion across different sensors to improve the performance of road segmentation is widely considered an important and irreplaceable solution. In this paper, we propose a novel structure to fuse image and LiDAR point cloud in an end-to-end semantic segmentation network, in which the fusion is performed at decoder stage instead of at, more commonly, encoder stage. During fusion, we improve the multi-scale LiDAR map generation to increase the precision of the multi-scale LiDAR map by introducing pyramid projection method. Additionally, we adapted the multi-path refinement network with our fusion strategy and improve the road prediction compared with transpose convolution with skip layers. Our approach has been tested on KITTI ROAD dataset and has competitive performance. | for DNN based road segmentation. discussed different fusion strategies, named early fusion and late fusion @cite_15 . Last year, proposed a new fusion architecture name corss-fusion to directly learn from data where to integrate information by using trainable cross connections between the LIDAR and the camera processing branches @cite_33 . A siamese fusion method for road segmentation is also proposed to fuse LiDAR and camera informations @cite_37 . On road segmentation with camera and LiDAR fusion tasks, the only suitable dataset and benchmark, to our knowledge, is KITTI ROAD dataset @cite_46 . This dataset provide various sensor data collected on a real car with careful calibration. Due to above limitation, we will test our method using KITTI benchmark. | {
"cite_N": [
"@cite_37",
"@cite_15",
"@cite_46",
"@cite_33"
],
"mid": [
"2904858039",
"2416791088",
"2167222293",
"2891733872"
],
"abstract": [
"Robust road detection is a key challenge in safe autonomous driving. Recently, with the rapid development of 3D sensors, more and more researchers are trying to fuse information across different sensors to improve the performance of road detection. Although many successful works have been achieved in this field, methods for data fusion under deep learning framework is still an open problem. In this paper, we propose a Siamese deep neural network based on FCN-8s to detect road region. Our method uses data collected from a monocular color camera and a Velodyne-64 LiDAR sensor. We project the LiDAR point clouds onto the image plane to generate LiDAR images and feed them into one of the branches of the network. The RGB images are fed into another branch of our proposed network. The feature maps that these two branches extract in multiple scales are fused before each pooling layer, via padding additional fusion layers. Extensive experimental results on public dataset KITTI ROAD demonstrate the effectiveness of our proposed approach.",
"In this paper, we explore various aspects of fusing LIDAR and color imagery for pedestrian detection in the context of convolutional neural networks (CNNs), which have recently become state-of-art for many vision problems. We incorporate LIDAR by up-sampling the point cloud to a dense depth map and then extracting three features representing different aspects of the 3D scene. We then use those features as extra image channels. Specifically, we leverage recent work on HHA [9] (horizontal disparity, height above ground, and angle) representations, adapting the code to work on up-sampled LIDAR rather than Microsoft Kinect depth maps. We show, for the first time, that such a representation is applicable to up-sampled LIDAR data, despite its sparsity. Since CNNs learn a deep hierarchy of feature representations, we then explore the question: At what level of representation should we fuse this additional information with the original RGB image channels? We use the KITTI pedestrian detection dataset for our exploration. We first replicate the finding that region-CNNs (R-CNNs) [8] can outperform the original proposal mechanism using only RGB images, but only if fine-tuning is employed. Then, we show that: 1) using HHA features and RGB images performs better than RGB-only, even without any fine-tuning using large RGB web data, 2) fusing RGB and HHA achieves the strongest results if done late, but, under a parameter or computational budget, is best done at the early to middle layers of the hierarchical representation, which tend to represent midlevel features rather than low (e.g. edges) or high (e.g. object class decision) level features, 3) some of the less successful methods have the most parameters, indicating that increased classification accuracy is not simply a function of increased capacity in the neural network.",
"Detecting the road area and ego-lane ahead of a vehicle is central to modern driver assistance systems. While lane-detection on well-marked roads is already available in modern vehicles, finding the boundaries of unmarked or weakly marked roads and lanes as they appear in inner-city and rural environments remains an unsolved problem due to the high variability in scene layout and illumination conditions, amongst others. While recent years have witnessed great interest in this subject, to date no commonly agreed upon benchmark exists, rendering a fair comparison amongst methods difficult. In this paper, we introduce a novel open-access dataset and benchmark for road area and ego-lane detection. Our dataset comprises 600 annotated training and test images of high variability from the KITTI autonomous driving project, capturing a broad spectrum of urban road scenes. For evaluation, we propose to use the 2D Bird's Eye View (BEV) space as vehicle control usually happens in this 2D world, requiring detection results to be represented in this very same space. Furthermore, we propose a novel, behavior-based metric which judges the utility of the extracted ego-lane area for driver assistance applications by fitting a driving corridor to the road detection results in the BEV. We believe this to be important for a meaningful evaluation as pixel-level performance is of limited value for vehicle control. State-of-the-art road detection algorithms are used to demonstrate results using classical pixel-level metrics in perspective and BEV space as well as the novel behavior-based performance measure. All data and annotations are made publicly available on the KITTI online evaluation website in order to serve as a common benchmark for road terrain detection algorithms.",
"Road detection from images is a key task in autonomous driving. The recent advent of deep learning (and in particular, CNN or convolutional neural networks) has greatly improved the performance of road detection algorithms. In this paper, we show how to fuse multiple different cues under the same convolutional network framework. Specifically, we adopt a pre-trained Resnet-lOl to extract feature maps from RGB images; we then connect it with three extra deconvolution layers. These deconvolution layers is trained conditioning on appropriate image cues, and in our case they are a height image (i.e. elevation map obtained by e.g. Lidar scanner), image gradient, and position map. We also design two skip layers to speed up the convergence. Experiments on KITTI benchmark show competitive performance of our new networks."
]
} |
1905.11503 | 2947199171 | Modern approaches to pose and body shape estimation have recently achieved strong performance even under challenging real-world conditions. Even from a single image of a clothed person, a realistic looking body shape can be inferred that captures a users' weight group and body shape type well. This opens up a whole spectrum of applications -- in particular in fashion -- where virtual try-on and recommendation systems can make use of these new and automatized cues. However, a realistic depiction of the undressed body is regarded highly private and therefore might not be consented by most people. Hence, we ask if the automatic extraction of such information can be effectively evaded. While adversarial perturbations have been shown to be effective for manipulating the output of machine learning models -- in particular, end-to-end deep learning approaches -- state of the art shape estimation methods are composed of multiple stages. We perform the first investigation of different strategies that can be used to effectively manipulate the automatic shape estimation while preserving the overall appearance of the original image. | Recovery of 3D human shape from a 2D image is a very challenging task due to ambiguities such as depth and unknown camera data. This task has been facilitated by the availability of 3D generative body models learned from thousands of scans of people @cite_53 @cite_65 @cite_19 , which capture anthropometric constraints of the population and therefore reduce ambiguities. Several works @cite_61 @cite_3 @cite_8 @cite_11 @cite_38 @cite_40 @cite_71 @cite_12 @cite_11 leverage these generative models to estimate 3D shape from single or multiple images, using shading cues, silhouettes and appearance. Recent model based approaches are using deep learning based 2D detections @cite_34 -- by either fitting a model to them at test time @cite_17 @cite_61 @cite_40 @cite_68 @cite_26 or by using them to supervise bottom-up 3D shape predictors @cite_27 @cite_5 @cite_60 @cite_62 @cite_57 @cite_17 . Hence, to evade recent shape estimators, we study different strategies to attack the 2D keypoint detections while preserving the overall appearance of the original image. | {
"cite_N": [
"@cite_3",
"@cite_71",
"@cite_5",
"@cite_38",
"@cite_8",
"@cite_60",
"@cite_68",
"@cite_17",
"@cite_26",
"@cite_57",
"@cite_19",
"@cite_40",
"@cite_27",
"@cite_34",
"@cite_12",
"@cite_61",
"@cite_62",
"@cite_53",
"@cite_65",
"@cite_11"
],
"mid": [
"2103025041",
"2963355540",
"2798637590",
"1522277130",
"2545173102",
"2778680124",
"2793768642",
"2921745007",
"2894878561",
"2892165078",
"1967554269",
"2483862638",
"2962754033",
"2559085405",
"",
"2853932467",
"",
"1989191365",
"2049462432",
"2075834168"
],
"abstract": [
"Estimation of three-dimensional articulated human pose and motion from images is a central problem in computer vision. Much of the previous work has been limited by the use of crude generative models of humans represented as articulated collections of simple parts such as cylinders. Automatic initialization of such models has proved difficult and most approaches assume that the size and shape of the body parts are known a priori. In this paper we propose a method for automatically recovering a detailed parametric model of non-rigid body shape and pose from monocular imagery. Specifically, we represent the body using a parameterized triangulated mesh model that is learned from a database of human range scans. We demonstrate a discriminative method to directly recover the model parameters from monocular images using a conditional mixture of kernel regressors. This predicted pose and shape are used to initialize a generative model for more detailed pose and shape estimation. The resulting approach allows fully automatic pose and shape recovery from monocular and multi-camera imagery. Experimental results show that our method is capable of robustly recovering articulated pose, shape and biometric measurements (e.g. height, weight, etc.) in both calibrated and uncalibrated camera environments.",
"Existing markerless motion capture methods often assume known backgrounds, static cameras, and sequence specific motion priors, limiting their application scenarios. Here we present a fully automatic method that, given multi-view videos, estimates 3D human pose and body shape. We take the recently proposed SMPLify method bogo2016keep as the base method and extend it in several ways. First we fit a 3D human body model to 2D features detected in multi-view images. Second, we use a CNN method to segment the person in each image and fit the 3D body model to the contours, further improving accuracy. Third we utilize a generic and robust DCT temporal prior to handle the left and right side swapping issue sometimes introduced by the 2D pose estimator. Validation on standard benchmarks shows our results are comparable to the state of the art and also provide a realistic 3D shape avatar. We also demonstrate accurate results on HumanEva and on challenging monocular sequences of dancing from YouTube.",
"This work addresses the problem of estimating the full body 3D human pose and shape from a single color image. This is a task where iterative optimization-based solutions have typically prevailed, while Convolutional Networks (ConvNets) have suffered because of the lack of training data and their low resolution 3D predictions. Our work aims to bridge this gap and proposes an efficient and effective direct prediction method based on ConvNets. Central part to our approach is the incorporation of a parametric statistical body shape model (SMPL) within our end-to-end framework. This allows us to get very detailed 3D mesh results, while requiring estimation only of a small number of parameters, making it friendly for direct network prediction. Interestingly, we demonstrate that these parameters can be predicted reliably only from 2D keypoints and masks. These are typical outputs of generic 2D human analysis ConvNets, allowing us to relax the massive requirement that images with 3D shape ground truth are available for training. Simultaneously, by maintaining differentiability, at training time we generate the 3D mesh from the estimated parameters and optimize explicitly for the surface using a 3D per-vertex loss. Finally, a differentiable renderer is employed to project the 3D mesh to the image, which enables further refinement of the network, by optimizing for the consistency of the projection with 2D annotations (i.e., 2D keypoints or masks). The proposed approach outperforms previous baselines on this task and offers an attractive solution for direct prediction of3D shape from a single color image.",
"In this paper we propose a probabilistic framework that models shape variations and infers dense and detailed 3D shapes from a single silhouette. We model two types of shape variations, the object phenotype variation and its pose variation using two independent Gaussian Process Latent Variable Models (GPLVMs) respectively. The proposed shape variation models are learnt from 3D samples without prior knowledge about object class, e.g. object parts and skeletons, and are combined to fully span the 3D shape space. A novel probabilistic inference algorithm for 3D shape estimation is proposed by maximum likelihood estimates of the GPLVM latent variables and the camera parameters that best fit generated 3D shapes to given silhouettes. The proposed inference involves a small number of latent variables and it is computationally efficient. Experiments on both human body and shark data demonstrate the efficacy of our new approach.",
"We describe a solution to the challenging problem of estimating human body shape from a single photograph or painting. Our approach computes shape and pose parameters of a 3D human body model directly from monocular image cues and advances the state of the art in several directions. First, given a user-supplied estimate of the subject's height and a few clicked points on the body we estimate an initial 3D articulated body pose and shape. Second, using this initial guess we generate a tri-map of regions inside, outside and on the boundary of the human, which is used to segment the image using graph cuts. Third, we learn a low-dimensional linear model of human shape in which variations due to height are concentrated along a single dimension, enabling height-constrained estimation of body shape. Fourth, we formulate the problem of parametric human shape from shading. We estimate the body pose, shape and reflectance as well as the scene lighting that produces a synthesized body that robustly matches the image evidence. Quantitative experiments demonstrate how smooth shading provides powerful constraints on human shape. We further demonstrate a novel application in which we extract 3D human models from archival photographs and paintings.",
"We describe Human Mesh Recovery (HMR), an end-to-end framework for reconstructing a full 3D mesh of a human body from a single RGB image. In contrast to most current methods that compute 2D or 3D joint locations, we produce a richer and more useful mesh representation that is parameterized by shape and 3D joint angles. The main objective is to minimize the reprojection loss of keypoints, which allow our model to be trained using in-the-wild images that only have ground truth 2D annotations. However, reprojection loss alone is highly under constrained. In this work we address this problem by introducing an adversary trained to tell whether a human body parameter is real or not using a large database of 3D human meshes. We show that HMR can be trained with and without using any coupled 2D-to-3D supervision. We do not rely on intermediate 2D keypoint detection and infer 3D pose and shape parameters directly from image pixels. Our model runs in real-time given a bounding box containing the person. We demonstrate our approach on various images in-the-wild and out-perform previous optimizationbased methods that output 3D meshes and show competitive results on tasks such as 3D joint location estimation and part segmentation.",
"This paper describes a method to obtain accurate 3D body models and texture of arbitrary people from a single, monocular video in which a person is moving. Based on a parametric body model, we present a robust processing pipeline to infer 3D model shapes including clothed people with 4.5mm reconstruction accuracy. At the core of our approach is the transformation of dynamic body pose into a canonical frame of reference. Our main contribution is a method to transform the silhouette cones corresponding to dynamic human silhouettes to obtain a visual hull in a common reference frame. This enables efficient estimation of a consensus 3D shape, texture and implanted animation skeleton based on a large number of frames. Results on 4 different datasets demonstrate the effectiveness of our approach to produce accurate 3D models. Requiring only an RGB camera, our method enables everyone to create their own fully animatable digital double, e.g., for social VR applications or virtual try-on for online fashion shopping.",
"",
"We present the first real-time human performance capture approach that reconstructs dense, space-time coherent deforming geometry of entire humans in general everyday clothing from just a single RGB video. We propose a novel two-stage analysis-by-synthesis optimization whose formulation and implementation are designed for high performance. In the first stage, a skinned template model is jointly fitted to background subtracted input video, 2D and 3D skeleton joint positions found using a deep neural network, and a set of sparse facial landmark detections. In the second stage, dense non-rigid 3D deformations of skin and even loose apparel are captured based on a novel real-time capable algorithm for non-rigid tracking using dense photometric and silhouette constraints. Our novel energy formulation leverages automatically identified material regions on the template to model the differing non-rigid deformation behavior of skin and apparel. The two resulting non-linear optimization problems per frame are solved with specially tailored data-parallel Gauss-Newton solvers. To achieve real-time performance of over 25Hz, we design a pipelined parallel architecture using the CPU and two commodity GPUs. Our method is the first real-time monocular approach for full-body performance capture. Our method yields comparable accuracy with off-line performance capture techniques while being orders of magnitude faster.",
"",
"We present a learned model of human body shape and pose-dependent shape variation that is more accurate than previous models and is compatible with existing graphics pipelines. Our Skinned Multi-Person Linear model (SMPL) is a skinned vertex-based model that accurately represents a wide variety of body shapes in natural human poses. The parameters of the model are learned from data including the rest pose template, blend weights, pose-dependent blend shapes, identity-dependent blend shapes, and a regressor from vertices to joint locations. Unlike previous models, the pose-dependent blend shapes are a linear function of the elements of the pose rotation matrices. This simple formulation enables training the entire model from a relatively large number of aligned 3D meshes of different people in different poses. We quantitatively evaluate variants of SMPL using linear or dual-quaternion blend skinning and show that both are more accurate than a Blend-SCAPE model trained on the same data. We also extend SMPL to realistically model dynamic soft-tissue deformations. Because it is based on blend skinning, SMPL is compatible with existing rendering engines and we make it available for research purposes.",
"We describe the first method to automatically estimate the 3D pose of the human body as well as its 3D shape from a single unconstrained image. We estimate a full 3D mesh and show that 2D joints alone carry a surprising amount of information about body shape. The problem is challenging because of the complexity of the human body, articulation, occlusion, clothing, lighting, and the inherent ambiguity in inferring 3D from 2D. To solve this, we first use a recently published CNN-based method, DeepCut, to predict (bottom-up) the 2D body joint locations. We then fit (top-down) a recently published statistical body shape model, called SMPL, to the 2D joints. We do so by minimizing an objective function that penalizes the error between the projected 3D model joints and detected 2D joints. Because SMPL captures correlations in human shape across the population, we are able to robustly fit it to very little data. We further leverage the 3D model to prevent solutions that cause interpenetration. We evaluate our method, SMPLify, on the Leeds Sports, HumanEva, and Human3.6M datasets, showing superior pose accuracy with respect to the state of the art.",
"Direct prediction of 3D body pose and shape parameters remains a challenge even for highly parameterized, deep learning models. The representation of the prediction space is difficult to map to from the plain 2D image space, perspective ambiguities make the loss function noisy and training data is scarce. In this paper, we propose a novel approach (Neural Body Fitting (NBF)) that integrates a statistical body model as a layer within a CNN leveraging both reliable bottom-up body part segmentation and robust top-down body model constraints. NBF is fully differentiable and can be trained end-to-end from both 2D and 3D annotations. In detailed experiments we analyze how the components of our model improve model performance and present a robust, easy to use, end-to-end trainable framework for 3D human pose estimation from single 2D images.",
"We present an approach to efficiently detect the 2D pose of multiple people in an image. The approach uses a nonparametric representation, which we refer to as Part Affinity Fields (PAFs), to learn to associate body parts with individuals in the image. The architecture encodes global context, allowing a greedy bottom-up parsing step that maintains high accuracy while achieving realtime performance, irrespective of the number of people in the image. The architecture is designed to jointly learn part locations and their association via two branches of the same sequential prediction process. Our method placed first in the inaugural COCO 2016 keypoints challenge, and significantly exceeds the previous state-of-the-art result on the MPII Multi-Person benchmark, both in performance and efficiency.",
"",
"To study the correlation between clothing garments and body shape, we collected a new dataset (Fashion Takes Shape), which includes images of female users with clothing category annotations. Despite the progress in body shape estimation from images, it turns out to be challenging to infer body shape from such diverse, real-world photos. Hence, we propose a novel and robust multi-photo approach to estimate body shapes of each user and build a conditional model of clothing categories given body-shape. We demonstrate that in real-world data, clothing categories and body-shapes are correlated and show that our multi-photo approach leads to a better predictive model for clothing categories compared to models based on single-view shape estimates or manually annotated body types. We see our method as the first step towards the large-scale understanding of clothing preferences from body shape.",
"",
"We introduce the SCAPE method (Shape Completion and Animation for PEople)---a data-driven method for building a human shape model that spans variation in both subject shape and pose. The method is based on a representation that incorporates both articulated and non-rigid deformations. We learn a pose deformation model that derives the non-rigid surface deformation as a function of the pose of the articulated skeleton. We also learn a separate model of variation based on body shape. Our two models can be combined to produce 3D surface models with realistic muscle deformation for different people in different poses, when neither appear in the training set. We show how the model can be used for shape completion --- generating a complete surface mesh given a limited set of markers specifying the target shape. We present applications of shape completion to partial view completion and motion capture animation. In particular, our method is capable of constructing a high-quality animated surface model of a moving person, with realistic muscle deformation, using just a single static scan and a marker motion capture sequence of the person.",
"To look human, digital full-body avatars need to have soft-tissue deformations like those of real people. We learn a model of soft-tissue deformations from examples using a high-resolution 4D capture system and a method that accurately registers a template mesh to sequences of 3D scans. Using over 40,000 scans of ten subjects, we learn how soft-tissue motion causes mesh triangles to deform relative to a base 3D body model. Our Dyna model uses a low-dimensional linear subspace to approximate soft-tissue deformation and relates the subspace coefficients to the changing pose of the body. Dyna uses a second-order auto-regressive model that predicts soft-tissue deformations based on previous deformations, the velocity and acceleration of the body, and the angular velocities and accelerations of the limbs. Dyna also models how deformations vary with a person's body mass index (BMI), producing different deformations for people with different shapes. Dyna realistically represents the dynamics of soft tissue for previously unseen subjects and motions. We provide tools for animators to modify the deformations and apply them to new stylized characters.",
"We present an easy-to-use image retouching technique for realistic reshaping of human bodies in a single image. A model-based approach is taken by integrating a 3D whole-body morphable model into the reshaping process to achieve globally consistent editing effects. A novel body-aware image warping approach is introduced to reliably transfer the reshaping effects from the model to the image, even under moderate fitting errors. Thanks to the parametric nature of the model, our technique parameterizes the degree of reshaping by a small set of semantic attributes, such as weight and height. It allows easy creation of desired reshaping effects by changing the full-body attributes, while producing visually pleasing results even for loosely-dressed humans in casual photographs with a variety of poses and shapes."
]
} |
1905.11531 | 2946840550 | Semantic parsing is the process of translating natural language utterances into logical forms, which has many important applications such as question answering and instruction following. Sequence-to-sequence models have been very successful across many NLP tasks. However, a lack of task-specific prior knowledge can be detrimental to the performance of these models. Prior work has used frameworks for inducing grammars over the training examples, which capture conditional independence properties that the model can leverage. Inspired by the recent success stories such as BERT we set out to extend this augmentation framework into two stages. The first stage is to pre-train using a corpus of augmented examples in an unsupervised manner. The second stage is to fine-tune to a domain-specific task. In addition, since the pre-training stage is separate from the training on the main task we also expand the universe of possible augmentations without causing catastrophic inference. We also propose a novel data augmentation strategy that interchanges tokens that co-occur in similar contexts to produce new training pairs. We demonstrate that the proposed two-stage framework is beneficial for improving the parsing accuracy in a standard dataset called GeoQuery for the task of generating logical forms from a set of questions about the US geography. | The approach used in this work is a continuation of the work by @cite_8 where the authors proposed a sequence-to-sequence model with an attention-based copying mechanism. This supervised approach leverages the flexibility of of the encoder-decoder architecture and the authors demonstrate that the model can learn very accurate parsers across three standard semantic parsing datasets. The augmentation strategy used in this work allows for injecting prior knowledge which improve the generalization power of the model. | {
"cite_N": [
"@cite_8"
],
"mid": [
"2444236087"
],
"abstract": [
"Modeling crisp logical regularities is crucial in semantic parsing, making it difficult for neural models with no task-specific prior knowledge to achieve good results. In this paper, we introduce data recombination, a novel framework for injecting such prior knowledge into a model. From the training data, we induce a high-precision synchronous context-free grammar, which captures important conditional independence properties commonly found in semantic parsing. We then train a sequence-to-sequence recurrent network (RNN) model with a novel attention-based copying mechanism on datapoints sampled from this grammar, thereby teaching the model about these structural properties. Data recombination improves the accuracy of our RNN model on three semantic parsing datasets, leading to new state-of-the-art performance on the standard GeoQuery dataset for models with comparable supervision."
]
} |
1905.11531 | 2946840550 | Semantic parsing is the process of translating natural language utterances into logical forms, which has many important applications such as question answering and instruction following. Sequence-to-sequence models have been very successful across many NLP tasks. However, a lack of task-specific prior knowledge can be detrimental to the performance of these models. Prior work has used frameworks for inducing grammars over the training examples, which capture conditional independence properties that the model can leverage. Inspired by the recent success stories such as BERT we set out to extend this augmentation framework into two stages. The first stage is to pre-train using a corpus of augmented examples in an unsupervised manner. The second stage is to fine-tune to a domain-specific task. In addition, since the pre-training stage is separate from the training on the main task we also expand the universe of possible augmentations without causing catastrophic inference. We also propose a novel data augmentation strategy that interchanges tokens that co-occur in similar contexts to produce new training pairs. We demonstrate that the proposed two-stage framework is beneficial for improving the parsing accuracy in a standard dataset called GeoQuery for the task of generating logical forms from a set of questions about the US geography. | Moreover, building annotated semantic parsing datasets is highly labor-intensive and parsers built for one domain do not necessarily transfer across domains. propose a multi-task setup and demonstrate that training using this setup can improve the accuracy in domains with smaller labeled datasets @cite_10 . This approach is aligned with one of the main contributions in our research. Our proposed framework allows for pre-training the model in an unsupervied manner with data from multiple tasks that enables transfer learning. | {
"cite_N": [
"@cite_10"
],
"mid": [
"2625324377"
],
"abstract": [
"The goal of semantic parsing is to map natural language to a machine interpretable meaning representation language (MRL). One of the constraints that limits full exploration of deep learning technologies for semantic parsing is the lack of sufficient annotation training data. In this paper, we propose using sequence-to-sequence in a multi-task setup for semantic parsing with a focus on transfer learning. We explore three multi-task architectures for sequence-to-sequence modeling and compare their performance with an independently trained model. Our experiments show that the multi-task setup aids transfer learning from an auxiliary task with large labeled data to a target task with smaller labeled data. We see absolute accuracy gains ranging from 1.0 to 4.4 in our in- house data set, and we also see good gains ranging from 2.5 to 7.0 on the ATIS semantic parsing tasks with syntactic and semantic auxiliary tasks."
]
} |
1905.11280 | 2946527621 | In this work we consider the interplay between multiprover interactive proofs, quantum entanglement, and zero knowledge proofs - notions that are central pillars of complexity theory, quantum information and cryptography. In particular, we study the relationship between the complexity class MIP @math , the set of languages decidable by multiprover interactive proofs with quantumly entangled provers, and the class PZKMIP @math , which is the set of languages decidable by MIP @math protocols that furthermore possess the perfect zero knowledge property. Our main result is that the two classes are equal, i.e., MIP @math PZKMIP @math . This result provides a quantum analogue of the celebrated result of Ben-Or, Goldwasser, Kilian, and Wigderson (STOC 1988) who show that MIP @math PZKMIP (in other words, all classical multiprover interactive protocols can be made zero knowledge). We prove our result by showing that every MIP @math protocol can be efficiently transformed into an equivalent zero knowledge MIP @math protocol in a manner that preserves the completeness-soundness gap. Combining our transformation with previous results by Slofstra (Forum of Mathematics, Pi 2019) and Fitzsimons, Ji, Vidick and Yuen (STOC 2019), we obtain the corollary that all co-recursively enumerable languages (which include undecidable problems as well as all decidable problems) have zero knowledge MIP @math protocols with vanishing promise gap. | In quantum information theory, zero knowledge proofs have been primarily studied in the context of quantum interactive proofs. This setting was first formalized by Watrous @cite_31 , and has been an active area of research over the years. Various aspects of zero knowledge quantum interactive proofs have been studied, including honest verifier models @cite_31 @cite_2 , computational zero knowledge proof systems for @math @cite_11 , and more. | {
"cite_N": [
"@cite_31",
"@cite_11",
"@cite_2"
],
"mid": [
"2120279343",
"2341328795",
"1550821730"
],
"abstract": [
"In this paper we propose a definition for (honest verifier) quantum statistical zero-knowledge interactive proof systems and study the resulting complexity class, which we denote QSZK sub HV . We prove several facts regarding this class, including: the following problem is a complete promise problem for QSZKHV: given instructions for preparing two mixed quantum states, are the states close together or far apart in the trace norm metric? This problem is a quantum generalization of the complete promise problem of Sahai and Vadhan (1997) for (classical) statistical zero-knowledge; QSZK sub HV is closed under complement; QSZK sub HV spl sube PSPACE. (At present it is not known if arbitrary quantum interactive proof systems can be simulated in PSPACE even for one-round proof systems); any polynomial-round honest verifier quantum statistical zero-knowledge proof system can be simulated by a two-message (i.e., one-round) honest verifier quantum statistical zero-knowledge proof system. Similarly, any polynomial-round honest verifier quantum statistical zero-knowledge proof system can be simulated by a three-message public-coin honest verifier quantum statistical zero-knowledge proof system. These facts establish close connections between classical statistical zero-knowledge and our definition for quantum statistical zero-knowledge, and give some insight regarding the effect of this zero-knowledge restriction on quantum interactive proof systems. The relationship between our definition and possible definitions of general (i.e., not necessarily honest) quantum statistical zero-knowledge are also discussed.",
"Prior work has established that all problems in NP admit classical zero-knowledge proof systems, and under reasonable hardness assumptions for quantum computations, these proof systems can be made secure against quantum attacks. We prove a result representing a further quantum generalization of this fact, which is that every problem in the complexity class QMA has a quantum zero-knowledge proof system. More specifically, assuming the existence of an unconditionally binding and quantum computationally concealing commitment scheme, we prove that every problem in the complexity class QMA has a quantum interactive proof system that is zero-knowledge with respect to efficient quantum computations. Our QMA proof system is sound against arbitrary quantum provers, but only requires an honest prover to perform polynomial-time quantum computations, provided that it holds a quantum witness for a given instance of the QMA problem under consideration.",
"In quantum zero knowledge, the assumption was made that the verifier is only using unitary operations. Under this assumption, many nice properties have been shown about quantum zero knowledge, including the fact that Honest-Verifier Quantum Statistical Zero Knowledge ( @math ) is equal to Cheating-Verifier Quantum Statistical Zero Knowledge ( @math ) (see Wat02,Wat06 ). In this paper, we study what happens when we allow an honest verifier to flip some coins in addition to using unitary operations. Flipping a coin is a non-unitary operation but doesn 't seem at first to enhance the cheating possibilities of the verifier since a classical honest verifier can flip coins. In this setting, we show an unexpected result: any classical Interactive Proof has an Honest-Verifier Quantum Statistical Zero Knowledge proof with coins. Note that in the classical case, honest verifier @math is no more powerful than @math and hence it is not believed to contain even @math . On the other hand, in the case of cheating verifiers, we show that Quantum Statistical Zero Knowledge where the verifier applies any non-unitary operation is equal to Quantum Zero-Knowledge where the verifier uses only unitaries. One can think of our results in two complementary ways. If we would like to use the honest verifier model as a means to study the general model by taking advantage of their equivalence, then it is imperative to use the unitary definition without coins, since with the general one this equivalence is most probably not true. On the other hand, if we would like to use quantum zero knowledge protocols in a cryptographic scenario where the honest-but-curious model is sufficient, then adding the unitary constraint severely decreases the power of quantum zero knowledge protocols."
]
} |
1905.11280 | 2946527621 | In this work we consider the interplay between multiprover interactive proofs, quantum entanglement, and zero knowledge proofs - notions that are central pillars of complexity theory, quantum information and cryptography. In particular, we study the relationship between the complexity class MIP @math , the set of languages decidable by multiprover interactive proofs with quantumly entangled provers, and the class PZKMIP @math , which is the set of languages decidable by MIP @math protocols that furthermore possess the perfect zero knowledge property. Our main result is that the two classes are equal, i.e., MIP @math PZKMIP @math . This result provides a quantum analogue of the celebrated result of Ben-Or, Goldwasser, Kilian, and Wigderson (STOC 1988) who show that MIP @math PZKMIP (in other words, all classical multiprover interactive protocols can be made zero knowledge). We prove our result by showing that every MIP @math protocol can be efficiently transformed into an equivalent zero knowledge MIP @math protocol in a manner that preserves the completeness-soundness gap. Combining our transformation with previous results by Slofstra (Forum of Mathematics, Pi 2019) and Fitzsimons, Ji, Vidick and Yuen (STOC 2019), we obtain the corollary that all co-recursively enumerable languages (which include undecidable problems as well as all decidable problems) have zero knowledge MIP @math protocols with vanishing promise gap. | In the multiprover setting, Chiesa, Forbes, Gur and Spooner @cite_6 showed that all problems in @math (and thus @math ) are in @math . Their approach is considerably different of ours. They achieve their result by showing that model of interactive proofs called An interactive PCP is a protocol where the verifier and a single prover first commit to an oracle, which the verifier can query a bounded number of times. Then, the verifier and prover engage in an interactive proof. An interactive PCP is one where the committed oracle has a desired algebraic structure. We refer to @cite_6 for an in-depth discussion of these models. can be lifted to the entangled provers setting in a way that preserves zero knowledge, and then showing that languages in @math have zero knowledge algebraic interactive PCPs. | {
"cite_N": [
"@cite_6"
],
"mid": [
"2962993321"
],
"abstract": [
"Zero knowledge plays a central role in cryptography and complexity. The seminal work of Ben- (STOC 1988) shows that zero knowledge can be achieved unconditionally for any language in NEXP, as long as one is willing to make a suitable *physical assumption*: if the provers are spatially isolated, then they can be assumed to be playing independent strategies. Quantum mechanics, however, tells us that this assumption is unrealistic, because spatially-isolated provers could share a quantum entangled state and realize a non-local correlated strategy. The MIP^* model captures this setting. In this work we study the following question: does spatial isolation still suffice to unconditionally achieve zero knowledge even in the presence of quantum entanglement? We answer this question in the affirmative: we prove that every language in NEXP has a 2-prover *zero knowledge* interactive proof that is sound against entangled provers; that is, NEXP ⊆ ZK-MIP^*. Our proof consists of constructing a zero knowledge interactive PCP with a strong algebraic structure, and then lifting it to the MIP^* model. This lifting relies on a new framework that builds on recent advances in low-degree testing against entangled strategies, and clearly separates classical and quantum tools. Our main technical contribution is the development of algebraic techniques for obtaining unconditional zero knowledge; this includes a zero knowledge variant of the celebrated sumcheck protocol, a key building block in many probabilistic proof systems. A core component of our sumcheck protocol is a new algebraic commitment scheme, whose analysis relies on algebraic complexity theory."
]
} |
1905.11280 | 2946527621 | In this work we consider the interplay between multiprover interactive proofs, quantum entanglement, and zero knowledge proofs - notions that are central pillars of complexity theory, quantum information and cryptography. In particular, we study the relationship between the complexity class MIP @math , the set of languages decidable by multiprover interactive proofs with quantumly entangled provers, and the class PZKMIP @math , which is the set of languages decidable by MIP @math protocols that furthermore possess the perfect zero knowledge property. Our main result is that the two classes are equal, i.e., MIP @math PZKMIP @math . This result provides a quantum analogue of the celebrated result of Ben-Or, Goldwasser, Kilian, and Wigderson (STOC 1988) who show that MIP @math PZKMIP (in other words, all classical multiprover interactive protocols can be made zero knowledge). We prove our result by showing that every MIP @math protocol can be efficiently transformed into an equivalent zero knowledge MIP @math protocol in a manner that preserves the completeness-soundness gap. Combining our transformation with previous results by Slofstra (Forum of Mathematics, Pi 2019) and Fitzsimons, Ji, Vidick and Yuen (STOC 2019), we obtain the corollary that all co-recursively enumerable languages (which include undecidable problems as well as all decidable problems) have zero knowledge MIP @math protocols with vanishing promise gap. | The results of @cite_6 are, strictly speaking, incomparable to ours. We show that all languages in @math have single-round @math protocols with four additional provers, whereas @cite_6 show that @math (which is a subset of @math ) have @math protocols with provers and polynomially many rounds. Improving our result to only two provers seems to be quite a daunting challenge, as it is not even known how @math relates to @math -- it could potentially be the case that adding more entangled provers yields a strictly larger complexity class! | {
"cite_N": [
"@cite_6"
],
"mid": [
"2962993321"
],
"abstract": [
"Zero knowledge plays a central role in cryptography and complexity. The seminal work of Ben- (STOC 1988) shows that zero knowledge can be achieved unconditionally for any language in NEXP, as long as one is willing to make a suitable *physical assumption*: if the provers are spatially isolated, then they can be assumed to be playing independent strategies. Quantum mechanics, however, tells us that this assumption is unrealistic, because spatially-isolated provers could share a quantum entangled state and realize a non-local correlated strategy. The MIP^* model captures this setting. In this work we study the following question: does spatial isolation still suffice to unconditionally achieve zero knowledge even in the presence of quantum entanglement? We answer this question in the affirmative: we prove that every language in NEXP has a 2-prover *zero knowledge* interactive proof that is sound against entangled provers; that is, NEXP ⊆ ZK-MIP^*. Our proof consists of constructing a zero knowledge interactive PCP with a strong algebraic structure, and then lifting it to the MIP^* model. This lifting relies on a new framework that builds on recent advances in low-degree testing against entangled strategies, and clearly separates classical and quantum tools. Our main technical contribution is the development of algebraic techniques for obtaining unconditional zero knowledge; this includes a zero knowledge variant of the celebrated sumcheck protocol, a key building block in many probabilistic proof systems. A core component of our sumcheck protocol is a new algebraic commitment scheme, whose analysis relies on algebraic complexity theory."
]
} |
1905.11280 | 2946527621 | In this work we consider the interplay between multiprover interactive proofs, quantum entanglement, and zero knowledge proofs - notions that are central pillars of complexity theory, quantum information and cryptography. In particular, we study the relationship between the complexity class MIP @math , the set of languages decidable by multiprover interactive proofs with quantumly entangled provers, and the class PZKMIP @math , which is the set of languages decidable by MIP @math protocols that furthermore possess the perfect zero knowledge property. Our main result is that the two classes are equal, i.e., MIP @math PZKMIP @math . This result provides a quantum analogue of the celebrated result of Ben-Or, Goldwasser, Kilian, and Wigderson (STOC 1988) who show that MIP @math PZKMIP (in other words, all classical multiprover interactive protocols can be made zero knowledge). We prove our result by showing that every MIP @math protocol can be efficiently transformed into an equivalent zero knowledge MIP @math protocol in a manner that preserves the completeness-soundness gap. Combining our transformation with previous results by Slofstra (Forum of Mathematics, Pi 2019) and Fitzsimons, Ji, Vidick and Yuen (STOC 2019), we obtain the corollary that all co-recursively enumerable languages (which include undecidable problems as well as all decidable problems) have zero knowledge MIP @math protocols with vanishing promise gap. | Furthermore, the proof techniques of @cite_6 are very different from ours: they heavily rely on algebraic PCP techniques, as well as the analysis of the low degree test against entangled provers @cite_37 . Our proof relies on techniques from fault tolerant quantum computing and the protocol compression procedure of @cite_4 @cite_35 , which in turn rely heavily on self-testing and history state Hamiltonians. | {
"cite_N": [
"@cite_35",
"@cite_37",
"@cite_4",
"@cite_6"
],
"mid": [
"2807350215",
"2761821748",
"2530940551",
"2962993321"
],
"abstract": [
"We show that any language in nondeterministic time @math , where the number of iterated exponentials is an arbitrary function @math , can be decided by a multiprover interactive proof system with a classical polynomial-time verifier and a constant number of quantum entangled provers, with completeness @math and soundness @math , where the number of iterated exponentials is @math and @math is a universal constant. The result was previously known for @math and @math ; we obtain it for any time-constructible function @math . The result is based on a compression technique for interactive proof systems with entangled provers that significantly simplifies and strengthens a protocol compression result of Ji (STOC'17). As a separate consequence of this technique we obtain a different proof of Slofstra's recent result (unpublished) on the uncomputability of the entangled value of multiprover games. Finally, we show that even minor improvements to our compression result would yield remarkable consequences in computational complexity theory and the foundations of quantum mechanics: first, it would imply that the class MIP* contains all computable languages; second, it would provide a negative resolution to a multipartite version of Tsirelson's problem on the relation between the commuting operator and tensor product models for quantum correlations.",
"We show that it is NP-hard to approximate, to within an additive constant, the maximum success probability of players sharing quantum entanglement in a two-player game with classical questions of logarithmic length and classical answers of constant length. As a corollary, the inclusion NEXP ⊆ MIP*, first shown by Ito and Vidick (FOCS'12) with three provers, holds with two provers only. The proof is based on a simpler, improved analysis of the low-degree test of Raz and Safra (STOC'97) against two entangled provers.",
"We present a protocol that transforms any quantum multi-prover interactive proof into a nonlocal game in which questions consist of logarithmic number of bits and answers of constant number of bits. As a corollary, it follows that the promise problem corresponding to the approximation of the nonlocal value to inverse polynomial accuracy is complete for QMIP*, and therefore NEXP-hard. This establishes that nonlocal games are provably harder than classical games without any complexity theory assumptions. Our result also indicates that gap amplification for nonlocal games may be impossible in general and provides a negative evidence for the feasibility of the gap amplification approach to the multi-prover variant of the quantum PCP conjecture.",
"Zero knowledge plays a central role in cryptography and complexity. The seminal work of Ben- (STOC 1988) shows that zero knowledge can be achieved unconditionally for any language in NEXP, as long as one is willing to make a suitable *physical assumption*: if the provers are spatially isolated, then they can be assumed to be playing independent strategies. Quantum mechanics, however, tells us that this assumption is unrealistic, because spatially-isolated provers could share a quantum entangled state and realize a non-local correlated strategy. The MIP^* model captures this setting. In this work we study the following question: does spatial isolation still suffice to unconditionally achieve zero knowledge even in the presence of quantum entanglement? We answer this question in the affirmative: we prove that every language in NEXP has a 2-prover *zero knowledge* interactive proof that is sound against entangled provers; that is, NEXP ⊆ ZK-MIP^*. Our proof consists of constructing a zero knowledge interactive PCP with a strong algebraic structure, and then lifting it to the MIP^* model. This lifting relies on a new framework that builds on recent advances in low-degree testing against entangled strategies, and clearly separates classical and quantum tools. Our main technical contribution is the development of algebraic techniques for obtaining unconditional zero knowledge; this includes a zero knowledge variant of the celebrated sumcheck protocol, a key building block in many probabilistic proof systems. A core component of our sumcheck protocol is a new algebraic commitment scheme, whose analysis relies on algebraic complexity theory."
]
} |
1905.11280 | 2946527621 | In this work we consider the interplay between multiprover interactive proofs, quantum entanglement, and zero knowledge proofs - notions that are central pillars of complexity theory, quantum information and cryptography. In particular, we study the relationship between the complexity class MIP @math , the set of languages decidable by multiprover interactive proofs with quantumly entangled provers, and the class PZKMIP @math , which is the set of languages decidable by MIP @math protocols that furthermore possess the perfect zero knowledge property. Our main result is that the two classes are equal, i.e., MIP @math PZKMIP @math . This result provides a quantum analogue of the celebrated result of Ben-Or, Goldwasser, Kilian, and Wigderson (STOC 1988) who show that MIP @math PZKMIP (in other words, all classical multiprover interactive protocols can be made zero knowledge). We prove our result by showing that every MIP @math protocol can be efficiently transformed into an equivalent zero knowledge MIP @math protocol in a manner that preserves the completeness-soundness gap. Combining our transformation with previous results by Slofstra (Forum of Mathematics, Pi 2019) and Fitzsimons, Ji, Vidick and Yuen (STOC 2019), we obtain the corollary that all co-recursively enumerable languages (which include undecidable problems as well as all decidable problems) have zero knowledge MIP @math protocols with vanishing promise gap. | Another qualitative difference between the zero knowledge protocol of @cite_6 and ours is that the honest prover strategy for their protocol does not require any entanglement; the provers can behave classically. In our protocol, however, the provers are required to use entanglement; this is what enables the class @math and @math to contain classes beyond @math , such as @math (and beyond). | {
"cite_N": [
"@cite_6"
],
"mid": [
"2962993321"
],
"abstract": [
"Zero knowledge plays a central role in cryptography and complexity. The seminal work of Ben- (STOC 1988) shows that zero knowledge can be achieved unconditionally for any language in NEXP, as long as one is willing to make a suitable *physical assumption*: if the provers are spatially isolated, then they can be assumed to be playing independent strategies. Quantum mechanics, however, tells us that this assumption is unrealistic, because spatially-isolated provers could share a quantum entangled state and realize a non-local correlated strategy. The MIP^* model captures this setting. In this work we study the following question: does spatial isolation still suffice to unconditionally achieve zero knowledge even in the presence of quantum entanglement? We answer this question in the affirmative: we prove that every language in NEXP has a 2-prover *zero knowledge* interactive proof that is sound against entangled provers; that is, NEXP ⊆ ZK-MIP^*. Our proof consists of constructing a zero knowledge interactive PCP with a strong algebraic structure, and then lifting it to the MIP^* model. This lifting relies on a new framework that builds on recent advances in low-degree testing against entangled strategies, and clearly separates classical and quantum tools. Our main technical contribution is the development of algebraic techniques for obtaining unconditional zero knowledge; this includes a zero knowledge variant of the celebrated sumcheck protocol, a key building block in many probabilistic proof systems. A core component of our sumcheck protocol is a new algebraic commitment scheme, whose analysis relies on algebraic complexity theory."
]
} |
1905.11280 | 2946527621 | In this work we consider the interplay between multiprover interactive proofs, quantum entanglement, and zero knowledge proofs - notions that are central pillars of complexity theory, quantum information and cryptography. In particular, we study the relationship between the complexity class MIP @math , the set of languages decidable by multiprover interactive proofs with quantumly entangled provers, and the class PZKMIP @math , which is the set of languages decidable by MIP @math protocols that furthermore possess the perfect zero knowledge property. Our main result is that the two classes are equal, i.e., MIP @math PZKMIP @math . This result provides a quantum analogue of the celebrated result of Ben-Or, Goldwasser, Kilian, and Wigderson (STOC 1988) who show that MIP @math PZKMIP (in other words, all classical multiprover interactive protocols can be made zero knowledge). We prove our result by showing that every MIP @math protocol can be efficiently transformed into an equivalent zero knowledge MIP @math protocol in a manner that preserves the completeness-soundness gap. Combining our transformation with previous results by Slofstra (Forum of Mathematics, Pi 2019) and Fitzsimons, Ji, Vidick and Yuen (STOC 2019), we obtain the corollary that all co-recursively enumerable languages (which include undecidable problems as well as all decidable problems) have zero knowledge MIP @math protocols with vanishing promise gap. | Recently, Kinoshita @cite_20 showed that a model of honest-verifier'' zero knowledge QMIP can be lifted to general zero knowledge QMIP protocols. He also shows that @math have interactive proofs with zero knowledge proofs under a computational assumption. | {
"cite_N": [
"@cite_20"
],
"mid": [
"2915900549"
],
"abstract": [
"Zero-knowledge and multi-prover systems are both central notions in classical and quantum complexity theory. There is, however, little research in quantum multi-prover zero-knowledge systems. This paper studies complexity-theoretical aspects of the quantum multi-prover zero-knowledge systems. This paper has two results: 1.QMIP* systems with honest zero-knowledge can be converted into general zero-knowledge systems without any assumptions. 2.QMIP* has computational quantum zero-knowledge systems if a natural computational conjecture holds. One of the main tools is a test (called the GHZ test) that uses GHZ states shared by the provers, which prevents the verifier's attack in the above two results. Another main tool is what we call the Local Hamiltonian based Interactive protocol (LHI protocol). The LHI protocol makes previous research for Local Hamiltonians applicable to check the history state of interactive proofs, and we then apply 's zero-knowledge protocol for QMA BJSW to quantum multi-prover systems in order to obtain the second result."
]
} |
1905.11280 | 2946527621 | In this work we consider the interplay between multiprover interactive proofs, quantum entanglement, and zero knowledge proofs - notions that are central pillars of complexity theory, quantum information and cryptography. In particular, we study the relationship between the complexity class MIP @math , the set of languages decidable by multiprover interactive proofs with quantumly entangled provers, and the class PZKMIP @math , which is the set of languages decidable by MIP @math protocols that furthermore possess the perfect zero knowledge property. Our main result is that the two classes are equal, i.e., MIP @math PZKMIP @math . This result provides a quantum analogue of the celebrated result of Ben-Or, Goldwasser, Kilian, and Wigderson (STOC 1988) who show that MIP @math PZKMIP (in other words, all classical multiprover interactive protocols can be made zero knowledge). We prove our result by showing that every MIP @math protocol can be efficiently transformed into an equivalent zero knowledge MIP @math protocol in a manner that preserves the completeness-soundness gap. Combining our transformation with previous results by Slofstra (Forum of Mathematics, Pi 2019) and Fitzsimons, Ji, Vidick and Yuen (STOC 2019), we obtain the corollary that all co-recursively enumerable languages (which include undecidable problems as well as all decidable problems) have zero knowledge MIP @math protocols with vanishing promise gap. | Coudron and Slofstra prove a similar result to @cite_35 for multiprover proofs with commuting operator strategies, showing that this class also contains languages of arbitrarily large time complexity, if the promise gap is allowed to be arbitrarily small @cite_23 . Their results (achieved via a completely different method from ours) also show that there are two-prover zero knowledge proofs for languages of arbitrarily large time complexity, albeit in the commuting operator model and with a quantitatively worse lower bound than Corollary . | {
"cite_N": [
"@cite_35",
"@cite_23"
],
"mid": [
"2807350215",
"2947949186"
],
"abstract": [
"We show that any language in nondeterministic time @math , where the number of iterated exponentials is an arbitrary function @math , can be decided by a multiprover interactive proof system with a classical polynomial-time verifier and a constant number of quantum entangled provers, with completeness @math and soundness @math , where the number of iterated exponentials is @math and @math is a universal constant. The result was previously known for @math and @math ; we obtain it for any time-constructible function @math . The result is based on a compression technique for interactive proof systems with entangled provers that significantly simplifies and strengthens a protocol compression result of Ji (STOC'17). As a separate consequence of this technique we obtain a different proof of Slofstra's recent result (unpublished) on the uncomputability of the entangled value of multiprover games. Finally, we show that even minor improvements to our compression result would yield remarkable consequences in computational complexity theory and the foundations of quantum mechanics: first, it would imply that the class MIP* contains all computable languages; second, it would provide a negative resolution to a multipartite version of Tsirelson's problem on the relation between the commuting operator and tensor product models for quantum correlations.",
"We study the problem of approximating the commuting-operator value of a two-player non-local game. It is well-known that it is @math -complete to decide whether the classical value of a non-local game is 1 or @math . Furthermore, as long as @math is small enough, this result does not depend on the gap @math . In contrast, a recent result of Fitzsimons, Ji, Vidick, and Yuen shows that the complexity of computing the quantum value grows without bound as the gap @math decreases. In this paper, we show that this also holds for the commuting-operator value of a game. Specifically, in the language of multi-prover interactive proofs, we show that the power of @math (proofs with two provers, one round, completeness probability @math , soundness probability @math , and commuting-operator strategies) can increase without bound as the gap @math gets arbitrarily small. Our results also extend naturally in two ways, to perfect zero-knowledge protocols, and to lower bounds on the complexity of computing the approximately-commuting value of a game. Thus we get lower bounds on the complexity class @math - @math of perfect zero-knowledge multi-prover proofs with approximately-commuting operator strategies, as the gap @math gets arbitrarily small. While we do not know any computable time upper bound on the class @math , a result of the first author and Vidick shows that for @math and @math , the class @math , with constant communication from the provers, is contained in @math . We give a lower bound of @math (ignoring constants inside the function) for this class, which is tight up to polynomial factors assuming the exponential time hypothesis."
]
} |
1905.11280 | 2946527621 | In this work we consider the interplay between multiprover interactive proofs, quantum entanglement, and zero knowledge proofs - notions that are central pillars of complexity theory, quantum information and cryptography. In particular, we study the relationship between the complexity class MIP @math , the set of languages decidable by multiprover interactive proofs with quantumly entangled provers, and the class PZKMIP @math , which is the set of languages decidable by MIP @math protocols that furthermore possess the perfect zero knowledge property. Our main result is that the two classes are equal, i.e., MIP @math PZKMIP @math . This result provides a quantum analogue of the celebrated result of Ben-Or, Goldwasser, Kilian, and Wigderson (STOC 1988) who show that MIP @math PZKMIP (in other words, all classical multiprover interactive protocols can be made zero knowledge). We prove our result by showing that every MIP @math protocol can be efficiently transformed into an equivalent zero knowledge MIP @math protocol in a manner that preserves the completeness-soundness gap. Combining our transformation with previous results by Slofstra (Forum of Mathematics, Pi 2019) and Fitzsimons, Ji, Vidick and Yuen (STOC 2019), we obtain the corollary that all co-recursively enumerable languages (which include undecidable problems as well as all decidable problems) have zero knowledge MIP @math protocols with vanishing promise gap. | Finally, Cr ' e peau and Yang @cite_27 refined the notion of zero knowledge, requiring the simulator to be local, i.e., that there are non-communicating classical simulators that simulate the (joint) output distribution of the provers. We note that our result does not fulfill this modified definition, and we leave it as an open problem (dis)proving that all @math can be made zero knowledge in this setting. | {
"cite_N": [
"@cite_27"
],
"mid": [
"2950183140"
],
"abstract": [
"In multi-prover interactive proofs (MIPs), the verifier is usually non-adaptive. This stems from an implicit problem which we call contamination'' by the verifier. We make explicit the verifier contamination problem, and identify a solution by constructing a generalization of the MIP model. This new model quantifies non-locality as a new dimension in the characterization of MIPs. A new property of zero-knowledge emerges naturally as a result by also quantifying the non-locality of the simulator."
]
} |
1905.11533 | 2947944300 | Today a canonical approach to reduce the computation cost of Deep Neural Networks (DNNs) is to pre-define an over-parameterized model before training to guarantee the learning capacity, and then prune unimportant learning units (filters and neurons) during training to improve model compactness. We argue it is unnecessary to introduce redundancy at the beginning of the training but then reduce redundancy for the ultimate inference model. In this paper, we propose a Continuous Growth and Pruning (CGaP) scheme to minimize the redundancy from the beginning. CGaP starts the training from a small network seed, then expands the model continuously by reinforcing important learning units, and finally prunes the network to obtain a compact and accurate model. As the growth phase favors important learning units, CGaP provides a clear learning purpose to the pruning phase. Experimental results on representative datasets and DNN architectures demonstrate that CGaP outperforms previous pruning-only approaches that deal with pre-defined structures. For VGG-19 on CIFAR-100 and SVHN datasets, CGaP reduces the number of parameters by 78.9 and 85.8 , FLOPs by 53.2 and 74.2 , respectively; For ResNet-110 On CIFAR-10, CGaP reduces 64.0 number of parameters and 63.3 FLOPs. | There has been intense interest in reducing the over-parameterization of DNNs. The structure surgery has been widely used, including destructive and constructive methods. Destructive approaches remove connections or filters neurons from the structure, generating sparse models. For instance, @cite_4 pruned connections by removing less important weights determined by magnitude. @cite_16 pruned individual filters layer by layer based on the saliency metrics of each filter. Other similar saliency-based filter pruning approaches include @cite_30 @cite_8 @cite_26 @cite_28 . Penalty-based sparsity regularization have been explored by @cite_6 @cite_22 and structured sparsity was achieved. Our method is different from all above pruning schemes from two perspectives: We start training from a small seed other than an over-parameterized network. Apart from discarding unimportant filters neurons, we also strengthen important ones to further raise accuracy and compactness. | {
"cite_N": [
"@cite_30",
"@cite_26",
"@cite_4",
"@cite_22",
"@cite_8",
"@cite_28",
"@cite_6",
"@cite_16"
],
"mid": [
"2526468814",
"2495425901",
"2963674932",
"1935978687",
"2963363373",
"",
"2513419314",
"2515385951"
],
"abstract": [
"",
"State-of-the-art neural networks are getting deeper and wider. While their performance increases with the increasing number of layers and neurons, it is crucial to design an efficient deep architecture in order to reduce computational and memory costs. Designing an efficient neural network, however, is labor intensive requiring many experiments, and fine-tunings. In this paper, we introduce network trimming which iteratively optimizes the network by pruning unimportant neurons based on analysis of their outputs on a large dataset. Our algorithm is inspired by an observation that the outputs of a significant portion of neurons in a large network are mostly zero, regardless of what inputs the network received. These zero activation neurons are redundant, and can be removed without affecting the overall accuracy of the network. After pruning the zero activation neurons, we retrain the network using the weights before pruning as initialization. We alternate the pruning and retraining to further reduce zero activations in a network. Our experiments on the LeNet and VGG-16 show that we can achieve high compression ratio of parameters without losing or even achieving higher accuracy than the original network.",
"Neural networks are both computationally intensive and memory intensive, making them difficult to deploy on embedded systems. Also, conventional networks fix the architecture before training starts; as a result, training cannot improve the architecture. To address these limitations, we describe a method to reduce the storage and computation required by neural networks by an order of magnitude without affecting their accuracy by learning only the important connections. Our method prunes redundant connections using a three-step method. First, we train the network to learn which connections are important. Next, we prune the unimportant connections. Finally, we retrain the network to fine tune the weights of the remaining connections. On the ImageNet dataset, our method reduced the number of parameters of AlexNet by a factor of 9x, from 61 million to 6.7 million, without incurring accuracy loss. Similar experiments with VGG-16 found that the total number of parameters can be reduced by 13x, from 138 million to 10.3 million, again with no loss of accuracy.",
"Deep neural networks have achieved remarkable performance in both image classification and object detection problems, at the cost of a large number of parameters and computational complexity. In this work, we show how to reduce the redundancy in these parameters using a sparse decomposition. Maximum sparsity is obtained by exploiting both inter-channel and intra-channel redundancy, with a fine-tuning step that minimize the recognition loss caused by maximizing sparsity. This procedure zeros out more than 90 of parameters, with a drop of accuracy that is less than 1 on the ILSVRC2012 dataset. We also propose an efficient sparse matrix multiplication algorithm on CPU for Sparse Convolutional Neural Networks (SCNN) models. Our CPU implementation demonstrates much higher efficiency than the off-the-shelf sparse matrix libraries, with a significant speedup realized over the original dense network. In addition, we apply the SCNN model to the object detection problem, in conjunction with a cascade model and sparse fully connected layers, to achieve significant speedups.",
"In this paper, we introduce a new channel pruning method to accelerate very deep convolutional neural networks. Given a trained CNN model, we propose an iterative two-step algorithm to effectively prune each layer, by a LASSO regression based channel selection and least square reconstruction. We further generalize this algorithm to multi-layer and multi-branch cases. Our method reduces the accumulated error and enhance the compatibility with various architectures. Our pruned VGG-16 achieves the state-of-the-art results by 5× speed-up along with only 0.3 increase of error. More importantly, our method is able to accelerate modern networks like ResNet, Xception and suffers only 1.4 , 1.0 accuracy loss under 2× speedup respectively, which is significant.",
"",
"High demand for computation resources severely hinders deployment of large-scale Deep Neural Networks (DNN) in resource constrained devices. In this work, we propose a Structured Sparsity Learning (SSL) method to regularize the structures (i.e., filters, channels, filter shapes, and layer depth) of DNNs. SSL can: (1) learn a compact structure from a bigger DNN to reduce computation cost; (2) obtain a hardware-friendly structured sparsity of DNN to efficiently accelerate the DNNs evaluation. Experimental results show that SSL achieves on average 5.1x and 3.1x speedups of convolutional layer computation of AlexNet against CPU and GPU, respectively, with off-the-shelf libraries. These speedups are about twice speedups of non-structured sparsity; (3) regularize the DNN structure to improve classification accuracy. The results show that for CIFAR-10, regularization on layer depth can reduce 20 layers of a Deep Residual Network (ResNet) to 18 layers while improve the accuracy from 91.25 to 92.60 , which is still slightly higher than that of original ResNet with 32 layers. For AlexNet, structure regularization by SSL also reduces the error by around 1 . Open source code is in this https URL",
"The success of CNNs in various applications is accompanied by a significant increase in the computation and parameter storage costs. Recent efforts toward reducing these overheads involve pruning and compressing the weights of various layers without hurting original accuracy. However, magnitude-based pruning of weights reduces a significant number of parameters from the fully connected layers and may not adequately reduce the computation costs in the convolutional layers due to irregular sparsity in the pruned networks. We present an acceleration method for CNNs, where we prune filters from CNNs that are identified as having a small effect on the output accuracy. By removing whole filters in the network together with their connecting feature maps, the computation costs are reduced significantly. In contrast to pruning weights, this approach does not result in sparse connectivity patterns. Hence, it does not need the support of sparse convolution libraries and can work with existing efficient BLAS libraries for dense matrix multiplications. We show that even simple filter pruning techniques can reduce inference costs for VGG-16 by up to 34 and ResNet-110 by up to 38 on CIFAR10 while regaining close to the original accuracy by retraining the networks."
]
} |
1905.11533 | 2947944300 | Today a canonical approach to reduce the computation cost of Deep Neural Networks (DNNs) is to pre-define an over-parameterized model before training to guarantee the learning capacity, and then prune unimportant learning units (filters and neurons) during training to improve model compactness. We argue it is unnecessary to introduce redundancy at the beginning of the training but then reduce redundancy for the ultimate inference model. In this paper, we propose a Continuous Growth and Pruning (CGaP) scheme to minimize the redundancy from the beginning. CGaP starts the training from a small network seed, then expands the model continuously by reinforcing important learning units, and finally prunes the network to obtain a compact and accurate model. As the growth phase favors important learning units, CGaP provides a clear learning purpose to the pruning phase. Experimental results on representative datasets and DNN architectures demonstrate that CGaP outperforms previous pruning-only approaches that deal with pre-defined structures. For VGG-19 on CIFAR-100 and SVHN datasets, CGaP reduces the number of parameters by 78.9 and 85.8 , FLOPs by 53.2 and 74.2 , respectively; For ResNet-110 On CIFAR-10, CGaP reduces 64.0 number of parameters and 63.3 FLOPs. | On the other hand, constructive methods add connections or neurons to increase capacity. @cite_3 @cite_14 enlarged network capacity with fresh neurons and evaluated rudimentary problems such as XOR problems and multi-layer perceptron (MLP), without datasets from a real scenario. Unlike their work, we expand the network with elaborately selected units and validate on advanced DNNs and datasets. @cite_18 picked a set of convolutional filters from a bundle of randomly generated ones to enlarge the convolutional layers. However, how to find one set of filters that reduces the most loss among several sets is by trial-and-error method, consuming a significant amount of resources. Compared to it, directly growing up the convolutional layers from one seed, like CGaP does, is more efficient in real-world applications. To our knowledge, CGaP is the first scheme to employ continuous saliency-based growth for the purpose of efficient deep learning. | {
"cite_N": [
"@cite_18",
"@cite_14",
"@cite_3"
],
"mid": [
"2768083806",
"134986842",
"2042481239"
],
"abstract": [
"Deep neural networks (DNNs) have begun to have a pervasive impact on various applications of machine learning. However, the problem of finding an optimal DNN architecture for large applications is challenging. Common approaches go for deeper and larger DNN architectures but may incur substantial redundancy. To address these problems, we introduce a network growth algorithm that complements network pruning to learn both weights and compact DNN architectures during training. We propose a DNN synthesis tool (NeST) that combines both methods to automate the generation of compact and accurate DNNs. NeST starts with a randomly initialized sparse network called the seed architecture. It iteratively tunes the architecture with gradient-based growth and magnitude-based pruning of neurons and connections. Our experimental results show that NeST yields accurate, yet very compact DNNs, with a wide range of seed architecture selection. For the LeNet-300-100 (LeNet-5) architecture, we reduce network parameters by 70.2x (74.3x) and floating-point operations (FLOPs) by 79.4x (43.7x). For the AlexNet and VGG-16 architectures, we reduce network parameters (FLOPs) by 15.7x (4.6x) and 30.2x (8.6x), respectively. NeST's grow-and-prune paradigm delivers significant additional parameter and FLOPs reduction relative to pruning-only methods.",
"",
"Abstract This paper introduces a new method called Dynamic Node Creation (DNC) which automatically grows BP networks until the target problem is solved. DNC sequentially adds nodes one at a time to the hidden layer(s) of the network until the desired approximation accuracy is achieved. Simulation results for parity, symmetry, binary addition, and the encoder problem are presented. The procedure was capable of finding known minimal topologies in many cases, and was always within three nodes of the minimum. Computational expense for finding the solutions was comparable to training normal BP networks with the same final topologies. Starting out with fewer nodes than needed to solve the problem actually seems to help find a solution. The method yielded a solution for every problem tried."
]
} |
1905.11533 | 2947944300 | Today a canonical approach to reduce the computation cost of Deep Neural Networks (DNNs) is to pre-define an over-parameterized model before training to guarantee the learning capacity, and then prune unimportant learning units (filters and neurons) during training to improve model compactness. We argue it is unnecessary to introduce redundancy at the beginning of the training but then reduce redundancy for the ultimate inference model. In this paper, we propose a Continuous Growth and Pruning (CGaP) scheme to minimize the redundancy from the beginning. CGaP starts the training from a small network seed, then expands the model continuously by reinforcing important learning units, and finally prunes the network to obtain a compact and accurate model. As the growth phase favors important learning units, CGaP provides a clear learning purpose to the pruning phase. Experimental results on representative datasets and DNN architectures demonstrate that CGaP outperforms previous pruning-only approaches that deal with pre-defined structures. For VGG-19 on CIFAR-100 and SVHN datasets, CGaP reduces the number of parameters by 78.9 and 85.8 , FLOPs by 53.2 and 74.2 , respectively; For ResNet-110 On CIFAR-10, CGaP reduces 64.0 number of parameters and 63.3 FLOPs. | Orthogonal methods to our work include low-precision quantization and low-rank decomposition. Low-precision quantization @cite_27 @cite_2 quantized the parameter and gradients to fewer bits and thus reduced the memory size and access. @cite_23 @cite_24 compressed each convolutional layer by finding an appropriate low-rank approximation. It is worth mentioning that our CGaP scheme can leverage the aforementioned orthogonal methods to further compress and accelerate the sparse model. | {
"cite_N": [
"@cite_24",
"@cite_27",
"@cite_23",
"@cite_2"
],
"mid": [
"2962833442",
"1724438581",
"2167215970",
"2524428287"
],
"abstract": [
"",
"Deep convolutional neural networks (CNN) has become the most promising method for object recognition, repeatedly demonstrating record breaking results for image classification and object detection in recent years. However, a very deep CNN generally involves many layers with millions of parameters, making the storage of the network model to be extremely large. This prohibits the usage of deep CNNs on resource limited hardware, especially cell phones or other embedded devices. In this paper, we tackle this model storage issue by investigating information theoretical vector quantization methods for compressing the parameters of CNNs. In particular, we have found in terms of compressing the most storage demanding dense connected layers, vector quantization methods have a clear gain over existing matrix factorization methods. Simply applying k-means clustering to the weights or conducting product quantization can lead to a very good balance between model size and recognition accuracy. For the 1000-category classification task in the ImageNet challenge, we are able to achieve 16-24 times compression of the network with only 1 loss of classification accuracy using the state-of-the-art CNN.",
"We present techniques for speeding up the test-time evaluation of large convolutional networks, designed for object recognition tasks. These models deliver impressive accuracy, but each image evaluation requires millions of floating point operations, making their deployment on smartphones and Internet-scale clusters problematic. The computation is dominated by the convolution operations in the lower layers of the model. We exploit the redundancy present within the convolutional filters to derive approximations that significantly reduce the required computation. Using large state-of-the-art models, we demonstrate speedups of convolutional layers on both CPU and GPU by a factor of 2 x, while keeping the accuracy within 1 of the original model.",
"We introduce a method to train Quantized Neural Networks (QNNs) -- neural networks with extremely low precision (e.g., 1-bit) weights and activations, at run-time. At traintime the quantized weights and activations are used for computing the parameter gradients. During the forward pass, QNNs drastically reduce memory size and accesses, and replace most arithmetic operations with bit-wise operations. As a result, power consumption is expected to be drastically reduced. We trained QNNs over the MNIST, CIFAR-10, SVHN and ImageNet datasets. The resulting QNNs achieve prediction accuracy comparable to their 32-bit counterparts. For example, our quantized version of AlexNet with 1-bit weights and 2-bit activations achieves 51 top-1 accuracy. Moreover, we quantize the parameter gradients to 6-bits as well which enables gradients computation using only bit-wise operation. Quantized recurrent neural networks were tested over the Penn Treebank dataset, and achieved comparable accuracy as their 32-bit counterparts using only 4-bits. Last but not least, we programmed a binary matrix multiplication GPU kernel with which it is possible to run our MNIST QNN 7 times faster than with an unoptimized GPU kernel, without suffering any loss in classification accuracy. The QNN code is available online."
]
} |
1905.11528 | 2947593239 | As the complexity of neural network models has grown, it has become increasingly important to optimize their design automatically through met alearning. Methods for discovering hyperparameters, topologies, and learning rate schedules have lead to significant increases in performance. This paper shows that loss functions can be optimized with met alearning as well, and result in similar improvements. The method, Genetic Loss-function Optimization (GLO), discovers loss functions de novo, and optimizes them for a target task. Leveraging techniques from genetic programming, GLO builds loss functions hierarchically from a set of operators and leaf nodes. These functions are repeatedly recombined and mutated to find an optimal structure, and then a covariance-matrix adaptation evolutionary strategy (CMA-ES) is used to find optimal coefficients. Networks trained with GLO loss functions are found to outperform the standard cross-entropy loss on standard image classification tasks. Training with these new loss functions requires fewer steps, results in lower test error, and allows for smaller datasets to be used. Loss-function optimization thus provides a new dimension of met alearning, and constitutes an important step towards AutoML. | In addition to hyperparameter optimization and neural architecture search, new opportunities for met alearning have recently emerged. In particular, learning rate scheduling and adaptation can have a significant impact on a model's performance. Learning rate schedules determine how the learning rate changes as training progresses. This functionality tends to be encapsulated away in practice by different gradient-descent optimizers, such as AdaGrad @cite_9 and Adam @cite_20 . While the general consensus has been that monotonically decreasing learning rates yield good results, new ideas, such as cyclical learning rates @cite_1 , have shown promise in learning better models in fewer epochs. | {
"cite_N": [
"@cite_9",
"@cite_1",
"@cite_20"
],
"mid": [
"2146502635",
"2964054038",
"1522301498"
],
"abstract": [
"We present a new family of subgradient methods that dynamically incorporate knowledge of the geometry of the data observed in earlier iterations to perform more informative gradient-based learning. Metaphorically, the adaptation allows us to find needles in haystacks in the form of very predictive but rarely seen features. Our paradigm stems from recent advances in stochastic optimization and online learning which employ proximal functions to control the gradient steps of the algorithm. We describe and analyze an apparatus for adaptively modifying the proximal function, which significantly simplifies setting a learning rate and results in regret guarantees that are provably as good as the best proximal function that can be chosen in hindsight. We give several efficient algorithms for empirical risk minimization problems with common and important regularization functions and domain constraints. We experimentally study our theoretical analysis and show that adaptive subgradient methods outperform state-of-the-art, yet non-adaptive, subgradient algorithms.",
"It is known that the learning rate is the most important hyper-parameter to tune for training deep neural networks. This paper describes a new method for setting the learning rate, named cyclical learning rates, which practically eliminates the need to experimentally find the best values and schedule for the global learning rates. Instead of monotonically decreasing the learning rate, this method lets the learning rate cyclically vary between reasonable boundary values. Training with cyclical learning rates instead of fixed values achieves improved classification accuracy without a need to tune and often in fewer iterations. This paper also describes a simple way to estimate \"reasonable bounds\" – linearly increasing the learning rate of the network for a few epochs. In addition, cyclical learning rates are demonstrated on the CIFAR-10 and CIFAR-100 datasets with ResNets, Stochastic Depth networks, and DenseNets, and the ImageNet dataset with the AlexNet and GoogLeNet architectures. These are practical tools for everyone who trains neural networks.",
"We introduce Adam, an algorithm for first-order gradient-based optimization of stochastic objective functions, based on adaptive estimates of lower-order moments. The method is straightforward to implement, is computationally efficient, has little memory requirements, is invariant to diagonal rescaling of the gradients, and is well suited for problems that are large in terms of data and or parameters. The method is also appropriate for non-stationary objectives and problems with very noisy and or sparse gradients. The hyper-parameters have intuitive interpretations and typically require little tuning. Some connections to related algorithms, on which Adam was inspired, are discussed. We also analyze the theoretical convergence properties of the algorithm and provide a regret bound on the convergence rate that is comparable to the best known results under the online convex optimization framework. Empirical results demonstrate that Adam works well in practice and compares favorably to other stochastic optimization methods. Finally, we discuss AdaMax, a variant of Adam based on the infinity norm."
]
} |
1905.11528 | 2947593239 | As the complexity of neural network models has grown, it has become increasingly important to optimize their design automatically through met alearning. Methods for discovering hyperparameters, topologies, and learning rate schedules have lead to significant increases in performance. This paper shows that loss functions can be optimized with met alearning as well, and result in similar improvements. The method, Genetic Loss-function Optimization (GLO), discovers loss functions de novo, and optimizes them for a target task. Leveraging techniques from genetic programming, GLO builds loss functions hierarchically from a set of operators and leaf nodes. These functions are repeatedly recombined and mutated to find an optimal structure, and then a covariance-matrix adaptation evolutionary strategy (CMA-ES) is used to find optimal coefficients. Networks trained with GLO loss functions are found to outperform the standard cross-entropy loss on standard image classification tasks. Training with these new loss functions requires fewer steps, results in lower test error, and allows for smaller datasets to be used. Loss-function optimization thus provides a new dimension of met alearning, and constitutes an important step towards AutoML. | Met alearning methods have also been recently developed for data augmentation, such as AutoAugment @cite_25 , a reinforcement learning based approach to find new data augmentation policies. In reinforcement learning tasks, EC has proven a successful approach. For instance, in evolving policy gradients @cite_22 , the policy loss is not represented symbolically, but rather as a neural network that convolves over a temporal sequence of context vectors. In reward function search @cite_26 , the task is framed as a genetic programming problem, leveraging PushGP @cite_6 . | {
"cite_N": [
"@cite_22",
"@cite_26",
"@cite_25",
"@cite_6"
],
"mid": [
"2964227899",
"2123408238",
"2804047946",
""
],
"abstract": [
"We propose a met alearning approach for learning gradient-based reinforcement learning (RL) algorithms. The idea is to evolve a differentiable loss function, such that an agent, which optimizes its policy to minimize this loss, will achieve high rewards. The loss is parametrized via temporal convolutions over the agent's experience. Because this loss is highly flexible in its ability to take into account the agent's history, it enables fast task learning. Empirical results show that our evolved policy gradient algorithm (EPG) achieves faster learning on several randomized environments compared to an off-the-shelf policy gradient method. We also demonstrate that EPG's learned loss can generalize to out-of-distribution test time tasks, and exhibits qualitatively different behavior from other popular met alearning algorithms.",
"Reward functions in reinforcement learning have largely been assumed given as part of the problem being solved by the agent. However, the psychological notion of intrinsic motivation has recently inspired inquiry into whether there exist alternate reward functions that enable an agent to learn a task more easily than the natural task-based reward function allows. This paper presents a genetic programming algorithm to search for alternate reward functions that improve agent learning performance. We present experiments that show the superiority of these reward functions, demonstrate the possible scalability of our method, and define three classes of problems where reward function search might be particularly useful: distributions of environments, nonstationary environments, and problems with short agent lifetimes.",
"Data augmentation is an effective technique for improving the accuracy of modern image classifiers. However, current data augmentation implementations are manually designed. In this paper, we describe a simple procedure called AutoAugment to automatically search for improved data augmentation policies. In our implementation, we have designed a search space where a policy consists of many sub-policies, one of which is randomly chosen for each image in each mini-batch. A sub-policy consists of two operations, each operation being an image processing function such as translation, rotation, or shearing, and the probabilities and magnitudes with which the functions are applied. We use a search algorithm to find the best policy such that the neural network yields the highest validation accuracy on a target dataset. Our method achieves state-of-the-art accuracy on CIFAR-10, CIFAR-100, SVHN, and ImageNet (without additional data). On ImageNet, we attain a Top-1 accuracy of 83.5 which is 0.4 better than the previous record of 83.1 . On CIFAR-10, we achieve an error rate of 1.5 , which is 0.6 better than the previous state-of-the-art. Augmentation policies we find are transferable between datasets. The policy learned on ImageNet transfers well to achieve significant improvements on other datasets, such as Oxford Flowers, Caltech-101, Oxford-IIT Pets, FGVC Aircraft, and Stanford Cars.",
""
]
} |
1905.11373 | 2945814190 | We consider a new extension of the extragradient method that is motivated by approximating implicit updates. Since in a recent work chavdarova2019reducing it was shown that the existing stochastic extragradient algorithm (called mirror-prox) of juditsky2011solving diverges on a simple bilinear problem, we prove guarantees for solving variational inequality that are more general than in juditsky2011solving . Furthermore, we illustrate numerically that the proposed variant converges faster than many other methods on the example of chavdarova2019reducing . We also discuss how extragradient can be applied to training Generative Adversarial Networks (GANs). Our experiments on GANs demonstrate that the introduced approach may make the training faster in terms of data passes, while its higher iteration complexity makes the advantage smaller. To further accelerate method's convergence on problems such as bilinear minimax, we combine the extragradient step with negative momentum gidel2018negative and discuss the optimal momentum value. | There are other techniques that allow to improve stability and achieve convergence for monotone operators. While gradient descent-ascent does not, in general, converge to a solution @cite_16 , the negative momentum trick proposed in @cite_16 can fix this. | {
"cite_N": [
"@cite_16"
],
"mid": [
"2848500096"
],
"abstract": [
"Games generalize the single-objective optimization paradigm by introducing different objective functions for different players. Differentiable games often proceed by simultaneous or alternating gradient updates. In machine learning, games are gaining new importance through formulations like generative adversarial networks (GANs) and actor-critic systems. However, compared to single-objective optimization, game dynamics are more complex and less understood. In this paper, we analyze gradient-based methods with momentum on simple games. We prove that alternating updates are more stable than simultaneous updates. Next, we show both theoretically and empirically that alternating gradient updates with a negative momentum term achieves convergence in a difficult toy adversarial problem, but also on the notoriously difficult to train saturating GANs."
]
} |
1905.11373 | 2945814190 | We consider a new extension of the extragradient method that is motivated by approximating implicit updates. Since in a recent work chavdarova2019reducing it was shown that the existing stochastic extragradient algorithm (called mirror-prox) of juditsky2011solving diverges on a simple bilinear problem, we prove guarantees for solving variational inequality that are more general than in juditsky2011solving . Furthermore, we illustrate numerically that the proposed variant converges faster than many other methods on the example of chavdarova2019reducing . We also discuss how extragradient can be applied to training Generative Adversarial Networks (GANs). Our experiments on GANs demonstrate that the introduced approach may make the training faster in terms of data passes, while its higher iteration complexity makes the advantage smaller. To further accelerate method's convergence on problems such as bilinear minimax, we combine the extragradient step with negative momentum gidel2018negative and discuss the optimal momentum value. | This work is not the first to consider a variant of stochastic extragradient. A stochastic version of the mirror-prox method @cite_9 was analyzed in @cite_19 under pretty restrictive assumptions. While deterministic extragradient approximates implicit update, the authors of @cite_19 chose to sample two different instances of the stochastic operator, which leads to a poor approximation of stochastic implicit update unless the variance is tiny. It was observed in @cite_14 that this approach leads to terrible practical performance, dubious convergence guarantees and divergence on bilinear problems. All later variants of stochastic extragradient, that we are aware of, consider the same update model. | {
"cite_N": [
"@cite_19",
"@cite_9",
"@cite_14"
],
"mid": [
"2964121860",
"2089559088",
"2939856750"
],
"abstract": [
"In this paper we consider iterative methods for stochastic variational inequalities (s.v.i.) with monotone operators. Our basic assumption is that the operator possesses both smooth and nonsmooth components. Further, only noisy observations of the problem data are available. We develop a novel Stochastic Mirror-Prox (SMP) algorithm for solving s.v.i. and show that with the convenient stepsize strategy it attains the optimal rates of convergence with respect to the problem parameters. We apply the SMP algorithm to Stochastic composite minimization and describe particular applications to Stochastic Semidefinite Feasibility problem and deterministic Eigenvalue minimization.",
"We propose a prox-type method with efficiency estimate @math for approximating saddle points of convex-concave C @math functions and solutions of variational inequalities with monotone Lipschitz continuous operators. Application examples include matrix games, eigenvalue minimization, and computing the Lovasz capacity number of a graph, and these are illustrated by numerical experiments with large-scale matrix games and Lovasz capacity problems.",
"We study the effect of the stochastic gradient noise on the training of generative adversarial networks (GANs) and show that it can prevent the convergence of standard game optimization methods, while the batch version converges. We address this issue with a novel stochastic variance-reduced extragradient (SVRE) optimization algorithm that improves upon the best convergence rates proposed in the literature. We observe empirically that SVRE performs similarly to a batch method on MNIST while being computationally cheaper, and that SVRE yields more stable GAN training on standard datasets."
]
} |
1905.11373 | 2945814190 | We consider a new extension of the extragradient method that is motivated by approximating implicit updates. Since in a recent work chavdarova2019reducing it was shown that the existing stochastic extragradient algorithm (called mirror-prox) of juditsky2011solving diverges on a simple bilinear problem, we prove guarantees for solving variational inequality that are more general than in juditsky2011solving . Furthermore, we illustrate numerically that the proposed variant converges faster than many other methods on the example of chavdarova2019reducing . We also discuss how extragradient can be applied to training Generative Adversarial Networks (GANs). Our experiments on GANs demonstrate that the introduced approach may make the training faster in terms of data passes, while its higher iteration complexity makes the advantage smaller. To further accelerate method's convergence on problems such as bilinear minimax, we combine the extragradient step with negative momentum gidel2018negative and discuss the optimal momentum value. | Surprisingly, a variant of extragradient was also rediscovered by practitioners @cite_10 as a way to stabilize training of GANs. The main different of the method in @cite_10 to what we consider is in applying extrasteps only on one of two neural networks. In addition, @cite_10 proposed to use more than one extra step and claim that in on specific problems 5 steps is a good trade-off between results quality and computation. | {
"cite_N": [
"@cite_10"
],
"mid": [
"2953246223"
],
"abstract": [
"We introduce a method to stabilize Generative Adversarial Networks (GANs) by defining the generator objective with respect to an unrolled optimization of the discriminator. This allows training to be adjusted between using the optimal discriminator in the generator's objective, which is ideal but infeasible in practice, and using the current value of the discriminator, which is often unstable and leads to poor solutions. We show how this technique solves the common problem of mode collapse, stabilizes training of GANs with complex recurrent generators, and increases diversity and coverage of the data distribution by the generator."
]
} |
1905.11472 | 2947094797 | Latent fingerprint recognition is not a new topic but it has attracted a lot of attention from researchers in both academia and industry over the past 50 years. With the rapid development of pattern recognition techniques, automated fingerprint identification systems (AFIS) have become more and more ubiquitous. However, most AFIS are utilized for live-scan or rolled slap prints while only a few systems can work on latent fingerprints with reasonable accuracy. The question of whether taking higher resolution scans of latent fingerprints and their rolled slap mate prints could help improve the identification accuracy still remains an open question in the forensic community. Because pores are one of the most reliable features besides minutiae to identify latent fingerprints, we propose an end-to-end automatic pore extraction and matching system to analyze the utility of pores in latent fingerprint identification. Hence, this paper answers two questions in the latent fingerprint domain: (i) does the incorporation of pores as level-3 features improve the system performance significantly? and (ii) does the 1,000 ppi image resolution improve the recognition results? We believe that our proposed end-to-end pore extraction and matching system will be a concrete baseline for future latent AFIS development. | An end-to-end AFIS contains two stages: feature extraction and matching. The feature extraction stage converts input fingerprint images into corresponding templates while the matching stage uses those templates to compute a similarity score between the input fingerprint templates, resulting in a ranked list of candidate matches. Although many studies have tried to address level 3 feature extraction and matching on fingerprints, these only used high resolution images with no background noise @cite_31 or scanned fingerprints. | {
"cite_N": [
"@cite_31"
],
"mid": [
"2133393105"
],
"abstract": [
"Sweat pores on fingerprints have proven to be useful features for personal identification. Several methods have been proposed for pore matching. The state-of-the-art method first matches minutiae on the fingerprints and then matches the pores based on the minutia matching results. A problem of such minutia-based pore matching method is that the pore matching is dependent on the minutia matching. Such dependency limits the pore matching performance and impairs the effectiveness of the fusion of minutia and pore match scores. In this paper, we propose a novel direct approach for matching fingerprint pores. It first determines the correspondences between pores based on their local features. It then uses the RANSAC (RANdom SAmple Consensus) algorithm to refine the pore correspondences obtained in the first step. A similarity score is finally calculated based on the pore matching results. The proposed pore matching method successfully avoids the dependency of pore matching on minutia matching results. Experiments have shown that the fingerprint recognition accuracy can be greatly improved by using the method proposed in this paper."
]
} |
1905.11472 | 2947094797 | Latent fingerprint recognition is not a new topic but it has attracted a lot of attention from researchers in both academia and industry over the past 50 years. With the rapid development of pattern recognition techniques, automated fingerprint identification systems (AFIS) have become more and more ubiquitous. However, most AFIS are utilized for live-scan or rolled slap prints while only a few systems can work on latent fingerprints with reasonable accuracy. The question of whether taking higher resolution scans of latent fingerprints and their rolled slap mate prints could help improve the identification accuracy still remains an open question in the forensic community. Because pores are one of the most reliable features besides minutiae to identify latent fingerprints, we propose an end-to-end automatic pore extraction and matching system to analyze the utility of pores in latent fingerprint identification. Hence, this paper answers two questions in the latent fingerprint domain: (i) does the incorporation of pores as level-3 features improve the system performance significantly? and (ii) does the 1,000 ppi image resolution improve the recognition results? We believe that our proposed end-to-end pore extraction and matching system will be a concrete baseline for future latent AFIS development. | With the advancement of capture technology, obtaining high resolution images is not an obstacle in forensic community. One of the pioneers in the extraction of pores as fingerprint features is Ray al @cite_32 , who used a modified minimum squared error approach on the NIST 4 @cite_22 dataset. Jain al @cite_30 developed their techniques based on high quality scanned partial fingerprint images. Zhao al @cite_14 proposed adaptive modeling for sweat pore extraction. They divided the image into small blocks and constructed local pore models for each region. | {
"cite_N": [
"@cite_30",
"@cite_22",
"@cite_14",
"@cite_32"
],
"mid": [
"2158359334",
"",
"2048425530",
"2119702682"
],
"abstract": [
"Fingerprint friction ridge details are generally described in a hierarchical order at three different levels, namely, level 1 (pattern), level 2 (minutia points), and level 3 (pores and ridge contours). Although latent print examiners frequently take advantage of level 3 features to assist in identification, automated fingerprint identification systems (AFIS) currently rely only on level 1 and level 2 features. In fact, the Federal Bureau of Investigation's (FBI) standard of fingerprint resolution for AFIS is 500 pixels per inch (ppi), which is inadequate for capturing level 3 features, such as pores. With the advances in fingerprint sensing technology, many sensors are now equipped with dual resolution (500 ppi 1,000 ppi) scanning capability. However, increasing the scan resolution alone does not necessarily provide any performance improvement in fingerprint matching, unless an extended feature set is utilized. As a result, a systematic study to determine how much performance gain one can achieve by introducing level 3 features in AFIS is highly desired. We propose a hierarchical matching system that utilizes features at all the three levels extracted from 1,000 ppi fingerprint scans. Level 3 features, including pores and ridge contours, are automatically extracted using Gabor filters and wavelet transform and are locally matched using the iterative closest point (ICP) algorithm. Our experiments show that level 3 features carry significant discriminatory information. There is a relative reduction of 20 percent in the equal error rate (EER) of the matching system when level 3 features are employed in combination with level 1 and 2 features. This significant performance gain is consistently observed across various quality fingerprint images",
"",
"Sweat pores on fingerprints have proven to be discriminative features and have recently been successfully employed in automatic fingerprint recognition systems (AFRS), where the extraction of fingerprint pores is a critical step. Most of the existing pore extraction methods detect pores by using a static isotropic pore model; however, their detection accuracy is not satisfactory due to the limited approximation capability of static isotropic models to various types of pores. This paper presents a dynamic anisotropic pore model to describe pores more accurately by using orientation and scale parameters. An adaptive pore extraction method is then developed based on the proposed dynamic anisotropic pore model. The fingerprint image is first partitioned into well-defined, ill-posed, and background blocks. According to the dominant ridge orientation and frequency on each foreground block, a local instantiation of appropriate pore model is obtained. Finally, the pores are extracted by filtering the block with the adaptively generated pore model. Extensive experiments are performed on the high resolution fingerprint databases we established. The results demonstrate that the proposed method can detect pores more accurately and robustly, and consequently improve the fingerprint recognition accuracy of pore-based AFRS.",
"Automated fingerprint identification systems (AFIS) have become a popular tool in many security and law enforcement applications. Most of these systems rely on the matching of fingerprints using the position and orientation of ridge endings and bifurcations within the fingerprint image. While this information is sufficient for matching fingerprints in small databases, it is not discriminatory enough to provide good results on large collections of fingerprint images. This paper presents a means of obtaining additional discriminatory information from fingerprint images by demonstrating a novel method to extract the locations of sweat pores from the grayscale fingerprint image. This is achieved through the implementation of a modified minimum squared error approach. The proposed algorithm is capable of obtaining good results even from images obtained from basic 500 dpi optical live-scan devices despite the common belief that images obtained at this resolution are not of high enough quality. Results of the proposed method are demonstrated on fingerprint images taken both from a common live-scan device and the inked prints of the NIST 4 database. An explanation of the approach is presented, the results are discussed, and future research possibilities are put forth."
]
} |
1905.11472 | 2947094797 | Latent fingerprint recognition is not a new topic but it has attracted a lot of attention from researchers in both academia and industry over the past 50 years. With the rapid development of pattern recognition techniques, automated fingerprint identification systems (AFIS) have become more and more ubiquitous. However, most AFIS are utilized for live-scan or rolled slap prints while only a few systems can work on latent fingerprints with reasonable accuracy. The question of whether taking higher resolution scans of latent fingerprints and their rolled slap mate prints could help improve the identification accuracy still remains an open question in the forensic community. Because pores are one of the most reliable features besides minutiae to identify latent fingerprints, we propose an end-to-end automatic pore extraction and matching system to analyze the utility of pores in latent fingerprint identification. Hence, this paper answers two questions in the latent fingerprint domain: (i) does the incorporation of pores as level-3 features improve the system performance significantly? and (ii) does the 1,000 ppi image resolution improve the recognition results? We believe that our proposed end-to-end pore extraction and matching system will be a concrete baseline for future latent AFIS development. | With the development of neural networks and high-powered GPUs, some studies have used a deep network approach to extract pores. Jang al @cite_21 proposed a pore extraction technique based on convolutional neural networks. Again, they used high resolution partial fingerprints, not latents, to evaluate their approach. Genovese al @cite_17 utilized a neural network to extract pores from live-scanned fingerprints. Labati al @cite_15 used a similar idea for pore extraction in heterogeneous fingerprint images. However, the lab-collected pseudo latent'' dataset mentioned in the paper is very different from those sourced from forensic agencies. | {
"cite_N": [
"@cite_15",
"@cite_21",
"@cite_17"
],
"mid": [
"2604762485",
"2763549164",
"2557932930"
],
"abstract": [
"Abstract Most fingerprint recognition systems use Level 1 characteristics (ridge flow, orientation, and frequency) and Level 2 features (minutiae points) to recognize individuals. Level 3 features (sweat pores, incipient ridges and ultra-thin characteristics of the ridges) are less frequently adopted because they can be extracted only from high resolution images, but they have the potential of improving all the steps of the biometric recognition process. In particular, sweat pores can be used for quality assessment, liveness detection, biometric matching in live applications, and matching of partial latent fingerprints in forensic applications. Currently, each type of fingerprint acquisition technique (touch-based, touchless, or latent) requires a different algorithm for pore extraction. In this paper, we propose the first method in the literature able to extract the coordinates of the pores from touch-based, touchless, and latent fingerprint images. Our method uses specifically designed and trained Convolutional Neural Networks (CNN) to estimate and refine the centroid of each pore. Results show that our method is feasible and achieved satisfactory accuracy for all the types of evaluated images, with a better performance with respect to the compared state-of-the-art methods.",
"As technological developments have enabled high-quality fingerprint scanning, sweat pores, one of the Level 3 features of fingerprints, have been successfully used in automatic fingerprint recognition systems (AFRS). Since the pore extraction process is a critical step for AFRS, high accuracy is required. However, it is difficult to extract the pore correctly because the pore shape depends on the person, region, and pore type. To solve the problem, we have presented a pore extraction method using deep convolutional neural networks and pore intensity refinement. The deep networks are used to detect pores in detail using a large area of a fingerprint image. We then refine the pore information by finding local maxima to identify pores with different intensities in the fingerprint image. The experimental results show that our pore extraction method performs better than the state-of-the-art methods.",
"Touchless fingerprint recognition systems are being increasingly used for a fast, hygienic, and distortion-free recognition. However, due to the greater complexity of the algorithms required for processing touchless fingerprint samples, currently only Level 1 and Level 2 features are being used for recognition, and Level 3 features are used only in touch-based optical devices with about 1000 ppi resolution. In this paper, we propose the first innovative method in the literature able to extract Level 3 features, in particular sweat pores, from fingerprint images captured with a touchless acquisition using a commercial off-the-shelf camera. The method uses image processing algorithms to extract a set of candidate sweat pores. Then, computational intelligence techniques based on neural networks are used to learn the local features of the real pores, and select only the actual sweat pores from the set of candidate points. The results show the validity of the proposed methodology, with the majority of the pores correctly extracted, indicating that a touchless fingerprint recognition using Level 3 features is feasible."
]
} |
1905.11472 | 2947094797 | Latent fingerprint recognition is not a new topic but it has attracted a lot of attention from researchers in both academia and industry over the past 50 years. With the rapid development of pattern recognition techniques, automated fingerprint identification systems (AFIS) have become more and more ubiquitous. However, most AFIS are utilized for live-scan or rolled slap prints while only a few systems can work on latent fingerprints with reasonable accuracy. The question of whether taking higher resolution scans of latent fingerprints and their rolled slap mate prints could help improve the identification accuracy still remains an open question in the forensic community. Because pores are one of the most reliable features besides minutiae to identify latent fingerprints, we propose an end-to-end automatic pore extraction and matching system to analyze the utility of pores in latent fingerprint identification. Hence, this paper answers two questions in the latent fingerprint domain: (i) does the incorporation of pores as level-3 features improve the system performance significantly? and (ii) does the 1,000 ppi image resolution improve the recognition results? We believe that our proposed end-to-end pore extraction and matching system will be a concrete baseline for future latent AFIS development. | There have been several attempts made to use pores as a feature for matching. Jain al @cite_30 designed a hand-crafted approach to utilize features from all 3 fingerprint feature levels. They used pre-defined thresholds for their matching system. Again, they only reported experimental results on their high-resolution lab-collected dataset. Chen and Jain @cite_9 extracted an extended feature set as input for the matching step. They used a commercial off-the-shelf (COTS) matcher to evaluate their feature extraction. The dataset on which they evaluated is the partial fingerprint dataset from NIST SD30 This dataset is no longer publicly available. . Zhao al @cite_31 proposed a pore matcher based on a combination of the ICP @cite_24 and RANSAC @cite_16 algorithms. Again, they only evaluated their algorithm on their high-resolution live-scanned partial fingerprint images. | {
"cite_N": [
"@cite_30",
"@cite_9",
"@cite_24",
"@cite_31",
"@cite_16"
],
"mid": [
"2158359334",
"2079129408",
"2119851068",
"2133393105",
"2085261163"
],
"abstract": [
"Fingerprint friction ridge details are generally described in a hierarchical order at three different levels, namely, level 1 (pattern), level 2 (minutia points), and level 3 (pores and ridge contours). Although latent print examiners frequently take advantage of level 3 features to assist in identification, automated fingerprint identification systems (AFIS) currently rely only on level 1 and level 2 features. In fact, the Federal Bureau of Investigation's (FBI) standard of fingerprint resolution for AFIS is 500 pixels per inch (ppi), which is inadequate for capturing level 3 features, such as pores. With the advances in fingerprint sensing technology, many sensors are now equipped with dual resolution (500 ppi 1,000 ppi) scanning capability. However, increasing the scan resolution alone does not necessarily provide any performance improvement in fingerprint matching, unless an extended feature set is utilized. As a result, a systematic study to determine how much performance gain one can achieve by introducing level 3 features in AFIS is highly desired. We propose a hierarchical matching system that utilizes features at all the three levels extracted from 1,000 ppi fingerprint scans. Level 3 features, including pores and ridge contours, are automatically extracted using Gabor filters and wavelet transform and are locally matched using the iterative closest point (ICP) algorithm. Our experiments show that level 3 features carry significant discriminatory information. There is a relative reduction of 20 percent in the equal error rate (EER) of the matching system when level 3 features are employed in combination with level 1 and 2 features. This significant performance gain is consistently observed across various quality fingerprint images",
"There are fundamental differences in the way fingerprints are compared by forensic examiners and current automatic systems. For example, while automatic systems focus mainly on the quantitative measures of fingerprint minutiae (ridge ending and bifurcation points), forensic examiners often analyze details of intrinsic ridge characteristics and relational information. This process, known as qualitative friction ridge analysis [1], includes examination of ridge shape, pores, dots, incipient ridges, etc. This explains the challenges that current automatic systems face in processing partial fingerprints, mostly seen in latents. The forensics and automatic fingerprint identification systems (AFIS) communities have been active in standardizing the definition of extended feature set, as well as quantifying the relevance and reliability of these features for automatic matching systems. CDEFFS (committee to define an extended feature set) has proposed a working draft on possible definitions and representations of extended features [2]. However, benefits of utilizing these extended features in automatic systems are not yet known. While fingerprint matching technology is quite mature for matching tenprints [3], matching partial fingerprints, especially latents, still needs a lot of improvement. We propose an algorithm to extract two major level 3 feature types, dots and incipients, based on local phase symmetry and demonstrate their effectiveness in partial print matching. Since dots and incipients can be easily encoded by forensic examiners, we believe the results of this research will have benefits to next generation identification (NGI) systems.",
"The ICP (Iterative Closest Point) algorithm is widely used for geometric alignment of three-dimensional models when an initial estimate of the relative pose is known. Many variants of ICP have been proposed, affecting all phases of the algorithm from the selection and matching of points to the minimization strategy. We enumerate and classify many of these variants, and evaluate their effect on the speed with which the correct alignment is reached. In order to improve convergence for nearly-flat meshes with small features, such as inscribed surfaces, we introduce a new variant based on uniform sampling of the space of normals. We conclude by proposing a combination of ICP variants optimized for high speed. We demonstrate an implementation that is able to align two range images in a few tens of milliseconds, assuming a good initial guess. This capability has potential application to real-time 3D model acquisition and model-based tracking.",
"Sweat pores on fingerprints have proven to be useful features for personal identification. Several methods have been proposed for pore matching. The state-of-the-art method first matches minutiae on the fingerprints and then matches the pores based on the minutia matching results. A problem of such minutia-based pore matching method is that the pore matching is dependent on the minutia matching. Such dependency limits the pore matching performance and impairs the effectiveness of the fusion of minutia and pore match scores. In this paper, we propose a novel direct approach for matching fingerprint pores. It first determines the correspondences between pores based on their local features. It then uses the RANSAC (RANdom SAmple Consensus) algorithm to refine the pore correspondences obtained in the first step. A similarity score is finally calculated based on the pore matching results. The proposed pore matching method successfully avoids the dependency of pore matching on minutia matching results. Experiments have shown that the fingerprint recognition accuracy can be greatly improved by using the method proposed in this paper.",
"A new paradigm, Random Sample Consensus (RANSAC), for fitting a model to experimental data is introduced. RANSAC is capable of interpreting smoothing data containing a significant percentage of gross errors, and is thus ideally suited for applications in automated image analysis where interpretation is based on the data provided by error-prone feature detectors. A major portion of this paper describes the application of RANSAC to the Location Determination Problem (LDP): Given an image depicting a set of landmarks with known locations, determine that point in space from which the image was obtained. In response to a RANSAC requirement, new results are derived on the minimum number of landmarks needed to obtain a solution, and algorithms are presented for computing these minimum-landmark solutions in closed form. These results provide the basis for an automatic system that can solve the LDP under difficult viewing"
]
} |
1905.11089 | 2944849037 | CoAP (Constrained Application Protocol) with block-wise transfer (BWT) option is a known protocol choice for large data transfer in general lossy IoT network environments. Lossy transmission environments on the other hand lead to CoAP resending multiple blocks, which creates overheads. To tackle this problem, we design a BWT with network coding (NC), with the goal to reducing the number of unnecessary retransmissions. The results show the reduction in the number of block retransmissions for different values of blocksize, implying the reduced transfer time. For the maximum blocksize of 1024 bytes and total probability loss of 0.5, CoAP with NC can resend up to 5 times less blocks. | The authors in @cite_5 extend BWT using NC, while authors in @cite_1 propose a scheme where multiple blocks can be retrieved by one request, focusing more on the problem of reducing latency. The goal of these schemes is to reduce communication time. Our paper focuses on another approach, combining NC and BWT based on the work in @cite_2 to reduce the amount of traffic that needs to be resent. | {
"cite_N": [
"@cite_5",
"@cite_1",
"@cite_2"
],
"mid": [
"2566572824",
"",
"2921647295"
],
"abstract": [
"This paper presents a smooth way to include Network Coding in the Constrained Application Protocol (CoAP) for large resource transmissions. Devices in the Internet of Things usually communicate using short messages with little data. In some cases, for example, requesting firmware updates, bigger resources need to be transferred. CoAP's recently finalized blockwise transfer scheme can handle large resources, but is not efficient in lossy environments. Network Coding has proven to be more error resistant. This paper demonstrates the limitations of CoAP's existing blockwise transfer scheme and presents a new approach, based on Network Coding. The evaluation compares CoAP's regular blockwise transfer to the new Network Coding extension. Measurements on an implemented client-server application with simulated losses and delay, confirm the benefits of the extension, resulting in reduced transfer durations.",
"",
"The rising number of IoT devices is accelerating the research on new solutions that will be able to efficiently deal with unreliable connectivity in highly dynamic computing applications. To improve the overall performance in IoT applications, there are multiple communication solutions available, either proprietary or open source, all of which satisfy different communication requirements. Most commonly, for this kind of communication, developers choose REST HTTP protocol as a result of its ease of use and compatibility with the existing computing infrastructure. In applications where mobility and unreliable connectivity play a significant role, ensuring a reliable exchange of data with the stateless REST HTTP protocol completely depends on the developer itself. This often means resending multiple request messages when the connection fails, constantly trying to access the service until the connection reestablishes. In order to alleviate this problem, in this paper, we combine REST HTTP with random linear network coding (RLNC) to reduce the number of additional retransmissions. We show how using RLNC with REST HTTP requests can decrease the reconnection time by reducing the additional packet retransmissions in unreliable highly dynamic scenarios."
]
} |
1905.11166 | 2944904450 | The graph parameters highway dimension and skeleton dimension were introduced to capture the properties of transportation networks. As many important optimization problems like Travelling Salesperson, Steiner Tree or @math -Center arise in such networks, it is worthwhile to study them on graphs of bounded highway or skeleton dimension. We investigate the relationships between mentioned parameters and how they are related to other important graph parameters that have been applied successfully to various optimization problems. We show that the skeleton dimension is incomparable to any of the parameters distance to linear forest, bandwidth, treewidth and highway dimension and hence, it is worthwhile to study mentioned problems also on graphs of bounded skeleton dimension. Moreover, we prove that the skeleton dimension is upper bounded by the max leaf number and that for any graph on at least three vertices there are edge weights such that both parameters are equal. Then we show that computing the highway dimension according to most recent definition is NP-hard, which answers an open question stated by Finally we prove that on graphs @math of skeleton dimension @math it is NP-hard to approximate the @math -Center problem within a factor less than @math . | We now briefly sum-up some algorithmic results in the context of optimization problems in transportation networks. Arora @cite_25 developed a general framework that enables PTASs for several geometric problems where the network is embedded in the Euclidean plane. Building upon the work of Arora, Talwar @cite_26 developed QPTASs for TSP , Steiner Tree , @math -Median and Facility Location on graphs of low (for a formal definition, see Definition ). This was improved by @cite_15 , who obtained a PTAS for TSP . As the skeleton dimension of a graph upper bounds its doubling dimension (cf. ) the aforementioned results immediately imply a PTAS for TSP and QPTASs for Steiner Tree , @math -Median and Facility Location . | {
"cite_N": [
"@cite_15",
"@cite_26",
"@cite_25"
],
"mid": [
"2515220202",
"",
"2165142526"
],
"abstract": [
"The traveling salesman problem (TSP) is among the most famous NP-hard optimization problems. We design for this problem an algorithm that for any fixed @math computes in randomized polynomial time a @math -approximation to the optimal tour in TSP instances that form an arbitrary metric space with bounded intrinsic dimension. The celebrated results of Arora [J. ACM, 45 (1998), pp. 753--782] and Mitchell [SIAM J. Comput., 28 (1999), pp. 1298--1309] prove that the above result holds in the special case of TSP in a fixed-dimensional Euclidean space. Thus, our algorithm demonstrates that the algorithmic tractability of metric TSP depends on the dimensionality of the space and not on its specific geometry. This result resolves a problem that has been open since the quasi-polynomial time algorithm of Talwar [Proceedings of the 36th Annual ACM Symposium on Theory of Computing, 2004, pp. 281--290].",
"",
"We present a polynomial time approximation scheme for Euclidean TSP in fixed dimensions. For every fixed c > 1 and given any n nodes in R 2 , a randomized version of the scheme finds a (1 + 1 c )-approximation to the optimum traveling salesman tour in O(n (log n ) O(c) ) time. When the nodes are in R d , the running time increases to O(n (log n ) (O( d c)) d-1 ). For every fixed c, d the running time is n • poly(log n ), that is nearly linear in n . The algorithmm can be derandomized, but this increases the running time by a factor O(n d ). The previous best approximation algorithm for the problem (due to Christofides) achieves a 3 2-aproximation in polynomial time. We also give similar approximation schemes for some other NP-hard Euclidean problems: Minimum Steiner Tree, k -TSP, and k -MST. (The running times of the algorithm for k -TSP and k -MST involve an additional multiplicative factor k .) The previous best approximation algorithms for all these problems achieved a constant-factor approximation. We also give efficient approximation schemes for Euclidean Min-Cost Matching, a problem that can be solved exactly in polynomial time. All our algorithms also work, with almost no modification, when distance is measured using any geometric norm (such as e p for p ≥ 1 or other Minkowski norms). They also have simple parallel (i.e., NC) implementations."
]
} |
1905.10952 | 2944978536 | Deep neural networks (DNNs) have become a widely deployed model for numerous machine learning applications. However, their fixed architecture, substantial training cost, and significant model redundancy make it difficult to efficiently update them to accommodate previously unseen data. To solve these problems, we propose an incremental learning framework based on a grow-and-prune neural network synthesis paradigm. When new data arrive, the neural network first grows new connections based on the gradients to increase the network capacity to accommodate new data. Then, the framework iteratively prunes away connections based on the magnitude of weights to enhance network compactness, and hence recover efficiency. Finally, the model rests at a lightweight DNN that is both ready for inference and suitable for future grow-and-prune updates. The proposed framework improves accuracy, shrinks network size, and significantly reduces the additional training cost for incoming data compared to conventional approaches, such as training from scratch and network fine-tuning. For the LeNet-300-100 and LeNet-5 neural network architectures derived for the MNIST dataset, the framework reduces training cost by up to 64 (63 ) and 67 (63 ) compared to training from scratch (network fine-tuning), respectively. For the ResNet-18 architecture derived for the ImageNet dataset and DeepSpeech2 for the AN4 dataset, the corresponding training cost reductions against training from scratch (network fine-tunning) are 64 (60 ) and 67 (62 ), respectively. Our derived models contain fewer network parameters but achieve higher accuracy relative to conventional baselines. | Efficient and compact neural networks Most DNNs are computationally intensive and over-parameterized @cite_39 . Several different approaches have been put forward in the literature to design efficient DNNs. These approaches include designing novel compact neural network architectures and compressing existing models. We summarize them next. | {
"cite_N": [
"@cite_39"
],
"mid": [
"2963674932"
],
"abstract": [
"Neural networks are both computationally intensive and memory intensive, making them difficult to deploy on embedded systems. Also, conventional networks fix the architecture before training starts; as a result, training cannot improve the architecture. To address these limitations, we describe a method to reduce the storage and computation required by neural networks by an order of magnitude without affecting their accuracy by learning only the important connections. Our method prunes redundant connections using a three-step method. First, we train the network to learn which connections are important. Next, we prune the unimportant connections. Finally, we retrain the network to fine tune the weights of the remaining connections. On the ImageNet dataset, our method reduced the number of parameters of AlexNet by a factor of 9x, from 61 million to 6.7 million, without incurring accuracy loss. Similar experiments with VGG-16 found that the total number of parameters can be reduced by 13x, from 138 million to 10.3 million, again with no loss of accuracy."
]
} |
1905.10952 | 2944978536 | Deep neural networks (DNNs) have become a widely deployed model for numerous machine learning applications. However, their fixed architecture, substantial training cost, and significant model redundancy make it difficult to efficiently update them to accommodate previously unseen data. To solve these problems, we propose an incremental learning framework based on a grow-and-prune neural network synthesis paradigm. When new data arrive, the neural network first grows new connections based on the gradients to increase the network capacity to accommodate new data. Then, the framework iteratively prunes away connections based on the magnitude of weights to enhance network compactness, and hence recover efficiency. Finally, the model rests at a lightweight DNN that is both ready for inference and suitable for future grow-and-prune updates. The proposed framework improves accuracy, shrinks network size, and significantly reduces the additional training cost for incoming data compared to conventional approaches, such as training from scratch and network fine-tuning. For the LeNet-300-100 and LeNet-5 neural network architectures derived for the MNIST dataset, the framework reduces training cost by up to 64 (63 ) and 67 (63 ) compared to training from scratch (network fine-tuning), respectively. For the ResNet-18 architecture derived for the ImageNet dataset and DeepSpeech2 for the AN4 dataset, the corresponding training cost reductions against training from scratch (network fine-tunning) are 64 (60 ) and 67 (62 ), respectively. Our derived models contain fewer network parameters but achieve higher accuracy relative to conventional baselines. | : Exploiting efficient building blocks and operations can significantly cut down on the DNN computation cost. For example, MobileNetV2 effectively shrinks the model size and cuts down the number of floating-point operations (FLOPs) with inverted residual building blocks @cite_10 . propose another compact CNN architecture that utilizes channel shuffle operation and depth-wise convolution @cite_12 . suggest replacing spatial convolution layers with shift-based modules that have zero FLOPs. The generated ShiftNet has substantially reduced computation and storage costs @cite_5 . Besides, automated compact architecture design also provides a promising solution @cite_37 . develop an automated architecture adaptation and search framework based on efficient performance predictors @cite_13 . The searched models deliver up to 8.5 MobileNetV2 while reducing latency. | {
"cite_N": [
"@cite_37",
"@cite_10",
"@cite_5",
"@cite_13",
"@cite_12"
],
"mid": [
"2886953980",
"2963163009",
"2963844898",
"2905692112",
"2883780447"
],
"abstract": [
"Designing convolutional neural networks (CNN) models for mobile devices is challenging because mobile models need to be small and fast, yet still accurate. Although significant effort has been dedicated to design and improve mobile models on all three dimensions, it is challenging to manually balance these trade-offs when there are so many architectural possibilities to consider. In this paper, we propose an automated neural architecture search approach for designing resource-constrained mobile CNN models. We propose to explicitly incorporate latency information into the main objective so that the search can identify a model that achieves a good trade-off between accuracy and latency. Unlike in previous work, where mobile latency is considered via another, often inaccurate proxy (e.g., FLOPS), in our experiments, we directly measure real-world inference latency by executing the model on a particular platform, e.g., Pixel phones. To further strike the right balance between flexibility and search space size, we propose a novel factorized hierarchical search space that permits layer diversity throughout the network. Experimental results show that our approach consistently outperforms state-of-the-art mobile CNN models across multiple vision tasks. On the ImageNet classification task, our model achieves 74.0 top-1 accuracy with 76ms latency on a Pixel phone, which is 1.5x faster than MobileNetV2 ( 2018) and 2.4x faster than NASNet ( 2018) with the same top-1 accuracy. On the COCO object detection task, our model family achieves both higher mAP quality and lower latency than MobileNets.",
"In this paper we describe a new mobile architecture, MobileNetV2, that improves the state of the art performance of mobile models on multiple tasks and benchmarks as well as across a spectrum of different model sizes. We also describe efficient ways of applying these mobile models to object detection in a novel framework we call SSDLite. Additionally, we demonstrate how to build mobile semantic segmentation models through a reduced form of DeepLabv3 which we call Mobile DeepLabv3. is based on an inverted residual structure where the shortcut connections are between the thin bottleneck layers. The intermediate expansion layer uses lightweight depthwise convolutions to filter features as a source of non-linearity. Additionally, we find that it is important to remove non-linearities in the narrow layers in order to maintain representational power. We demonstrate that this improves performance and provide an intuition that led to this design. Finally, our approach allows decoupling of the input output domains from the expressiveness of the transformation, which provides a convenient framework for further analysis. We measure our performance on ImageNet [1] classification, COCO object detection [2], VOC image segmentation [3]. We evaluate the trade-offs between accuracy, and number of operations measured by multiply-adds (MAdd), as well as actual latency, and the number of parameters.",
"Neural networks rely on convolutions to aggregate spatial information. However, spatial convolutions are expensive in terms of model size and computation, both of which grow quadratically with respect to kernel size. In this paper, we present a parameter-free, FLOP-free \"shift\" operation as an alternative to spatial convolutions. We fuse shifts and point-wise convolutions to construct end-to-end trainable shift-based modules, with a hyperparameter characterizing the tradeoff between accuracy and efficiency. To demonstrate the operation's efficacy, we replace ResNet's 3x3 convolutions with shift-based modules for improved CIFAR10 and CIFAR100 accuracy using 60 fewer parameters; we additionally demonstrate the operation's resilience to parameter reduction on ImageNet, outperforming ResNet family members. We finally show the shift operation's applicability across domains, achieving strong performance with fewer parameters on image classification, face verification and style transfer.",
"This paper proposes an efficient neural network (NN) architecture design methodology called Chameleon that honors given resource constraints. Instead of developing new building blocks or using computationally-intensive reinforcement learning algorithms, our approach leverages existing efficient network building blocks and focuses on exploiting hardware traits and adapting computation resources to fit target latency and or energy constraints. We formulate platform-aware NN architecture search in an optimization framework and propose a novel algorithm to search for optimal architectures aided by efficient accuracy and resource (latency and or energy) predictors. At the core of our algorithm lies an accuracy predictor built atop Gaussian Process with Bayesian optimization for iterative sampling. With a one-time building cost for the predictors, our algorithm produces state-of-the-art model architectures on different platforms under given constraints in just minutes. Our results show that adapting computation resources to building blocks is critical to model performance. Without the addition of any bells and whistles, our models achieve significant accuracy improvements against state-of-the-art hand-crafted and automatically designed architectures. We achieve 73.8 and 75.3 top-1 accuracy on ImageNet at 20ms latency on a mobile CPU and DSP. At reduced latency, our models achieve up to 8.5 (4.8 ) and 6.6 (9.3 ) absolute top-1 accuracy improvements compared to MobileNetV2 and MnasNet, respectively, on a mobile CPU (DSP), and 2.7 (4.6 ) and 5.6 (2.6 ) accuracy gains over ResNet-101 and ResNet-152, respectively, on an Nvidia GPU (Intel CPU).",
"Currently, the neural network architecture design is mostly guided by the indirect metric of computation complexity, i.e., FLOPs. However, the direct metric, e.g., speed, also depends on the other factors such as memory access cost and platform characterics. Thus, this work proposes to evaluate the direct metric on the target platform, beyond only considering FLOPs. Based on a series of controlled experiments, this work derives several practical guidelines for efficient network design. Accordingly, a new architecture is presented, called ShuffleNet V2. Comprehensive ablation experiments verify that our model is the state-of-the-art in terms of speed and accuracy tradeoff."
]
} |
1905.10952 | 2944978536 | Deep neural networks (DNNs) have become a widely deployed model for numerous machine learning applications. However, their fixed architecture, substantial training cost, and significant model redundancy make it difficult to efficiently update them to accommodate previously unseen data. To solve these problems, we propose an incremental learning framework based on a grow-and-prune neural network synthesis paradigm. When new data arrive, the neural network first grows new connections based on the gradients to increase the network capacity to accommodate new data. Then, the framework iteratively prunes away connections based on the magnitude of weights to enhance network compactness, and hence recover efficiency. Finally, the model rests at a lightweight DNN that is both ready for inference and suitable for future grow-and-prune updates. The proposed framework improves accuracy, shrinks network size, and significantly reduces the additional training cost for incoming data compared to conventional approaches, such as training from scratch and network fine-tuning. For the LeNet-300-100 and LeNet-5 neural network architectures derived for the MNIST dataset, the framework reduces training cost by up to 64 (63 ) and 67 (63 ) compared to training from scratch (network fine-tuning), respectively. For the ResNet-18 architecture derived for the ImageNet dataset and DeepSpeech2 for the AN4 dataset, the corresponding training cost reductions against training from scratch (network fine-tunning) are 64 (60 ) and 67 (62 ), respectively. Our derived models contain fewer network parameters but achieve higher accuracy relative to conventional baselines. | : Apart from compact architecture design, compressing and simplifying existing models have also emerged as a promising approach @cite_27 . By removing redundant connections and neurons, network pruning has been shown to be very successful at DNN compression. For instance, have shown that more than 92 @cite_39 . A recent work combines pruning with network growth and improves the compression ratio of VGG-16 by another 2.5 @math @cite_32 . Furthermore, structured sparsity and pruning can reduce the run-time latency significantly @cite_1 . For example, achieve @math latency reduction for speech recognition on an Nvidia GPU based on hardware-aware structured column and row sparsity @cite_25 . Besides, low-bit quantization is another powerful tool for reducing the storage cost @cite_17 . For instance, show that replacing a full-precision (32-bit) weight representation with ternary weight quantization only incurs a minor accuracy loss for ResNet-18, but significantly reduces the storage and memory costs @cite_19 . | {
"cite_N": [
"@cite_1",
"@cite_32",
"@cite_39",
"@cite_19",
"@cite_27",
"@cite_25",
"@cite_17"
],
"mid": [
"2963000224",
"2963319203",
"2963674932",
"2560017826",
"2938315635",
"2913686213",
"2119144962"
],
"abstract": [
"High demand for computation resources severely hinders deployment of large-scale Deep Neural Networks (DNN) in resource constrained devices. In this work, we propose a Structured Sparsity Learning (SSL) method to regularize the structures (i.e., filters, channels, filter shapes, and layer depth) of DNNs. SSL can: (1) learn a compact structure from a bigger DNN to reduce computation cost; (2) obtain a hardware-friendly structured sparsity of DNN to efficiently accelerate the DNN's evaluation. Experimental results show that SSL achieves on average 5.1 × and 3.1 × speedups of convolutional layer computation of AlexNet against CPU and GPU, respectively, with off-the-shelf libraries. These speedups are about twice speedups of non-structured sparsity; (3) regularize the DNN structure to improve classification accuracy. The results show that for CIFAR-10, regularization on layer depth reduces a 20-layer Deep Residual Network (ResNet) to 18 layers while improves the accuracy from 91.25 to 92.60 , which is still higher than that of original ResNet with 32 layers. For AlexNet, SSL reduces the error by 1 .",
"Deep neural networks (DNNs) have begun to have a pervasive impact on various applications of machine learning. However, the problem of finding an optimal DNN architecture for large applications is challenging. Common approaches go for deeper and larger DNN architectures but may incur substantial redundancy. To address these problems, we introduce a network growth algorithm that complements network pruning to learn both weights and compact DNN architectures during training. We propose a DNN synthesis tool (NeST) that combines both methods to automate the generation of compact and accurate DNNs. NeST starts with a randomly initialized sparse network called the seed architecture. It iteratively tunes the architecture with gradient-based growth and magnitude-based pruning of neurons and connections. Our experimental results show that NeST yields accurate, yet very compact DNNs, with a wide range of seed architecture selection. For the LeNet-300-100 (LeNet-5) architecture, we reduce network parameters by @math 70.2× ( @math 74.3×) and floating-point operations (FLOPs) by @math 79.4× ( @math 43.7×). For the AlexNet, VGG-16, and ResNet-50 architectures, we reduce network parameters (FLOPs) by @math 15.7× ( @math 4.6×), @math 33.2× ( @math 8.9×), and @math 4.1× ( @math 2.1×) respectively. NeST's grow-and-prune paradigm delivers significant additional parameter and FLOPs reduction relative to pruning-only methods.",
"Neural networks are both computationally intensive and memory intensive, making them difficult to deploy on embedded systems. Also, conventional networks fix the architecture before training starts; as a result, training cannot improve the architecture. To address these limitations, we describe a method to reduce the storage and computation required by neural networks by an order of magnitude without affecting their accuracy by learning only the important connections. Our method prunes redundant connections using a three-step method. First, we train the network to learn which connections are important. Next, we prune the unimportant connections. Finally, we retrain the network to fine tune the weights of the remaining connections. On the ImageNet dataset, our method reduced the number of parameters of AlexNet by a factor of 9x, from 61 million to 6.7 million, without incurring accuracy loss. Similar experiments with VGG-16 found that the total number of parameters can be reduced by 13x, from 138 million to 10.3 million, again with no loss of accuracy.",
"Deep neural networks are widely used in machine learning applications. However, the deployment of large neural networks models can be difficult to deploy on mobile devices with limited power budgets. To solve this problem, we propose Trained Ternary Quantization (TTQ), a method that can reduce the precision of weights in neural networks to ternary values. This method has very little accuracy degradation and can even improve the accuracy of some models (32, 44, 56-layer ResNet) on CIFAR-10 and AlexNet on ImageNet. And our AlexNet model is trained from scratch, which means it's as easy as to train normal full precision model. We highlight our trained quantization method that can learn both ternary values and ternary assignment. During inference, only ternary values (2-bit weights) and scaling factors are needed, therefore our models are nearly 16x smaller than full-precision models. Our ternary models can also be viewed as sparse binary weight networks, which can potentially be accelerated with custom circuit. Experiments on CIFAR-10 show that the ternary models obtained by trained quantization method outperform full-precision models of ResNet-32,44,56 by 0.04 , 0.16 , 0.36 , respectively. On ImageNet, our model outperforms full-precision AlexNet model by 0.3 of Top-1 accuracy and outperforms previous ternary models by 3 .",
"Artificial neural networks (ANNs) have become the driving force behind recent artificial intelligence (AI) research. An important problem with implementing a neural network is the design of its architecture. Typically, such an architecture is obtained manually by exploring its hyperparameter space and kept fixed during training. This approach is both time-consuming and inefficient. Furthermore, modern neural networks often contain millions of parameters, whereas many applications require small inference models. Also, while ANNs have found great success in big-data applications, there is also significant interest in using ANNs for medium- and small-data applications that can be run on energy-constrained edge devices. To address these challenges, we propose a neural network synthesis methodology (SCANN) that can generate very compact neural networks without loss in accuracy for small and medium-size datasets. We also use dimensionality reduction methods to reduce the feature size of the datasets, so as to alleviate the curse of dimensionality. Our final synthesis methodology consists of three steps: dataset dimensionality reduction, neural network compression in each layer, and neural network compression with SCANN. We evaluate SCANN on the medium-size MNIST dataset by comparing our synthesized neural networks to the well-known LeNet-5 baseline. Without any loss in accuracy, SCANN generates a @math smaller network than the LeNet-5 Caffe model. We also evaluate the efficiency of using dimensionality reduction alongside SCANN on nine small to medium-size datasets. Using this methodology enables us to reduce the number of connections in the network by up to @math (geometric mean: @math ), with little to no drop in accuracy. We also show that our synthesis methodology yields neural networks that are much better at navigating the accuracy vs. energy efficiency space.",
"Many long short-term memory (LSTM) applications need fast yet compact models. Neural network compression approaches, such as the grow-and-prune paradigm, have proved to be promising for cutting down network complexity by skipping insignificant weights. However, current compression strategies are mostly hardware-agnostic and network complexity reduction does not always translate into execution efficiency. In this work, we propose a hardware-guided symbiotic training methodology for compact, accurate, yet execution-efficient inference models. It is based on our observation that hardware may introduce substantial non-monotonic behavior, which we call the latency hysteresis effect, when evaluating network size vs. inference latency. This observation raises question about the mainstream smaller-dimension-is-better compression strategy, which often leads to a sub-optimal model architecture. By leveraging the hardware-impacted hysteresis effect and sparsity, we are able to achieve the symbiosis of model compactness and accuracy with execution efficiency, thus reducing LSTM latency while increasing its accuracy. We have evaluated our algorithms on language modeling and speech recognition applications. Relative to the traditional stacked LSTM architecture obtained for the Penn Treebank dataset, we reduce the number of parameters by 18.0x (30.5x) and measured run-time latency by up to 2.4x (5.2x) on Nvidia GPUs (Intel Xeon CPUs) without any accuracy degradation. For the DeepSpeech2 architecture obtained for the AN4 dataset, we reduce the number of parameters by 7.0x (19.4x), word error rate from 12.9 to 9.9 (10.4 ), and measured run-time latency by up to 1.7x (2.4x) on Nvidia GPUs (Intel Xeon CPUs). Thus, our method yields compact, accurate, yet execution-efficient inference models.",
"Neural networks are both computationally intensive and memory intensive, making them difficult to deploy on embedded systems with limited hardware resources. To address this limitation, we introduce \"deep compression\", a three stage pipeline: pruning, trained quantization and Huffman coding, that work together to reduce the storage requirement of neural networks by 35x to 49x without affecting their accuracy. Our method first prunes the network by learning only the important connections. Next, we quantize the weights to enforce weight sharing, finally, we apply Huffman coding. After the first two steps we retrain the network to fine tune the remaining connections and the quantized centroids. Pruning, reduces the number of connections by 9x to 13x; Quantization then reduces the number of bits that represent each connection from 32 to 5. On the ImageNet dataset, our method reduced the storage required by AlexNet by 35x, from 240MB to 6.9MB, without loss of accuracy. Our method reduced the size of VGG-16 by 49x from 552MB to 11.3MB, again with no loss of accuracy. This allows fitting the model into on-chip SRAM cache rather than off-chip DRAM memory. Our compression method also facilitates the use of complex neural networks in mobile applications where application size and download bandwidth are constrained. Benchmarked on CPU, GPU and mobile GPU, compressed network has 3x to 4x layerwise speedup and 3x to 7x better energy efficiency."
]
} |
1905.11137 | 2946077140 | Unlabeled video in the wild presents a valuable, yet so far unharnessed, source of information for learning vision tasks. We present the first attempt of fully self-supervised learning of object detection from subtitled videos without any manual object annotation. To this end, we use the How2 multi-modal collection of instructional videos with English subtitles. We pose the problem as learning with a weakly- and noisily-labeled data, and propose a novel training model that can confront high noise levels, and yet train a classifier to localize the object of interest in the video frames, without any manual labeling involved. We evaluate our approach on a set of 11 manually annotated objects in over 5000 frames and compare it to an existing weakly-supervised approach as baseline. Benchmark data and code will be released upon acceptance of the paper. | Synchronization between the visual and audio or text as a source for self-supervised learning has been studied before @cite_28 @cite_20 . In @cite_28 , the authors suggest a method to learn the correspondence between audio captions and images, for the task of image retrieval. A cross-modal learning in @cite_20 is used to create links between audio and video frames, in order to retrieve audio samples that fit a given silent video. Another class of works suggest to train a model based on "pretext" task for which the ground-truth is available for free, such as solving a Jigsaw puzzle or image colorization @cite_26 @cite_18 . In this paper we address a different problem of self-supervised object detection from unlabeled and unconstrained videos. | {
"cite_N": [
"@cite_28",
"@cite_18",
"@cite_26",
"@cite_20"
],
"mid": [
"2556930864",
"2326925005",
"2943527153",
"2783457476"
],
"abstract": [
"Humans learn to speak before they can read or write, so why can’t computers do the same? In this paper, we present a deep neural network model capable of rudimentary spoken language acquisition using untranscribed audio training data, whose only supervision comes in the form of contextually relevant visual images. We describe the collection of our data comprised of over 120,000 spoken audio captions for the Places image dataset and evaluate our model on an image search and annotation task. We also provide some visualizations which suggest that our model is learning to recognize meaningful words within the caption spectrograms.",
"Given a grayscale photograph as input, this paper attacks the problem of hallucinating a plausible color version of the photograph. This problem is clearly underconstrained, so previous approaches have either relied on significant user interaction or resulted in desaturated colorizations. We propose a fully automatic approach that produces vibrant and realistic colorizations. We embrace the underlying uncertainty of the problem by posing it as a classification task and use class-rebalancing at training time to increase the diversity of colors in the result. The system is implemented as a feed-forward pass in a CNN at test time and is trained on over a million color images. We evaluate our algorithm using a “colorization Turing test,” asking human participants to choose between a generated and ground truth color image. Our method successfully fools humans on 32 of the trials, significantly higher than previous methods. Moreover, we show that colorization can be a powerful pretext task for self-supervised feature learning, acting as a cross-channel encoder. This approach results in state-of-the-art performance on several feature learning benchmarks.",
"Self-supervised learning aims to learn representations from the data itself without explicit manual supervision. Existing efforts ignore a crucial aspect of self-supervised learning - the ability to scale to large amount of data because self-supervision requires no manual labels. In this work, we revisit this principle and scale two popular self-supervised approaches to 100 million images. We show that by scaling on various axes (including data size and problem 'hardness'), one can largely match or even exceed the performance of supervised pre-training on a variety of tasks such as object detection, surface normal estimation (3D) and visual navigation using reinforcement learning. Scaling these methods also provides many interesting insights into the limitations of current self-supervised techniques and evaluations. We conclude that current self-supervised methods are not 'hard' enough to take full advantage of large scale data and do not seem to learn effective high level semantic representations. We also introduce an extensive benchmark across 9 different datasets and tasks. We believe that such a benchmark along with comparable evaluation settings is necessary to make meaningful progress. Code is at: this https URL.",
"The increasing amount of online videos brings several opportunities for training self-supervised neural networks. The creation of large scale datasets of videos such as the YouTube-8M allows us to deal with this large amount of data in manageable way. In this work, we find new ways of exploiting this dataset by taking advantage of the multi-modal information it provides. By means of a neural network, we are able to create links between audio and visual documents, by projecting them into a common region of the feature space, obtaining joint audio-visual embeddings. These links are used to retrieve audio samples that fit well to a given silent video, and also to retrieve images that match a given a query audio. The results in terms of Recall@K obtained over a subset of YouTube-8M videos show the potential of this unsupervised approach for cross-modal feature learning. We train embeddings for both scales and assess their quality in a retrieval problem, formulated as using the feature extracted from one modality to retrieve the most similar videos based on the features computed in the other modality."
]
} |
1905.10954 | 2809263292 | Transcribing content from structural images, e.g., writing notes from music scores, is a challenging task as not only the content objects should be recognized, but the internal structure should also be preserved. Existing image recognition methods mainly work on images with simple content (e.g., text lines with characters), but are not capable to identify ones with more complex content (e.g., structured code), which often follow a fine-grained grammar. To this end, in this paper, we propose a hierarchical Spotlight Transcribing Network (STN) framework followed by a two-stage "where-to-what'' solution. Specifically, we first decide "where-to-look'' through a novel spotlight mechanism to focus on different areas of the original image following its structure. Then, we decide "what-to-write'' by developing a GRU based network with the spotlight areas for transcribing the content accordingly. Moreover, we propose two implementations on the basis of STN, i.e., STNM and STNR, where the spotlight movement follows the Markov property and Recurrent modeling, respectively. We also design a reinforcement method to refine our STN framework by self-improving the spotlight mechanism. We conduct extensive experiments on many structural image datasets, where the results clearly demonstrate the effectiveness of STN framework. | The encoder-decoder system is a general framework, which has been applied to many applications, such as neural machine translation @cite_25 @cite_8 and image captioning @cite_31 @cite_36 . Generally, the system has two separate parts, one encoder for representing and encoding the input information into a feature vector, and one decoder for generating the output sequence according to the encoded representation. Due to its remarkable performance, many efforts have been made to apply it to scene text recognition @cite_17 , aiming at transcribing texts from natural images. Specifically, for encoder design, representative works leveraged deep CNN based networks, which have been the most popular methods due to their performance on hierarchical feature extraction @cite_4 , to learn the information encodings from images @cite_1 . Then for decoder selection, variations of recurrent neural networks (RNN), such as LSTM @cite_14 and GRU @cite_18 , were utilized to generate the output text sequence, both of which are able to preserve long-term dependencies for text representations @cite_19 . The whole architecture is end-to-end, which show the effectiveness in practice @cite_2 . | {
"cite_N": [
"@cite_18",
"@cite_14",
"@cite_4",
"@cite_8",
"@cite_36",
"@cite_1",
"@cite_19",
"@cite_2",
"@cite_31",
"@cite_25",
"@cite_17"
],
"mid": [
"1924770834",
"",
"2161381512",
"2964308564",
"2950178297",
"1922126009",
"2402268235",
"2194187530",
"1895577753",
"2172140247",
"1607307044"
],
"abstract": [
"In this paper we compare different types of recurrent units in recurrent neural networks (RNNs). Especially, we focus on more sophisticated units that implement a gating mechanism, such as a long short-term memory (LSTM) unit and a recently proposed gated recurrent unit (GRU). We evaluate these recurrent units on the tasks of polyphonic music modeling and speech signal modeling. Our experiments revealed that these advanced recurrent units are indeed better than more traditional recurrent units such as tanh units. Also, we found GRU to be comparable to LSTM.",
"",
"Convolutional neural networks (CNN) have recently shown outstanding image classification performance in the large- scale visual recognition challenge (ILSVRC2012). The suc- cess of CNNs is attributed to their ability to learn rich mid- level image representations as opposed to hand-designed low-level features used in other image classification meth- ods. Learning CNNs, however, amounts to estimating mil- lions of parameters and requires a very large number of annotated image samples. This property currently prevents application of CNNs to problems with limited training data. In this work we show how image representations learned with CNNs on large-scale annotated datasets can be effi- ciently transferred to other visual recognition tasks with limited amount of training data. We design a method to reuse layers trained on the ImageNet dataset to compute mid-level image representation for images in the PASCAL VOC dataset. We show that despite differences in image statistics and tasks in the two datasets, the transferred rep- resentation leads to significantly improved results for object and action classification, outperforming the current state of the art on Pascal VOC 2007 and 2012 datasets. We also show promising results for object and action localization.",
"Abstract: Neural machine translation is a recently proposed approach to machine translation. Unlike the traditional statistical machine translation, the neural machine translation aims at building a single neural network that can be jointly tuned to maximize the translation performance. The models proposed recently for neural machine translation often belong to a family of encoder-decoders and consists of an encoder that encodes a source sentence into a fixed-length vector from which a decoder generates a translation. In this paper, we conjecture that the use of a fixed-length vector is a bottleneck in improving the performance of this basic encoder-decoder architecture, and propose to extend this by allowing a model to automatically (soft-)search for parts of a source sentence that are relevant to predicting a target word, without having to form these parts as a hard segment explicitly. With this new approach, we achieve a translation performance comparable to the existing state-of-the-art phrase-based system on the task of English-to-French translation. Furthermore, qualitative analysis reveals that the (soft-)alignments found by the model agree well with our intuition.",
"Inspired by recent work in machine translation and object detection, we introduce an attention based model that automatically learns to describe the content of images. We describe how we can train this model in a deterministic manner using standard backpropagation techniques and stochastically by maximizing a variational lower bound. We also show through visualization how the model is able to automatically learn to fix its gaze on salient objects while generating the corresponding words in the output sequence. We validate the use of attention with state-of-the-art performance on three benchmark datasets: Flickr8k, Flickr30k and MS COCO.",
"In this work we present an end-to-end system for text spotting--localising and recognising text in natural scene images--and text based image retrieval. This system is based on a region proposal mechanism for detection and deep convolutional neural networks for recognition. Our pipeline uses a novel combination of complementary proposal generation techniques to ensure high recall, and a fast subsequent filtering stage for improving precision. For the recognition and ranking of proposals, we train very large convolutional neural networks to perform word recognition on the whole proposal region at the same time, departing from the character classifier based systems of the past. These networks are trained solely on data produced by a synthetic text generation engine, requiring no human labelled data. Analysing the stages of our pipeline, we show state-of-the-art performance throughout. We perform rigorous experiments across a number of standard end-to-end text spotting benchmarks and text-based image retrieval datasets, showing a large improvement over all previous methods. Finally, we demonstrate a real-world application of our text spotting system to allow thousands of hours of news footage to be instantly searchable via a text query.",
"Neural networks have become increasingly popular for the task of language modeling. Whereas feed-forward networks only exploit a fixed context length to predict the next word of a sequence, conceptually, standard recurrent neural networks can take into account all of the predecessor words. On the other hand, it is well known that recurrent networks are difficult to train and therefore are unlikely to show the full potential of recurrent models. These problems are addressed by a the Long Short-Term Memory neural network architecture. In this work, we analyze this type of network on an English and a large French language modeling task. Experiments show improvements of about 8 relative in perplexity over standard recurrent neural network LMs. In addition, we gain considerable improvements in WER on top of a state-of-the-art speech recognition system.",
"Image-based sequence recognition has been a long-standing research topic in computer vision. In this paper, we investigate the problem of scene text recognition, which is among the most important and challenging tasks in image-based sequence recognition. A novel neural network architecture, which integrates feature extraction, sequence modeling and transcription into a unified framework, is proposed. Compared with previous systems for scene text recognition, the proposed architecture possesses four distinctive properties: (1) It is end-to-end trainable, in contrast to most of the existing algorithms whose components are separately trained and tuned. (2) It naturally handles sequences in arbitrary lengths, involving no character segmentation or horizontal scale normalization. (3) It is not confined to any predefined lexicon and achieves remarkable performances in both lexicon-free and lexicon-based scene text recognition tasks. (4) It generates an effective yet much smaller model, which is more practical for real-world application scenarios. The experiments on standard benchmarks, including the IIIT-5K, Street View Text and ICDAR datasets, demonstrate the superiority of the proposed algorithm over the prior arts. Moreover, the proposed algorithm performs well in the task of image-based music score recognition, which evidently verifies the generality of it.",
"Automatically describing the content of an image is a fundamental problem in artificial intelligence that connects computer vision and natural language processing. In this paper, we present a generative model based on a deep recurrent architecture that combines recent advances in computer vision and machine translation and that can be used to generate natural sentences describing an image. The model is trained to maximize the likelihood of the target description sentence given the training image. Experiments on several datasets show the accuracy of the model and the fluency of the language it learns solely from image descriptions. Our model is often quite accurate, which we verify both qualitatively and quantitatively. For instance, while the current state-of-the-art BLEU-1 score (the higher the better) on the Pascal dataset is 25, our approach yields 59, to be compared to human performance around 69. We also show BLEU-1 score improvements on Flickr30k, from 56 to 66, and on SBU, from 19 to 28. Lastly, on the newly released COCO dataset, we achieve a BLEU-4 of 27.7, which is the current state-of-the-art.",
"Neural machine translation is a relatively new approach to statistical machine translation based purely on neural networks. The neural machine translation models often consist of an encoder and a decoder. The encoder extracts a fixed-length representation from a variable-length input sentence, and the decoder generates a correct translation from this representation. In this paper, we focus on analyzing the properties of the neural machine translation using two models; RNN Encoder--Decoder and a newly proposed gated recursive convolutional neural network. We show that the neural machine translation performs relatively well on short sentences without unknown words, but its performance degrades rapidly as the length of the sentence and the number of unknown words increase. Furthermore, we find that the proposed gated recursive convolutional network learns a grammatical structure of a sentence automatically.",
"Full end-to-end text recognition in natural images is a challenging problem that has received much attention recently. Traditional systems in this area have relied on elaborate models incorporating carefully hand-engineered features or large amounts of prior knowledge. In this paper, we take a different route and combine the representational power of large, multilayer neural networks together with recent developments in unsupervised feature learning, which allows us to use a common framework to train highly-accurate text detector and character recognizer modules. Then, using only simple off-the-shelf methods, we integrate these two modules into a full end-to-end, lexicon-driven, scene text recognition system that achieves state-of-the-art performance on standard benchmarks, namely Street View Text and ICDAR 2003."
]
} |
1905.10954 | 2809263292 | Transcribing content from structural images, e.g., writing notes from music scores, is a challenging task as not only the content objects should be recognized, but the internal structure should also be preserved. Existing image recognition methods mainly work on images with simple content (e.g., text lines with characters), but are not capable to identify ones with more complex content (e.g., structured code), which often follow a fine-grained grammar. To this end, in this paper, we propose a hierarchical Spotlight Transcribing Network (STN) framework followed by a two-stage "where-to-what'' solution. Specifically, we first decide "where-to-look'' through a novel spotlight mechanism to focus on different areas of the original image following its structure. Then, we decide "what-to-write'' by developing a GRU based network with the spotlight areas for transcribing the content accordingly. Moreover, we propose two implementations on the basis of STN, i.e., STNM and STNR, where the spotlight movement follows the Markov property and Recurrent modeling, respectively. We also design a reinforcement method to refine our STN framework by self-improving the spotlight mechanism. We conduct extensive experiments on many structural image datasets, where the results clearly demonstrate the effectiveness of STN framework. | However, in the original encoder-decoder systems, encoding the whole input into one vector usually makes the encoded information of images clumsy and confusing for the decoder to read from, leading to unsatisfactory transcription @cite_38 . To improve the encoder-decoder models addressing this problem, inspired by human visual system, researchers have tried to propose many attention mechanisms to highlight different parts of the encoder output by assigning weights to encoding vectors in each step of text generation @cite_8 @cite_36 @cite_15 or sequential prediction @cite_3 @cite_22 . For example, @cite_8 proposed a way to jointly generate and align words using attention mechanism. @cite_36 proposed soft and hard attention mechanisms for image captioning. @cite_24 used an attention-based encoder-decoder system for character recognition problems. | {
"cite_N": [
"@cite_38",
"@cite_22",
"@cite_8",
"@cite_36",
"@cite_3",
"@cite_24",
"@cite_15"
],
"mid": [
"2949335953",
"2808310571",
"2964308564",
"2950178297",
"2788574423",
"2294053032",
"2147527908"
],
"abstract": [
"An attentional mechanism has lately been used to improve neural machine translation (NMT) by selectively focusing on parts of the source sentence during translation. However, there has been little work exploring useful architectures for attention-based NMT. This paper examines two simple and effective classes of attentional mechanism: a global approach which always attends to all source words and a local one that only looks at a subset of source words at a time. We demonstrate the effectiveness of both approaches over the WMT translation tasks between English and German in both directions. With local attention, we achieve a significant gain of 5.0 BLEU points over non-attentional systems which already incorporate known techniques such as dropout. Our ensemble model using different attention architectures has established a new state-of-the-art result in the WMT'15 English to German translation task with 25.9 BLEU points, an improvement of 1.0 BLEU points over the existing best system backed by NMT and an n-gram reranker.",
"With a large amount of user activity data accumulated, it is crucial to exploit user sequential behavior for sequential recommendations. Conventionally, user general taste and recent demand are combined to promote recommendation performances. However, existing methods often neglect that user long-term preference keep evolving over time, and building a static representation for user general taste may not adequately reflect the dynamic characters. Moreover, they integrate user-item or itemitem interactions through a linear way which limits the capability of model. To this end, in this paper, we propose a novel two-layer hierarchical attention network, which takes the above properties into account, to recommend the next item user might be interested. Specifically, the first attention layer learns user long-term preferences based on the historical purchased item representation, while the second one outputs final user representation through coupling user long-term and short-term preferences. The experimental study demonstrates the superiority of our method compared with other state-of-the-art ones.",
"Abstract: Neural machine translation is a recently proposed approach to machine translation. Unlike the traditional statistical machine translation, the neural machine translation aims at building a single neural network that can be jointly tuned to maximize the translation performance. The models proposed recently for neural machine translation often belong to a family of encoder-decoders and consists of an encoder that encodes a source sentence into a fixed-length vector from which a decoder generates a translation. In this paper, we conjecture that the use of a fixed-length vector is a bottleneck in improving the performance of this basic encoder-decoder architecture, and propose to extend this by allowing a model to automatically (soft-)search for parts of a source sentence that are relevant to predicting a target word, without having to form these parts as a hard segment explicitly. With this new approach, we achieve a translation performance comparable to the existing state-of-the-art phrase-based system on the task of English-to-French translation. Furthermore, qualitative analysis reveals that the (soft-)alignments found by the model agree well with our intuition.",
"Inspired by recent work in machine translation and object detection, we introduce an attention based model that automatically learns to describe the content of images. We describe how we can train this model in a deterministic manner using standard backpropagation techniques and stochastically by maximizing a variational lower bound. We also show through visualization how the model is able to automatically learn to fix its gaze on salient objects while generating the corresponding words in the output sequence. We validate the use of attention with state-of-the-art performance on three benchmark datasets: Flickr8k, Flickr30k and MS COCO.",
"",
"We present recursive recurrent neural networks with attention modeling (R2AM) for lexicon-free optical character recognition in natural scene images. The primary advantages of the proposed method are: (1) use of recursive convolutional neural networks (CNNs), which allow for parametrically efficient and effective image feature extraction, (2) an implicitly learned character-level language model, embodied in a recurrent neural network which avoids the need to use N-grams, and (3) the use of a soft-attention mechanism, allowing the model to selectively exploit image features in a coordinated way, and allowing for end-to-end training within a standard backpropagation framework. We validate our method with state-of-the-art performance on challenging benchmark datasets: Street View Text, IIIT5k, ICDAR and Synth90k.",
"Applying convolutional neural networks to large images is computationally expensive because the amount of computation scales linearly with the number of image pixels. We present a novel recurrent neural network model that is capable of extracting information from an image or video by adaptively selecting a sequence of regions or locations and only processing the selected regions at high resolution. Like convolutional neural networks, the proposed model has a degree of translation invariance built-in, but the amount of computation it performs can be controlled independently of the input image size. While the model is non-differentiable, it can be trained using reinforcement learning methods to learn task-specific policies. We evaluate our model on several image classification tasks, where it significantly outperforms a convolutional neural network baseline on cluttered images, and on a dynamic visual control problem, where it learns to track a simple object without an explicit training signal for doing so."
]
} |
1905.10954 | 2809263292 | Transcribing content from structural images, e.g., writing notes from music scores, is a challenging task as not only the content objects should be recognized, but the internal structure should also be preserved. Existing image recognition methods mainly work on images with simple content (e.g., text lines with characters), but are not capable to identify ones with more complex content (e.g., structured code), which often follow a fine-grained grammar. To this end, in this paper, we propose a hierarchical Spotlight Transcribing Network (STN) framework followed by a two-stage "where-to-what'' solution. Specifically, we first decide "where-to-look'' through a novel spotlight mechanism to focus on different areas of the original image following its structure. Then, we decide "what-to-write'' by developing a GRU based network with the spotlight areas for transcribing the content accordingly. Moreover, we propose two implementations on the basis of STN, i.e., STNM and STNR, where the spotlight movement follows the Markov property and Recurrent modeling, respectively. We also design a reinforcement method to refine our STN framework by self-improving the spotlight mechanism. We conduct extensive experiments on many structural image datasets, where the results clearly demonstrate the effectiveness of STN framework. | Deep reinforcement learning is a kind of state-of-the-art technique, which has shown superior abilities in many fields, such as gaming and robotics @cite_0 . The main idea of them is to learn and refine model parameters according to task-specific reward signals. For example, @cite_20 used the whole sequence metrics to guide the sequence generation, using REINFORCE method; @cite_34 utilized the actor-critic algorithm for sequence prediction, refining the model to improve sentence BLEU score. | {
"cite_N": [
"@cite_0",
"@cite_34",
"@cite_20"
],
"mid": [
"2745868649",
"2487501366",
"2176263492"
],
"abstract": [
"Deep reinforcement learning (DRL) is poised to revolutionize the field of artificial intelligence (AI) and represents a step toward building autonomous systems with a higherlevel understanding of the visual world. Currently, deep learning is enabling reinforcement learning (RL) to scale to problems that were previously intractable, such as learning to play video games directly from pixels. DRL algorithms are also applied to robotics, allowing control policies for robots to be learned directly from camera inputs in the real world. In this survey, we begin with an introduction to the general field of RL, then progress to the main streams of value-based and policy-based methods. Our survey will cover central algorithms in deep RL, including the deep Q-network (DQN), trust region policy optimization (TRPO), and asynchronous advantage actor critic. In parallel, we highlight the unique advantages of deep neural networks, focusing on visual understanding via RL. To conclude, we describe several current areas of research within the field.",
"We present an approach to training neural networks to generate sequences using actor-critic methods from reinforcement learning (RL). Current log-likelihood training methods are limited by the discrepancy between their training and testing modes, as models must generate tokens conditioned on their previous guesses rather than the ground-truth tokens. We address this problem by introducing a network that is trained to predict the value of an output token, given the policy of an network. This results in a training procedure that is much closer to the test phase, and allows us to directly optimize for a task-specific score such as BLEU. Crucially, since we leverage these techniques in the supervised learning setting rather than the traditional RL setting, we condition the critic network on the ground-truth output. We show that our method leads to improved performance on both a synthetic task, and for German-English machine translation. Our analysis paves the way for such methods to be applied in natural language generation tasks, such as machine translation, caption generation, and dialogue modelling.",
"Many natural language processing applications use language models to generate text. These models are typically trained to predict the next word in a sequence, given the previous words and some context such as an image. However, at test time the model is expected to generate the entire sequence from scratch. This discrepancy makes generation brittle, as errors may accumulate along the way. We address this issue by proposing a novel sequence level training algorithm that directly optimizes the metric used at test time, such as BLEU or ROUGE. On three different tasks, our approach outperforms several strong baselines for greedy generation. The method is also competitive when these baselines employ beam search, while being several times faster."
]
} |
1905.11100 | 2945703303 | Value functions are crucial for model-free Reinforcement Learning (RL) to obtain a policy implicitly or guide the policy updates. Value estimation heavily depends on the stochasticity of environmental dynamics and the quality of reward signals. In this paper, we propose a two-step understanding of value estimation from the perspective of future prediction, through decomposing the value function into a reward-independent future dynamics part and a policy-independent trajectory return part. We then derive a practical deep RL algorithm from the above decomposition, consisting of a convolutional trajectory representation model, a conditional variational dynamics model to predict the expected representation of future trajectory and a convex trajectory return model that maps a trajectory representation to its return. Our algorithm is evaluated in MuJoCo continuous control tasks and shows superior results under both common settings and delayed reward settings. | Thinking about the future has been considered as an integral component of human cognition @cite_8 @cite_29 . In neuroscience, one concept named the prospective brain @cite_28 indicates that a crucial function of the brain is to use stored information to predict possible future events. The idea of future prediction is also studied in model-based RL @cite_26 @cite_0 . Simulated Policy Learning @cite_6 is proposed to learn one-step predictive models from the real environment and then train a policy within the simulated environment. Multi-steps and long-term future are also modeled in @cite_21 @cite_24 with recurrent variational dynamics models, after which actions are chosen through online planning with Model-Predictive Control (MPC). Besides, another related work is @cite_12 , in which a supervised model is trained to predict the residuals of goal-related measurements at a set of temporal offsets in the future. With a manually designed goal vector, actions are chosen to maximize the predicted outcomes. | {
"cite_N": [
"@cite_26",
"@cite_8",
"@cite_28",
"@cite_29",
"@cite_21",
"@cite_6",
"@cite_0",
"@cite_24",
"@cite_12"
],
"mid": [
"1540685400",
"2104128137",
"2095047274",
"2160623437",
"2950004691",
"2920362155",
"1980035368",
"2918394844",
"2553701371"
],
"abstract": [
"",
"Abstract Thinking about the future is an integral component of human cognition – one that has been claimed to distinguish us from other species. Building on the construct of episodic memory, we introduce the concept of ‘episodic future thinking': a projection of the self into the future to pre-experience an event. We argue that episodic future thinking has explanatory value when considering recent work in many areas of psychology: cognitive, social and personality, developmental, clinical and neuropsychology. Episodic future thinking can serve as a unifying concept, connecting aspects of diverse research findings and identifying key questions requiring further reflection and study.",
"A rapidly growing number of recent studies show that imagining the future depends on much of the same neural machinery that is needed for remembering the past. These findings have led to the concept of the prospective brain; an idea that a crucial function of the brain is to use stored information to imagine, simulate and predict possible future events. We suggest that processes such as memory can be productively re-conceptualized in light of this idea.",
"Episodic memory is widely conceived as a fundamentally constructive, rather than reproductive, process that is prone to various kinds of errors and illusions. With a view towards examining the functions served by a constructive episodic memory system, we consider recent neuropsychological and neuroimaging studies indicating that some types of memory distortions reflect the operation of adaptive processes. An important function of a constructive episodic memory is to allow individuals to simulate or imagine future episodes, happenings and scenarios. Since the future is not an exact repetition of the past, simulation of future episodes requires a system that can draw on the past in a manner that flexibly extracts and recombines elements of previous experiences. Consistent with this constructive episodic simulation hypothesis, we consider cognitive, neuropsychological and neuroimaging evidence showing that there is considerable overlap in the psychological and neural processes involved in remembering the past and imagining the future.",
"Planning has been very successful for control tasks with known environment dynamics. To leverage planning in unknown environments, the agent needs to learn the dynamics from interactions with the world. However, learning dynamics models that are accurate enough for planning has been a long-standing challenge, especially in image-based domains. We propose the Deep Planning Network (PlaNet), a purely model-based agent that learns the environment dynamics from images and chooses actions through fast online planning in latent space. To achieve high performance, the dynamics model must accurately predict the rewards ahead for multiple time steps. We approach this using a latent dynamics model with both deterministic and stochastic transition components. Moreover, we propose a multi-step variational inference objective that we name latent overshooting. Using only pixel observations, our agent solves continuous control tasks with contact dynamics, partial observability, and sparse rewards, which exceed the difficulty of tasks that were previously solved by planning with learned models. PlaNet uses substantially fewer episodes and reaches final performance close to and sometimes higher than strong model-free algorithms.",
"Model-free reinforcement learning (RL) can be used to learn effective policies for complex tasks, such as Atari games, even from image observations. However, this typically requires very large amounts of interaction -- substantially more, in fact, than a human would need to learn the same games. How can people learn so quickly? Part of the answer may be that people can learn how the game works and predict which actions will lead to desirable outcomes. In this paper, we explore how video prediction models can similarly enable agents to solve Atari games with fewer interactions than model-free methods. We describe Simulated Policy Learning (SimPLe), a complete model-based deep RL algorithm based on video prediction models and present a comparison of several model architectures, including a novel architecture that yields the best results in our setting. Our experiments evaluate SimPLe on a range of Atari games in low data regime of @math K interactions between the agent and the environment, which corresponds to two hours of real-time play. In most games SimPLe outperforms state-of-the-art model-free algorithms, in some games by over an order of magnitude.",
"Dyna is an AI architecture that integrates learning, planning, and reactive execution. Learning methods are used in Dyna both for compiling planning results and for updating a model of the effects of the agent's actions on the world. Planning is incremental and can use the probabilistic and ofttimes incorrect world models generated by learning processes. Execution is fully reactive in the sense that no planning intervenes between perception and action. Dyna relies on machine learning methods for learning from examples---these are among the basic building blocks making up the architecture---yet is not tied to any particular method. This paper briefly introduces Dyna and discusses its strengths and weaknesses with respect to other architectures.",
"In model-based reinforcement learning, the agent interleaves between model learning and planning. These two components are inextricably intertwined. If the model is not able to provide sensible long-term prediction, the executed planner would exploit model flaws, which can yield catastrophic failures. This paper focuses on building a model that reasons about the long-term future and demonstrates how to use this for efficient planning and exploration. To this end, we build a latent-variable autoregressive model by leveraging recent ideas in variational inference. We argue that forcing latent variables to carry future information through an auxiliary task substantially improves long-term predictions. Moreover, by planning in the latent space, the planner's solution is ensured to be within regions where the model is valid. An exploration strategy can be devised by searching for unlikely trajectories under the model. Our method achieves higher reward faster compared to baselines on a variety of tasks and environments in both the imitation learning and model-based reinforcement learning settings.",
"We present an approach to sensorimotor control in immersive environments. Our approach utilizes a high-dimensional sensory stream and a lower-dimensional measurement stream. The cotemporal structure of these streams provides a rich supervisory signal, which enables training a sensorimotor control model by interacting with the environment. The model is trained using supervised learning techniques, but without extraneous supervision. It learns to act based on raw sensory input from a complex three-dimensional environment. The presented formulation enables learning without a fixed goal at training time, and pursuing dynamically changing goals at test time. We conduct extensive experiments in three-dimensional simulations based on the classical first-person game Doom. The results demonstrate that the presented approach outperforms sophisticated prior formulations, particularly on challenging tasks. The results also show that trained models successfully generalize across environments and goals. A model trained using the presented approach won the Full Deathmatch track of the Visual Doom AI Competition, which was held in previously unseen environments."
]
} |
1905.11100 | 2945703303 | Value functions are crucial for model-free Reinforcement Learning (RL) to obtain a policy implicitly or guide the policy updates. Value estimation heavily depends on the stochasticity of environmental dynamics and the quality of reward signals. In this paper, we propose a two-step understanding of value estimation from the perspective of future prediction, through decomposing the value function into a reward-independent future dynamics part and a policy-independent trajectory return part. We then derive a practical deep RL algorithm from the above decomposition, consisting of a convolutional trajectory representation model, a conditional variational dynamics model to predict the expected representation of future trajectory and a convex trajectory return model that maps a trajectory representation to its return. Our algorithm is evaluated in MuJoCo continuous control tasks and shows superior results under both common settings and delayed reward settings. | Most model-free deep RL algorithms approximate value functions directly with deep neural networks, e.g., Proximal Policy Optimization (PPO) @cite_9 , Advantage Actor-Critic (A2C) @cite_14 and DDPG @cite_23 , without explicitly handling the coupling of environmental dynamics and rewards. One similar approach to our work is the Deep Successor Representation (DSR) @cite_20 , which factors the value function into the dot-product between the expected representation of state occupancy and a vector of immediate reward function. The representation is trained with TD algorithm @cite_17 and the vector is approximated from one-step transitions. In our work, we decompose the value function based on the composite form of trajectory dynamics and returns. In contrast to using TD algorithm, we use a conditional VAE to model the latent distribution and obtain expected trajectory representation. We demonstrate the superiority of our algorithm in the experimental section. | {
"cite_N": [
"@cite_14",
"@cite_9",
"@cite_23",
"@cite_20",
"@cite_17"
],
"mid": [
"2964043796",
"2736601468",
"2963864421",
"2417089653",
"2121863487"
],
"abstract": [
"We propose a conceptually simple and lightweight framework for deep reinforcement learning that uses asynchronous gradient descent for optimization of deep neural network controllers. We present asynchronous variants of four standard reinforcement learning algorithms and show that parallel actor-learners have a stabilizing effect on training allowing all four methods to successfully train neural network controllers. The best performing method, an asynchronous variant of actor-critic, surpasses the current state-of-the-art on the Atari domain while training for half the time on a single multi-core CPU instead of a GPU. Furthermore, we show that asynchronous actor-critic succeeds on a wide variety of continuous motor control problems as well as on a new task of navigating random 3D mazes using a visual input.",
"We propose a new family of policy gradient methods for reinforcement learning, which alternate between sampling data through interaction with the environment, and optimizing a \"surrogate\" objective function using stochastic gradient ascent. Whereas standard policy gradient methods perform one gradient update per data sample, we propose a novel objective function that enables multiple epochs of minibatch updates. The new methods, which we call proximal policy optimization (PPO), have some of the benefits of trust region policy optimization (TRPO), but they are much simpler to implement, more general, and have better sample complexity (empirically). Our experiments test PPO on a collection of benchmark tasks, including simulated robotic locomotion and Atari game playing, and we show that PPO outperforms other online policy gradient methods, and overall strikes a favorable balance between sample complexity, simplicity, and wall-time.",
"Abstract: We adapt the ideas underlying the success of Deep Q-Learning to the continuous action domain. We present an actor-critic, model-free algorithm based on the deterministic policy gradient that can operate over continuous action spaces. Using the same learning algorithm, network architecture and hyper-parameters, our algorithm robustly solves more than 20 simulated physics tasks, including classic problems such as cartpole swing-up, dexterous manipulation, legged locomotion and car driving. Our algorithm is able to find policies whose performance is competitive with those found by a planning algorithm with full access to the dynamics of the domain and its derivatives. We further demonstrate that for many of the tasks the algorithm can learn policies end-to-end: directly from raw pixel inputs.",
"Learning robust value functions given raw observations and rewards is now possible with model-free and model-based deep reinforcement learning algorithms. There is a third alternative, called Successor Representations (SR), which decomposes the value function into two components -- a reward predictor and a successor map. The successor map represents the expected future state occupancy from any given state and the reward predictor maps states to scalar rewards. The value function of a state can be computed as the inner product between the successor map and the reward weights. In this paper, we present DSR, which generalizes SR within an end-to-end deep reinforcement learning framework. DSR has several appealing properties including: increased sensitivity to distal reward changes due to factorization of reward and world dynamics, and the ability to extract bottleneck states (subgoals) given successor maps trained under a random policy. We show the efficacy of our approach on two diverse environments given raw pixel observations -- simple grid-world domains (MazeBase) and the Doom game engine.",
"Reinforcement learning, one of the most active research areas in artificial intelligence, is a computational approach to learning whereby an agent tries to maximize the total amount of reward it receives when interacting with a complex, uncertain environment. In Reinforcement Learning, Richard Sutton and Andrew Barto provide a clear and simple account of the key ideas and algorithms of reinforcement learning. Their discussion ranges from the history of the field's intellectual foundations to the most recent developments and applications. The only necessary mathematical background is familiarity with elementary concepts of probability. The book is divided into three parts. Part I defines the reinforcement learning problem in terms of Markov decision processes. Part II provides basic solution methods: dynamic programming, Monte Carlo methods, and temporal-difference learning. Part III presents a unified view of the solution methods and incorporates artificial neural networks, eligibility traces, and planning; the two final chapters present case studies and consider the future of reinforcement learning."
]
} |
1905.10963 | 2945278785 | We propose interpretable deep Gaussian Processes (GPs) that combine the expressiveness of deep Neural Networks (NNs) with quantified uncertainty of deep GPs. Our approach is based on approximating deep GP as a GP, which allows explicit, analytic forms for compositions of a wide variety of kernels. Consequently, our approach admits interpretation as both NNs with specified activation functions and as a variational approximation to deep GPs. We provide general recipes for deriving the effective kernels for deep GPs of two, three, or infinitely many layers, composed of homogeneous or heterogeneous kernels. Results illustrate the expressiveness of our effective kernels through samples from the prior and inference on simulated data and demonstrate advantages of interpretability by analysis of analytic forms, drawing relations and equivalences across kernels, and a priori identification of non-pathological regimes of hyperparameter space. | Williams @cite_18 was first to derive analytic kernels of one-layer neural networks, using sigmoidal and Gaussian activation functions. Later, Cho and Saul @cite_14 extended analytic results to polynomial activation functions and further made extension to deep models. Daniely al @cite_20 extended these results for homogeneous deep networks with 2nd hermite, step, and exponential activation functions and Poole al @cite_5 to @math activation functions. Our approach differs by focusing on deep GPs for Bayesian inference, and allowing heterogeneous kernels. | {
"cite_N": [
"@cite_5",
"@cite_18",
"@cite_14",
"@cite_20"
],
"mid": [
"2423689290",
"",
"2167608136",
"2962939986"
],
"abstract": [
"We combine Riemannian geometry with the mean field theory of high dimensional chaos to study the nature of signal propagation in generic, deep neural networks with random weights. Our results reveal an order-to-chaos expressivity phase transition, with networks in the chaotic phase computing nonlinear functions whose global curvature grows exponentially with depth but not width. We prove this generic class of deep random functions cannot be efficiently computed by any shallow network, going beyond prior work restricted to the analysis of single functions. Moreover, we formalize and quantitatively demonstrate the long conjectured idea that deep networks can disentangle highly curved manifolds in input space into flat manifolds in hidden space. Our theoretical analysis of the expressive power of deep networks broadly applies to arbitrary nonlinearities, and provides a quantitative underpinning for previously abstract notions about the geometry of deep functions.",
"",
"We introduce a new family of positive-definite kernel functions that mimic the computation in large, multilayer neural nets. These kernel functions can be used in shallow architectures, such as support vector machines (SVMs), or in deep kernel-based architectures that we call multilayer kernel machines (MKMs). We evaluate SVMs and MKMs with these kernel functions on problems designed to illustrate the advantages of deep architectures. On several problems, we obtain better results than previous, leading benchmarks from both SVMs with Gaussian kernels as well as deep belief nets.",
"We develop a general duality between neural networks and compositional kernel Hilbert spaces. We introduce the notion of a computation skeleton, an acyclic graph that succinctly describes both a family of neural networks and a kernel space. Random neural networks are generated from a skeleton through node replication followed by sampling from a normal distribution to assign weights. The kernel space consists of functions that arise by compositions, averaging, and non-linear transformations governed by the skeleton's graph topology and activation functions. We prove that random networks induce representations which approximate the kernel space. In particular, it follows that random weight initialization often yields a favorable starting point for optimization despite the worst-case intractability of training neural networks."
]
} |
1905.10963 | 2945278785 | We propose interpretable deep Gaussian Processes (GPs) that combine the expressiveness of deep Neural Networks (NNs) with quantified uncertainty of deep GPs. Our approach is based on approximating deep GP as a GP, which allows explicit, analytic forms for compositions of a wide variety of kernels. Consequently, our approach admits interpretation as both NNs with specified activation functions and as a variational approximation to deep GPs. We provide general recipes for deriving the effective kernels for deep GPs of two, three, or infinitely many layers, composed of homogeneous or heterogeneous kernels. Results illustrate the expressiveness of our effective kernels through samples from the prior and inference on simulated data and demonstrate advantages of interpretability by analysis of analytic forms, drawing relations and equivalences across kernels, and a priori identification of non-pathological regimes of hyperparameter space. | Duvenaud al @cite_8 studied pathologies in deep GPs, primarily focusing on squared exponential kernels and sampling functions from the prior. We study pathologies using our analytic approach identifying regimes of non-pathology in hyperparameter space, and we show results for prediction for a variety of kernels while controlling complexity through marginalization, which interestingly yields a different analytic form for deep squared exponential kernels. | {
"cite_N": [
"@cite_8"
],
"mid": [
"1589901016"
],
"abstract": [
"Choosing appropriate architectures and regularization strategies of deep networks is crucial to good predictive performance. To shed light on this problem, we analyze the analogous problem of constructing useful priors on compositions of functions. Specifically, we study the deep Gaussian process, a type of infinitely-wide, deep neural network. We show that in standard architectures, the representational capacity of the network tends to capture fewer degrees of freedom as the number of layers increases, retaining only a single degree of freedom in the limit. We propose an alternate network architecture which does not suffer from this pathology. We also examine deep covariance functions, obtained by composing infinitely many feature transforms. Lastly, we characterize the class of models obtained by performing dropout on Gaussian processes."
]
} |
1905.10999 | 2945464521 | Unbiased and objective architectural design decisions are crucial for the success of a software development project. Stakeholder inputs play an important role in arriving at such design decisions. However, the stakeholders may act in a biased manner in order to ensure that their requirements and concerns are addressed in a specific way in the software. Most of the existing methods of architectural decision-making do not adequately account for such biased behaviour of stakeholders. We view the software architecture design as a collective decision making (CDM) problem in social choice theory, and introduce the central ideas of mechanism design to our field. Our contributions are twofold: i) Using the impossibility results from social choice theory, we show that a rational stakeholder can game, to her advantage, almost every known method of software architecture decision making. ii) We also show that if the architect is willing to bear the extra cost of giving suitable incentives to stakeholders, then architectural decision making can be protected from strategic (biased) manipulations. For achieving this objective, we introduce the Vickrey-Clarke-Groves (VCG) mechanism from social choice theory to the software architecture domain. We illustrate our contributions by comparing and using the examples drawn from the well-known CBAM method of architectural decision making. | To the best of our knowledge, there exists no decision-making technique in software architecture which uses mechanism design or social choice theory. Hence, we consider those methodologies related to our work which takes stakeholders' concerns as input along with requirements and other environmental constraints. In methods like CBAM @cite_5 , @cite_12 and @cite_3 the authors made the customer decide which stakeholder to be included in the decision-making process, and then these stakeholders give some numeric number or score to the quality attributes, where these number should sum to 100. They made the stakeholder describe why or how a particular quality is essential. To choose between the various architectural strategy they calculate a desirability factor which decreases with cost. CBAM paper lists three important criteria to arrive at an architectural decision: costs, benefits, and uncertainty or risks. All data gathered and used during an architecture decision process is at best only an estimate, and therefore is best represented by associated probability distributions, rather than exact, definite values. The CBAM framework measures the relevance of a candidate architecture by the ratio of benefits to costs. | {
"cite_N": [
"@cite_5",
"@cite_3",
"@cite_12"
],
"mid": [
"2117324915",
"1502278114",
"1572181078"
],
"abstract": [
"The benefits of a software system are assessable only relative to the business goals the system has been developed to serve. In turn, these benefits result from interactions between the system's functionality and its quality attributes (such as performance, reliabilty and security). Its quality attributes are, in most cases, dictated by its architectural design decisions. Therefore, we argue in this paper that the software architecture is the crucial artifact to study in making design tradeoffs and in performing cost-benefit analyses. A substantial part of such an analysis is in determining the level of uncertainty with which we estimate both costs and benefits. In this paper we offer an architecture-centric approach to the economic modeling of software design decision making called CBAM (Cost Benefit Analysis Method), in which costs and benefits are traded off with system quality attributes. We present the CBAM, the early results from applying this method in a large-scale case study, and discuss the application of more sophisticated economic models to software decision making.",
"Abstract : The resources available to build any system are finite. The decisions involved in building any nontrivial system are complex and typically involve many stakeholders, many requirements, and many technical decisions. The stakeholders have an interest in ensuring that good design decisions are made decisions that meet their technical objectives and their tolerance for risk. These decisions should, as much as possible, maximize the benefit that the system provides and minimize its cost. The Cost Benefit Analysis Method (CBAM) was created to provide some structure to this decision-making process. The CBAM analyzes architectural decisions from the perspectives of cost, benefit, schedule, and risk. While the CBAM does not make decisions for the stakeholders, it does serve as a tool to inform managers and to structure the inquiry so that rational decisions can be made. This report describes the steps of the CBAM and its application to a real-world system.",
"Abstract : The software architecture forms an essential part of a complex software-intensive system. Architecture design decision-making involves addressing tradeoffs due to the presence of economic constraints. The problem is to develop a process that helps a designer choose amongst architectural options, during both initial design and its subsequent periods of upgrade, while being constrained to finite resources. To address this need for better decision-making, we have developed a method for performing economic modeling of software systems, centered on an analysis of their architecture. We call this method the Cost Benefit Analysis Method (CBAM). The CBAM incorporates the costs and benefits of architectural design decisions and provides an effective means of making such decisions. The CBAM provides a structured integrated assessment of the technical and economic issues and architectural decisions. The CBAM utilizes techniques in decision analysis, optimization, and statistics to help software architects characterize their uncertainty and choose a subset of changes that should be implemented from a larger set of alternatives. We also report on the application of this method to a real world case study."
]
} |
1905.10999 | 2945464521 | Unbiased and objective architectural design decisions are crucial for the success of a software development project. Stakeholder inputs play an important role in arriving at such design decisions. However, the stakeholders may act in a biased manner in order to ensure that their requirements and concerns are addressed in a specific way in the software. Most of the existing methods of architectural decision-making do not adequately account for such biased behaviour of stakeholders. We view the software architecture design as a collective decision making (CDM) problem in social choice theory, and introduce the central ideas of mechanism design to our field. Our contributions are twofold: i) Using the impossibility results from social choice theory, we show that a rational stakeholder can game, to her advantage, almost every known method of software architecture decision making. ii) We also show that if the architect is willing to bear the extra cost of giving suitable incentives to stakeholders, then architectural decision making can be protected from strategic (biased) manipulations. For achieving this objective, we introduce the Vickrey-Clarke-Groves (VCG) mechanism from social choice theory to the software architecture domain. We illustrate our contributions by comparing and using the examples drawn from the well-known CBAM method of architectural decision making. | In @cite_16 the architects in collaborations with project stakeholders define the cost and benefit functions. Each alternative has a utility score defined as the weighted sum of the stakeholders' preferences for each goal. Hence they measure the relevance of a candidate architecture by the net benefit,'' i.e., quantity obtained by subtracting the costs from the benefits. CEDA @cite_6 considers stakeholder's concerns to develop shared understanding aligned towards an organization's goals. Tools like RADAR @cite_1 use CBAM or ATAM as the starting point of developing an architecture and then handle particular aspects of architecture design decision like uncertainty using a Pareto optimal approach. Techniques like REARM @cite_8 accept quantifiable input from stakeholders, but do not provide the details of how is to be done. Further, techniques like @cite_14 integrate updated versions of CBAM-2 to handle structured analysis of architecture in agile development. @cite_11 suggests a workshop based approach to handle concerns of stakeholders and specifically mentions as automating the handling of concerns and conflict as the need of the hour. | {
"cite_N": [
"@cite_14",
"@cite_8",
"@cite_1",
"@cite_6",
"@cite_16",
"@cite_11"
],
"mid": [
"2652144948",
"",
"2617610500",
"1512700395",
"1999715773",
"2491100436"
],
"abstract": [
"Software architecture has many definitions. One widely accepted definition of software architecture is that it is a composition of a set of architectural design decisions. Hence, designing a software architecture is a decision-making process. Agile methods drastically changed the way of designing a software architecture. In projects using agile methods (e.g. Scrum), making architectural design decisions is not the responsibility of a single person, but rather the whole development team. Despite the popularity of such methods in the industry, little research exists on how to make these decisions from the perspective of a group effectively. Current techniques usually focus on the identification of quality attributes and design alternatives, not addressing the whole decision-making process. The quality of decisions directly reflects the quality of the software architecture. Therefore poor decisions lead to bad software architectures. In this paper, we discuss current research on group decision-making in software architecture and the proposal of a combination of concepts from two architecture definition methods into a single approach that can be used in agile projects and addresses the most critical concerns of group decision-making. This proposal is part of a master's research project.",
"",
"Uncertainty and conflicting stakeholders' objectives make many requirements and architecture decisions particularly hard. Quantitative probabilistic models allow software architects to analyse such decisions using stochastic simulation and multi-objective optimisation, but the difficulty of elaborating the models is an obstacle to the wider adoption of such techniques. To reduce this obstacle, this paper presents a novel modelling language and analysis tool, called RADAR, intended to facilitate requirements and architecture decision analysis. The language has relations to quantitative AND OR goal models used in requirements engineering and to feature models used in software product lines. However, it simplifies such models to a minimum set of language constructs essential for decision analysis. The paper presents RADAR's modelling language, automated support for decision analysis, and evaluates its application to four real-world examples.",
"",
"Uncertainty complicates early requirements and architecture decisions and may expose a software project to significant risk. Yet software architects lack support for evaluating uncertainty, its impact on risk, and the value of reducing uncertainty before making critical decisions. We propose to apply decision analysis and multi-objective optimisation techniques to provide such support. We present a systematic method allowing software architects to describe uncertainty about the impact of alternatives on stakeholders' goals; to calculate the consequences of uncertainty through Monte-Carlo simulation; to shortlist candidate architectures based on expected costs, benefits and risks; and to assess the value of obtaining additional information before deciding. We demonstrate our method on the design of a system for coordinating emergency response teams. Our approach highlights the need for requirements engineering and software cost estimation methods to disclose uncertainty instead of hiding it.",
"During the development of a software system, architects deal with a large number of stakeholders, each with differing concerns. This inevitably leads to inconsistency: goals, concerns, design decisions, and models are interrelated and overlapping. Existing approaches to support inconsistency management are limited in their applicability and usefulness in day to day practice due to the presence of incomplete, informal and heterogeneous models in software architecture. This paper presents a novel process in the form of a lightweight generic method, the Concern-Driven Inconsistency Management (CDIM) method, that is designed to address limitations of different related approaches. It aims to aid architects with management of intangible inconsistency in software architecture."
]
} |
1905.11169 | 2945476782 | We aim to perform unsupervised discovery of objects and their states such as location and velocity, as well as physical system parameters such as mass and gravity from video -- given only the differential equations governing the scene dynamics. Existing physical scene understanding methods require either object state supervision, or do not integrate with differentiable physics to learn interpretable system parameters and states. We address this problem through a @math approach that brings together vision-as-inverse-graphics and differentiable physics engines. This framework allows us to perform long term extrapolative video prediction, as well as vision-based model-predictive control. Our approach significantly outperforms related unsupervised methods in long-term future frame prediction of systems with interacting objects (such as ball-spring or 3-body gravitational systems). We further show the value of this tight vision-physics integration by demonstrating data-efficient learning of vision-actuated model-based control for a pendulum system. The controller's interpretability also provides unique capabilities in goal-driven control and physical reasoning for zero-data adaptation. | Within the differentiable physics literature , observed that a multi-layer perceptron (MLP) encoder-decoder with a physics engine, was not able to learn without supervising the physics engine's output with position velocity labels (c.f. Fig. 4 in @cite_4 ). While in their case 2 from 100 | {
"cite_N": [
"@cite_4"
],
"mid": [
"2891122218"
],
"abstract": [
"We present a differentiable physics engine that can be integrated as a module in deep neural networks for end-to-end learning. As a result, structured physics knowledge can be embedded into larger systems, allowing them, for example, to match observations by performing precise simulations, while achieves high sample efficiency. Specifically, in this paper we demonstrate how to perform backpropagation analytically through a physical simulator defined via a linear complementarity problem. Unlike traditional finite difference methods, such gradients can be computed analytically, which allows for greater flexibility of the engine. Through experiments in diverse domains, we highlight the system's ability to learn physical parameters from data, efficiently match and simulate observed visual behavior, and readily enable control via gradient-based planning methods."
]
} |
1905.11062 | 2945669993 | Autoencoders and their variations provide unsupervised models for learning low-dimensional representations for downstream tasks. Without proper regularization, autoencoder models are susceptible to the overfitting problem and the so-called posterior collapse phenomenon. In this paper, we introduce a quantization-based regularizer in the bottleneck stage of autoencoder models to learn meaningful latent representations. We combine both perspectives of Vector Quantized-Variational AutoEncoders (VQ-VAE) and classical denoising regularization schemes of neural networks. We interpret quantizers as regularizers that constrain latent representations while fostering a similarity mapping at the encoder. Before quantization, we impose noise on the latent variables and use a Bayesian estimator to optimize the quantizer-based representation. The introduced bottleneck Bayesian estimator outputs the posterior mean of the centroids to the decoder, and thus, is performing soft quantization of the latent variables. We show that our proposed regularization method results in improved latent representations for both supervised learning and clustering downstream tasks when compared to autoencoders using other bottleneck structures. | For extended work on VQ-VAE, @cite_23 uses the Expectation Maximization algorithm in the bottleneck stage to train the VQ-VAE and to achieve improved image generation results. However, the stability of the proposed algorithm may require to collect a large number of samples in the latent space. @cite_9 gives a probabilistic interpretation of the VQ-VAE and recovers its objective function using the variational inference principle combined with implicit assumptions made by the classical VQ-VAE model. | {
"cite_N": [
"@cite_9",
"@cite_23"
],
"mid": [
"2884607399",
"2804145368"
],
"abstract": [
"Generating versatile and appropriate synthetic speech requires control over the output expression separate from the spoken text. Important non-textual speech variation is seldom annotated, in which case output control must be learned in an unsupervised fashion. In this paper, we perform an in-depth study of methods for unsupervised learning of control in statistical speech synthesis. For example, we show that popular unsupervised training heuristics can be interpreted as variational inference in certain autoencoder models. We additionally connect these models to VQ-VAEs, another, recently-proposed class of deep variational autoencoders, which we show can be derived from a very similar mathematical argument. The implications of these new probabilistic interpretations are discussed. We illustrate the utility of the various approaches with an application to acoustic modelling for emotional speech synthesis, where the unsupervised methods for learning expression control (without access to emotional labels) are found to give results that in many aspects match or surpass the previous best supervised approach.",
"Deep neural networks with discrete latent variables offer the promise of better symbolic reasoning, and learning abstractions that are more useful to new tasks. There has been a surge in interest in discrete latent variable models, however, despite several recent improvements, the training of discrete latent variable models has remained challenging and their performance has mostly failed to match their continuous counterparts. Recent work on vector quantized autoencoders (VQ-VAE) has made substantial progress in this direction, with its perplexity almost matching that of a VAE on datasets such as CIFAR-10. In this work, we investigate an alternate training technique for VQ-VAE, inspired by its connection to the Expectation Maximization (EM) algorithm. Training the discrete bottleneck with EM helps us achieve better image generation results on CIFAR-10, and together with knowledge distillation, allows us to develop a non-autoregressive machine translation model whose accuracy almost matches a strong greedy autoregressive baseline Transformer, while being 3.3 times faster at inference."
]
} |
1905.11062 | 2945669993 | Autoencoders and their variations provide unsupervised models for learning low-dimensional representations for downstream tasks. Without proper regularization, autoencoder models are susceptible to the overfitting problem and the so-called posterior collapse phenomenon. In this paper, we introduce a quantization-based regularizer in the bottleneck stage of autoencoder models to learn meaningful latent representations. We combine both perspectives of Vector Quantized-Variational AutoEncoders (VQ-VAE) and classical denoising regularization schemes of neural networks. We interpret quantizers as regularizers that constrain latent representations while fostering a similarity mapping at the encoder. Before quantization, we impose noise on the latent variables and use a Bayesian estimator to optimize the quantizer-based representation. The introduced bottleneck Bayesian estimator outputs the posterior mean of the centroids to the decoder, and thus, is performing soft quantization of the latent variables. We show that our proposed regularization method results in improved latent representations for both supervised learning and clustering downstream tasks when compared to autoencoders using other bottleneck structures. | Several works have studied the end-to-end discrete representation learning model with different incorporated structures in the bottleneck stages. @cite_28 and @cite_21 introduce scalar quantization in the latent space and optimize jointly the entire model for rate-distortion performance over a database of training images. @cite_11 proposes a compression model by performing vector quantization on the network activations. The model uses a continuous relaxation of vector quantization which is annealed over time to obtain a hard clustering. In @cite_11 , the softmax function is used to give a soft assignment to the codewords where a single smoothing factor is used as an annealing factor. In our model, we learn different smoothing factors for each component. | {
"cite_N": [
"@cite_28",
"@cite_21",
"@cite_11"
],
"mid": [
"2593493485",
"2962676454",
"2964164354"
],
"abstract": [
"We propose a new approach to the problem of optimizing autoencoders for lossy image compression. New media formats, changing hardware technology, as well as diverse requirements and content types create a need for compression algorithms which are more flexible than existing codecs. Autoencoders have the potential to address this need, but are difficult to optimize directly due to the inherent non-differentiabilty of the compression loss. We here show that minimal changes to the loss are sufficient to train deep autoencoders competitive with JPEG 2000 and outperforming recently proposed approaches based on RNNs. Our network is furthermore computationally efficient thanks to a sub-pixel architecture, which makes it suitable for high-resolution images. This is in contrast to previous work on autoencoders for compression using coarser approximations, shallower architectures, computationally expensive methods, or focusing on small images.",
"We describe an image compression method, consisting of a nonlinear analysis transformation, a uniform quantizer, and a nonlinear synthesis transformation. The transforms are constructed in three successive stages of convolutional linear filters and nonlinear activation functions. Unlike most convolutional neural networks, the joint nonlinearity is chosen to implement a form of local gain control, inspired by those used to model biological neurons. Using a variant of stochastic gradient descent, we jointly optimize the entire model for rate-distortion performance over a database of training images, introducing a continuous proxy for the discontinuous loss function arising from the quantizer. Under certain conditions, the relaxed loss function may be interpreted as the log likelihood of a generative model, as implemented by a variational autoencoder. Unlike these models, however, the compression model must operate at any given point along the rate-distortion curve, as specified by a trade-off parameter. Across an independent set of test images, we find that the optimized method generally exhibits better rate-distortion performance than the standard JPEG and JPEG 2000 compression methods. More importantly, we observe a dramatic improvement in visual quality for all images at all bit rates, which is supported by objective quality estimates using MS-SSIM.",
"We present a new approach to learn compressible representations in deep architectures with an end-to-end training strategy. Our method is based on a soft (continuous) relaxation of quantization and entropy, which we anneal to their discrete counterparts throughout training. We showcase this method for two challenging applications: Image compression and neural network compression. While these tasks have typically been approached with different methods, our soft-to-hard quantization approach gives results competitive with the state-of-the-art for both."
]
} |
1905.11062 | 2945669993 | Autoencoders and their variations provide unsupervised models for learning low-dimensional representations for downstream tasks. Without proper regularization, autoencoder models are susceptible to the overfitting problem and the so-called posterior collapse phenomenon. In this paper, we introduce a quantization-based regularizer in the bottleneck stage of autoencoder models to learn meaningful latent representations. We combine both perspectives of Vector Quantized-Variational AutoEncoders (VQ-VAE) and classical denoising regularization schemes of neural networks. We interpret quantizers as regularizers that constrain latent representations while fostering a similarity mapping at the encoder. Before quantization, we impose noise on the latent variables and use a Bayesian estimator to optimize the quantizer-based representation. The introduced bottleneck Bayesian estimator outputs the posterior mean of the centroids to the decoder, and thus, is performing soft quantization of the latent variables. We show that our proposed regularization method results in improved latent representations for both supervised learning and clustering downstream tasks when compared to autoencoders using other bottleneck structures. | Various techniques for regularizing the autoencoders have been proposed recently. @cite_5 proposes an adversarial regularizer which encourages interpolation in the outputs and also improves the learned representation. @cite_29 interprets the VAEs as a amortized inference algorithm and proposed a procedure to constrain the expressiveness of the encoder. In addition, there is a increasing popularity of using information-theoretic principles to improve autoencoders. @cite_10 @cite_12 use the information bottleneck principle @cite_15 to recover the objective of @math -VAE and show that the KL divergence term in ELBO is an upper bound on the information rate between input and prior. @cite_2 is also inspired by the information bottleneck principle and introduces the information dropout method to penalize the transfer of information from data to the latents. @cite_24 proposes to use encoder-decoder structures to simulate the binary symmetric channel (BSC) to solve the joint source-channel coding problem while at the same time learn robust representations. | {
"cite_N": [
"@cite_29",
"@cite_24",
"@cite_2",
"@cite_5",
"@cite_15",
"@cite_10",
"@cite_12"
],
"mid": [
"2962964508",
"2901615357",
"2683470288",
"2952034151",
"2964184826",
"",
""
],
"abstract": [
"The variational autoencoder (VAE) is a popular model for density estimation and representation learning. Canonically, the variational principle suggests to prefer an expressive inference model so that the variational approximation is accurate. However, it is often overlooked that an overly-expressive inference model can be detrimental to the test set performance of both the amortized posterior approximator and, more importantly, the generative density estimator. In this paper, we leverage the fact that VAEs rely on amortized inference and propose techniques for amortized inference regularization (AIR) that control the smoothness of the inference model. We demonstrate that, by applying AIR, it is possible to improve VAE generalization on both inference and generative performance. Our paper challenges the belief that amortized inference is simply a mechanism for approximating maximum likelihood training and illustrates that regularization of the amortization family provides a new direction for understanding and improving generalization in VAEs.",
"For reliable transmission across a noisy communication channel, classical results from information theory show that it is asymptotically optimal to separate out the source and channel coding processes. However, this decomposition can fall short in the finite bit-length regime, as it requires non-trivial tuning of hand-crafted codes and assumes infinite computational power for decoding. In this work, we propose Neural Error Correcting and Source Trimming ( ) codes to jointly learn the encoding and decoding processes in an end-to-end fashion. By adding noise into the latent codes to simulate the channel during training, we learn to both compress and error-correct given a fixed bit-length and computational budget. We obtain codes that are not only competitive against several capacity-approaching channel codes, but also learn useful robust representations of the data for downstream tasks such as classification. Finally, we learn an extremely fast neural decoder, yielding almost an order of magnitude in speedup compared to standard decoding methods based on iterative belief propagation.",
"The cross-entropy loss commonly used in deep learning is closely related to the defining properties of optimal representations, but does not enforce some of the key properties. We show that this can be solved by adding a regularization term, which is in turn related to injecting multiplicative noise in the activations of a Deep Neural Network, a special case of which is the common practice of dropout. We show that our regularized loss function can be efficiently minimized using Information Dropout, a generalization of dropout rooted in information theoretic principles that automatically adapts to the data and can better exploit architectures of limited capacity. When the task is the reconstruction of the input, we show that our loss function yields a Variational Autoencoder as a special case, thus providing a link between representation learning, information theory and variational inference. Finally, we prove that we can promote the creation of optimal disentangled representations simply by enforcing a factorized prior, a fact that has been observed empirically in recent work. Our experiments validate the theoretical intuitions behind our method, and we find that Information Dropout achieves a comparable or better generalization performance than binary dropout, especially on smaller models, since it can automatically adapt the noise to the structure of the network, as well as to the test sample.",
"Autoencoders provide a powerful framework for learning compressed representations by encoding all of the information needed to reconstruct a data point in a latent code. In some cases, autoencoders can \"interpolate\": By decoding the convex combination of the latent codes for two datapoints, the autoencoder can produce an output which semantically mixes characteristics from the datapoints. In this paper, we propose a regularization procedure which encourages interpolated outputs to appear more realistic by fooling a critic network which has been trained to recover the mixing coefficient from interpolated data. We then develop a simple benchmark task where we can quantitatively measure the extent to which various autoencoders can interpolate and show that our regularizer dramatically improves interpolation in this setting. We also demonstrate empirically that our regularizer produces latent codes which are more effective on downstream tasks, suggesting a possible link between interpolation abilities and learning useful representations.",
"Deep Neural Networks (DNNs) are analyzed via the theoretical framework of the information bottleneck (IB) principle. We first show that any DNN can be quantified by the mutual information between the layers and the input and output variables. Using this representation we can calculate the optimal information theoretic limits of the DNN and obtain finite sample generalization bounds. The advantage of getting closer to the theoretical limit is quantifiable both by the generalization bound and by the network's simplicity. We argue that both the optimal architecture, number of layers and features connections at each layer, are related to the bifurcation points of the information bottleneck tradeoff, namely, relevant compression of the input layer with respect to the output layer. The hierarchical representations at the layered network naturally correspond to the structural phase transitions along the information curve. We believe that this new insight can lead to new optimality bounds and deep learning algorithms.",
"",
""
]
} |
1905.11027 | 2946332190 | Why do deep neural networks generalize with a very high dimensional parameter space? We took an information theoretic approach. We find that the dimensionality of the parameter space can be studied by singular semi-Riemannian geometry and is upper-bounded by the sample size. We adapt Fisher information to this singular neuromanifold. We use random matrix theory to derive a minimum description length of a deep learning model, where the spectrum of the Fisher information matrix plays a key role to improve generalisation. | The dynamics of supervised learning of a NN describes a trajectory on the parameter space of the NN geometrically modeled as a manifold when endowed with the FIM (e.g., ordinary natural gradient descent learning the parameters of a MLP). Singular regions of the neuromanifold @cite_23 correspond to non-identifiable parameters with rank-deficient FIM, and the learning trajectory typically exhibit chaotic patterns @cite_26 with the singularities which translate into slowdown plateau phenomena when plotting the loss function value against time. By building an elementary singular NN, @cite_26 (and references therein) show that GD learning dynamics yields a Milnor-type attractor with both attractor repulser subregions where the learning trajectory is attracted in the attractor region, then stay a long time there before escaping through the repulser region. The natural gradient is shown to be free of critical slowdowns. Furthermore, although large DNNs have potentially many singular regions, it is shown that the interaction of elementary units cancel out the Milnor-type attractors. It was shown @cite_48 that skip connections are helpful to reduce the effect of singularities. However, a full understanding of the learning dynamics @cite_20 for generic NN architectures with multiple output values or recurrent NNs is yet to be investigated. | {
"cite_N": [
"@cite_48",
"@cite_26",
"@cite_20",
"@cite_23"
],
"mid": [
"2623790698",
"2766357725",
"2915006549",
"2125911849"
],
"abstract": [
"Skip connections made the training of very deep networks possible and have become an indispensable component in a variety of neural architectures. A completely satisfactory explanation for their success remains elusive. Here, we present a novel explanation for the benefits of skip connections in training very deep networks. The difficulty of training deep networks is partly due to the singularities caused by the non-identifiability of the model. Several such singularities have been identified in previous works: (i) overlap singularities caused by the permutation symmetry of nodes in a given layer, (ii) elimination singularities corresponding to the elimination, i.e. consistent deactivation, of nodes, (iii) singularities generated by the linear dependence of the nodes. These singularities cause degenerate manifolds in the loss landscape that slow down learning. We argue that skip connections eliminate these singularities by breaking the permutation symmetry of nodes, by reducing the possibility of node elimination and by making the nodes less linearly dependent. Moreover, for typical initializations, skip connections move the network away from the \"ghosts\" of these singularities and sculpt the landscape around them to alleviate the learning slow-down. These hypotheses are supported by evidence from simplified models, as well as from experiments with deep networks trained on real-world datasets.",
"The dynamics of supervised learning play a main role in deep learning, which takes place in the parameter space of a multilayer perceptron MLP. We review the history of supervised stochastic gradient learning, focusing on its singular structure and natural gradient. The parameter space includes singular regions in which parameters are not identifiable. One of our results is a full exploration of the dynamical behaviors of stochastic gradient learning in an elementary singular network. The bad news is its pathological nature, in which part of the singular region becomes an attractor and another part a repulser at the same time, forming a Milnor attractor. A learning trajectory is attracted by the attractor region, staying in it for a long time, before it escapes the singular region through the repulser region. This is typical of plateau phenomena in learning. We demonstrate the strange topology of a singular region by introducing blow-down coordinates, which are useful for analyzing the natural gradient dynamics. We confirm that the natural gradient dynamics are free of critical slowdown. The second main result is the good news: the interactions of elementary singular networks eliminate the attractor part and the Milnor-type attractors disappear. This explains why large-scale networks do not suffer from serious critical slowdowns due to singularities. We finally show that the unit-wise natural gradient is effective for learning in spite of its low computational cost.",
"",
"We explicitly analyze the trajectories of learning near singularities in hierarchical networks, such as multilayer perceptrons and radial basis function networks, which include permutation symmetry of hidden nodes, and show their general properties. Such symmetry induces singularities in their parameter space, where the Fisher information matrix degenerates and odd learning behaviors, especially the existence of plateaus in gradient descent learning, arise due to the geometric structure of singularity. We plot dynamic vector fields to demonstrate the universal trajectories of learning near singularities. The singularity induces two types of plateaus, the on-singularity plateau and the near-singularity plateau, depending on the stability of the singularity and the initial parameters of learning. The results presented in this letter are universally applicable to a wide class of hierarchical models. Detailed stability analysis of the dynamics of learning in radial basis function networks and multilayer perceptrons will be presented in separate work."
]
} |
1905.10930 | 2946260471 | We propose a method for non-projective dependency parsing by incrementally predicting a set of edges. Since the edges do not have a pre-specified order, we propose a set-based learning method. Our method blends graph, transition, and easy-first parsing, including a prior state of the parser as a special case. The proposed transition-based method successfully parses near the state of the art on both projective and non-projective languages, without assuming a certain parsing order. | Transition-based dependency parsing has a rich history, with methods generally varying by the choice of transition system and feature representation. Traditional stack-based arc-standard and arc-eager @cite_5 @cite_20 transition systems only parse projectively, requiring additional operations for pseudo-non-projectivity @cite_21 or projectivity @cite_24 , while list-based non-projective systems have been developed @cite_15 . Recent variations assume a generation order such as top-down @cite_23 or left-to-right @cite_9 . Other recent models focus on unsupervised settings @cite_2 . Our focus here is a non-projective transition system and learning method which does not assume a particular generation order. | {
"cite_N": [
"@cite_9",
"@cite_21",
"@cite_24",
"@cite_23",
"@cite_2",
"@cite_5",
"@cite_15",
"@cite_20"
],
"mid": [
"2923547786",
"2106011930",
"2131395877",
"",
"2952203969",
"181643614",
"2105847779",
""
],
"abstract": [
"We propose a novel transition-based algorithm that straightforwardly parses sentences from left to right by building @math attachments, with @math being the length of the input sentence. Similarly to the recent stack-pointer parser by (2018), we use the pointer network framework that, given a word, can directly point to a position from the sentence. However, our left-to-right approach is simpler than the original top-down stack-pointer parser (not requiring a stack) and reduces transition sequence length in half, from 2 @math -1 actions to @math . This results in a quadratic non-projective parser that runs twice as fast as the original while achieving the best accuracy to date on the English PTB dataset (96.04 UAS, 94.43 LAS) among fully-supervised single-model dependency parsers, and improves over the former top-down transition system in the majority of languages tested.",
"The introduction of dynamic oracles has considerably improved the accuracy of greedy transition-based dependency parsers, without sacrificing parsing efficiency. However, this enhancement is limited to projective parsing, and dynamic oracles have not yet been implemented for parsers supporting non-projectivity. In this paper we introduce the first such oracle, for a non-projective parser based on Attardi’s parser. We show that training with this oracle improves parsing accuracy over a conventional (static) oracle on a wide range of datasets.",
"We present a novel transition system for dependency parsing, which constructs arcs only between adjacent words but can parse arbitrary non-projective trees by swapping the order of words in the input. Adding the swapping operation changes the time complexity for deterministic parsing from linear to quadratic in the worst case, but empirical estimates based on treebank data show that the expected running time is in fact linear for the range of data attested in the corpora. Evaluation on data from five languages shows state-of-the-art accuracy, with especially good results for the labeled exact match score.",
"",
"Recurrent neural network grammars (RNNG) are generative models of language which jointly model syntax and surface structure by incrementally generating a syntax tree and sentence in a top-down, left-to-right order. Supervised RNNGs achieve strong language modeling and parsing performance, but require an annotated corpus of parse trees. In this work, we experiment with unsupervised learning of RNNGs. Since directly marginalizing over the space of latent trees is intractable, we instead apply amortized variational inference. To maximize the evidence lower bound, we develop an inference network parameterized as a neural CRF constituency parser. On language modeling, unsupervised RNNGs perform as well their supervised counterparts on benchmarks in English and Chinese. On constituency grammar induction, they are competitive with recent neural language models that induce tree structures from words through attention mechanisms.",
"In this paper, we propose a method for analyzing word-word dependencies using deterministic bottom-up manner using Support Vector machines. We experimented with dependency trees converted from Penn treebank data, and achieved over 90 accuracy of word-word dependency. Though the result is little worse than the most up-to-date phrase structure based parsers, it looks satisfactorily accurate considering that our parser uses no information from phrase structures.",
"Parsing algorithms that process the input from left to right and construct a single derivation have often been considered inadequate for natural language parsing because of the massive ambiguity typically found in natural language grammars. Nevertheless, it has been shown that such algorithms, combined with treebank-induced classifiers, can be used to build highly accurate disambiguating parsers, in particular for dependency-based syntactic representations. In this article, we first present a general framework for describing and analyzing algorithms for deterministic incremental dependency parsing, formalized as transition systems. We then describe and analyze two families of such algorithms: stack-based and list-based algorithms. In the former family, which is restricted to projective dependency structures, we describe an arc-eager and an arc-standard variant; in the latter family, we present a projective and a non-projective variant. For each of the four algorithms, we give proofs of correctness and complexity. In addition, we perform an experimental evaluation of all algorithms in combination with SVM classifiers for predicting the next parsing action, using data from thirteen languages. We show that all four algorithms give competitive accuracy, although the non-projective list-based algorithm generally outperforms the projective algorithms for languages with a non-negligible proportion of non-projective constructions. However, the projective algorithms often produce comparable results when combined with the technique known as pseudo-projective parsing. The linear time complexity of the stack-based algorithms gives them an advantage with respect to efficiency both in learning and in parsing, but the projective list-based algorithm turns out to be equally efficient in practice. Moreover, when the projective algorithms are used to implement pseudo-projective parsing, they sometimes become less efficient in parsing (but not in learning) than the non-projective list-based algorithm. Although most of the algorithms have been partially described in the literature before, this is the first comprehensive analysis and evaluation of the algorithms within a unified framework.",
""
]
} |
1905.10930 | 2946260471 | We propose a method for non-projective dependency parsing by incrementally predicting a set of edges. Since the edges do not have a pre-specified order, we propose a set-based learning method. Our method blends graph, transition, and easy-first parsing, including a prior state of the parser as a special case. The proposed transition-based method successfully parses near the state of the art on both projective and non-projective languages, without assuming a certain parsing order. | A separate thread of research in sequential modeling has demonstrated that generation order can affect performance @cite_27 , both in tasks with set-structured outputs such as objects @cite_16 @cite_22 or graphs @cite_28 , and in sequential tasks such as language modeling @cite_4 . Developing models with relaxed or learned generation orders has picked up recent interest @cite_22 @cite_11 @cite_13 @cite_25 . We investigate this for dependency parsing, framing the problem as sequential set generation without a pre-specified order. | {
"cite_N": [
"@cite_4",
"@cite_22",
"@cite_28",
"@cite_27",
"@cite_16",
"@cite_13",
"@cite_25",
"@cite_11"
],
"mid": [
"2888693386",
"2963513093",
"2963555927",
"2173183968",
"2963576211",
"2913250058",
"2949644922",
"2912937082"
],
"abstract": [
"Neural language models are a critical component of state-of-the-art systems for machine translation, summarization, audio transcription, and other tasks. These language models are almost universally autoregressive in nature, generating sentences one token at a time from left to right. This paper studies the influence of token generation order on model quality via a novel two-pass language model that produces partially-filled sentence \"templates\" and then fills in missing tokens. We compare various strategies for structuring these two passes and observe a surprisingly large variation in model quality. We find the most effective strategy generates function words in the first pass followed by content words in the second. We believe these experimental results justify a more extensive investigation of generation order for neural language models.",
"We study the problem of multiset prediction. The goal of multiset prediction is to train a predictor that maps an input to a multiset consisting of multiple items. Unlike existing problems in supervised learning, such as classification, ranking and sequence generation, there is no known order among items in a target multiset, and each item in the multiset may appear more than once, making this problem extremely challenging. In this paper, we propose a novel multiset loss function by viewing this problem from the perspective of sequential decision making. The proposed multiset loss function is empirically evaluated on two families of datasets, one synthetic and the other real, with varying levels of difficulty, against various baseline loss functions including reinforcement learning, sequence, and aggregated distribution matching loss functions. The experiments reveal the effectiveness of the proposed loss function over the others.",
"Graphs are fundamental data structures required to model many important real-world data, from knowledge graphs, physical and social interactions to molecules and proteins. In this paper, we study the problem of learning generative models of graphs from a dataset of graphs of interest. After learning, these models can be used to generate samples with similar properties as the ones in the dataset. Such models can be useful in a lot of applications, e.g. drug discovery and knowledge graph construction. The task of learning generative models of graphs, however, has its unique challenges. In particular, how to handle symmetries in graphs and ordering of its elements during the generation process are important issues. We propose a generic graph neural net based model that is capable of generating any arbitrary graph. We study its performance on a few graph generation tasks compared to baselines that exploit domain knowledge. We discuss potential issues and open problems for such generative models going forward.",
"Sequences have become first class citizens in supervised learning thanks to the resurgence of recurrent neural networks. Many complex tasks that require mapping from or to a sequence of observations can now be formulated with the sequence-to-sequence (seq2seq) framework which employs the chain rule to efficiently represent the joint probability of sequences. In many cases, however, variable sized inputs and or outputs might not be naturally expressed as sequences. For instance, it is not clear how to input a set of numbers into a model where the task is to sort them; similarly, we do not know how to organize outputs when they correspond to random variables and the task is to model their unknown joint probability. In this paper, we first show using various examples that the order in which we organize input and or output data matters significantly when learning an underlying model. We then discuss an extension of the seq2seq framework that goes beyond sequences and handles input sets in a principled way. In addition, we propose a loss which, by searching over possible orders during training, deals with the lack of structure of output sets. We show empirical evidence of our claims regarding ordering, and on the modifications to the seq2seq framework on benchmark language modeling and parsing tasks, as well as two artificial tasks -- sorting numbers and estimating the joint probability of unknown graphical models.",
"Humans process visual scenes selectively and sequentially using attention. Central to models of human visual attention is the saliency map. We propose a hierarchical visual architecture that operates on a saliency map and uses a novel attention mechanism to sequentially focus on salient regions and take additional glimpses within those regions. The architecture is motivated by human visual attention, and is used for multi-label image classification on a novel multiset task, demonstrating that it achieves high precision and recall while localizing objects with its attention. Unlike conventional multi-label image classification models, the model supports multiset prediction due to a reinforcement-learning based training process that allows for arbitrary label permutation and multiple instances per label.",
"Conventional neural autoregressive decoding commonly assumes a fixed left-to-right generation order, which may be sub-optimal. In this work, we propose a novel decoding algorithm -- InDIGO -- which supports flexible sequence generation in arbitrary orders through insertion operations. We extend Transformer, a state-of-the-art sequence generation model, to efficiently implement the proposed approach, enabling it to be trained with either a pre-defined generation order or adaptive orders obtained from beam-search. Experiments on four real-world tasks, including word order recovery, machine translation, image caption and code generation, demonstrate that our algorithm can generate sequences following arbitrary orders, while achieving competitive or even better performance compared to the conventional left-to-right generation. The generated sequences show that InDIGO adopts adaptive generation orders based on input information.",
"We present the Insertion Transformer, an iterative, partially autoregressive model for sequence generation based on insertion operations. Unlike typical autoregressive models which rely on a fixed, often left-to-right ordering of the output, our approach accommodates arbitrary orderings by allowing for tokens to be inserted anywhere in the sequence during decoding. This flexibility confers a number of advantages: for instance, not only can our model be trained to follow specific orderings such as left-to-right generation or a binary tree traversal, but it can also be trained to maximize entropy over all valid insertions for robustness. In addition, our model seamlessly accommodates both fully autoregressive generation (one insertion at a time) and partially autoregressive generation (simultaneous insertions at multiple locations). We validate our approach by analyzing its performance on the WMT 2014 English-German machine translation task under various settings for training and decoding. We find that the Insertion Transformer outperforms many prior non-autoregressive approaches to translation at comparable or better levels of parallelism, and successfully recovers the performance of the original Transformer while requiring only logarithmically many iterations during decoding.",
""
]
} |
1905.10930 | 2946260471 | We propose a method for non-projective dependency parsing by incrementally predicting a set of edges. Since the edges do not have a pre-specified order, we propose a set-based learning method. Our method blends graph, transition, and easy-first parsing, including a prior state of the parser as a special case. The proposed transition-based method successfully parses near the state of the art on both projective and non-projective languages, without assuming a certain parsing order. | Finally, our work is inspired by techniques for improving upon maximum likelihood training through error exploration and dynamic oracles @cite_12 @cite_3 , and related techniques in imitation learning for structured prediction @cite_7 @cite_30 @cite_26 @cite_18 . In particular, our formulation is closely related to the framework of @cite_10 , where our oracle can be seen as an optimal roll-out policy which computes action costs without explicit roll-outs. | {
"cite_N": [
"@cite_30",
"@cite_18",
"@cite_26",
"@cite_7",
"@cite_3",
"@cite_10",
"@cite_12"
],
"mid": [
"2962957031",
"2518182934",
"2152588577",
"2949600092",
"2098532383",
"2952840881",
"595069947"
],
"abstract": [
"Sequential prediction problems such as imitation learning, where future observations depend on previous predictions (actions), violate the common i.i.d. assumptions made in statistical learning. This leads to poor performance in theory and often in practice. Some recent approaches (Daume , 2009; Ross and Bagnell, 2010) provide stronger guarantees in this setting, but remain somewhat unsatisfactory as they train either non-stationary or stochastic policies and require a large number of iterations. In this paper, we propose a new iterative algorithm, which trains a stationary deterministic policy, that can be seen as a no regret algorithm in an online learning setting. We show that any such no regret algorithm, combined with additional reduction assumptions, must find a policy with good performance under the distribution of observations it induces in such sequential settings. We demonstrate that this new approach outperforms previous approaches on two challenging imitation learning problems and a benchmark sequence labeling problem.",
"",
"Imitation Learning has been shown to be successful in solving many challenging real-world problems. Some recent approaches give strong performance guarantees by training the policy iteratively. However, it is important to note that these guarantees depend on how well the policy we found can imitate the oracle on the training data. When there is a substantial difference between the oracle's ability and the learner's policy space, we may fail to find a policy that has low error on the training set. In such cases, we propose to use a coach that demonstrates easy-to-learn actions for the learner and gradually approaches the oracle. By a reduction of learning by demonstration to online learning, we prove that coaching can yield a lower regret bound than using the oracle. We apply our algorithm to cost-sensitive dynamic feature selection, a hard decision problem that considers a user-specified accuracy-cost trade-off. Experimental results on UCI datasets show that our method outperforms state-of-the-art imitation learning methods in dynamic feature selection and two static feature selection methods.",
"We present Searn, an algorithm for integrating search and learning to solve complex structured prediction problems such as those that occur in natural language, speech, computational biology, and vision. Searn is a meta-algorithm that transforms these complex problems into simple classification problems to which any binary classifier may be applied. Unlike current algorithms for structured learning that require decomposition of both the loss function and the feature functions over the predicted structure, Searn is able to learn prediction functions for any loss function and any class of features. Moreover, Searn comes with a strong, natural theoretical guarantee: good performance on the derived classification problems implies good performance on the structured prediction problem.",
"Greedy transition-based parsers are very fast but tend to suffer from error propagation. This problem is aggravated by the fact that they are normally trained using oracles that are deterministic and incomplete in the sense that they assume a unique canonical path through the transition system and are only valid as long as the parser does not stray from this path. In this paper, we give a general characterization of oracles that are nondeterministic and complete, present a method for deriving such oracles for transition systems that satisfy a property we call arc decomposition, and instantiate this method for three well-known transition systems from the literature. We say that these oracles are dynamic, because they allow us to dynamically explore alternative and nonoptimal paths during training ‐ in contrast to oracles that statically assume a unique optimal path. Experimental evaluation on a wide range of data sets clearly shows that using dynamic oracles to train greedy parsers gives substantial improvements in accuracy. Moreover, this improvement comes at no cost in terms of efficiency, unlike other techniques like beam search.",
"Methods for learning to search for structured prediction typically imitate a reference policy, with existing theoretical guarantees demonstrating low regret compared to that reference. This is unsatisfactory in many applications where the reference policy is suboptimal and the goal of learning is to improve upon it. Can learning to search work even when the reference is poor? We provide a new learning to search algorithm, LOLS, which does well relative to the reference policy, but additionally guarantees low regret compared to deviations from the learned policy: a local-optimality guarantee. Consequently, LOLS can improve upon the reference policy, unlike previous algorithms. This enables us to develop structured contextual bandits, a partial information structured prediction setting with many potential applications.",
"The standard training regime for transition-based dependency parsers makes use of an oracle, which predicts an optimal transition sequence for a sentence and its gold tree. We present an improved oracle for the arc-eager transition system, which provides a set of optimal transitions for every valid parser configuration, including configurations from which the gold tree is not reachable. In such cases, the oracle provides transitions that will lead to the best reachable tree from the given configuration. The oracle is efficient to implement and provably correct. We use the oracle to train a deterministic left-to-right dependency parser that is less sensitive to error propagation, using an online training procedure that also explores parser configurations resulting from non-optimal sequences of transitions. This new parser outperforms greedy parsers trained using conventional oracles on a range of data sets, with an average improvement of over 1.2 LAS points and up to almost 3 LAS points on some data sets."
]
} |
1905.10706 | 2946544065 | Intelligent agents need a physical understanding of the world to predict the impact of their actions in the future. While learning-based models of the environment dynamics have contributed to significant improvements in sample efficiency compared to model-free reinforcement learning algorithms, they typically fail to generalize to system states beyond the training data, while often grounding their predictions on non-interpretable latent variables. We introduce Interactive Differentiable Simulation (IDS), a differentiable physics engine, that allows for efficient, accurate inference of physical properties of rigid-body systems. Integrated into deep learning architectures, our model is able to accomplish system identification using visual input, leading to an interpretable model of the world whose parameters have physical meaning. We present experiments showing automatic task-based robot design and parameter estimation for nonlinear dynamical systems by automatically calculating gradients in IDS. When integrated into an adaptive model-predictive control algorithm, our approach exhibits orders of magnitude improvements in sample efficiency over model-free reinforcement learning algorithms on challenging nonlinear control domains. | @cite_30 implemented a differentiable physics engine in the automatic differentiation framework Theano. IDS is implemented in C++ using Stan Math @cite_9 which enables reverse-mode automatic differentiation to efficiently compute gradients, even in cases where the code branches significantly. Analytical gradients of rigid-body dynamics algorithms have been implemented efficiently in the Pinnocchio library @cite_29 to facilitate optimal control and inverse kinematics. These are less general than our approach since they can only be used to optimize for a number of hand-engineered quantities. Simulating non-penetrative multi-point contacts between rigid bodies requires solving a linear complementarity problem (LCP), through which @cite_11 differentiate using the differentiable quadratic program solver OptNet @cite_4 . While our proposed model does not yet incorporate contact dynamics, we are able to demonstrate the scalability of our approach on versatile applications of differentiable physics to common 3D control domains. | {
"cite_N": [
"@cite_30",
"@cite_4",
"@cite_9",
"@cite_29",
"@cite_11"
],
"mid": [
"2556096037",
"2963970238",
"2195098255",
"2804623852",
"2891122218"
],
"abstract": [
"One of the most important fields in robotics is the optimization of controllers. Currently, robots are often treated as a black box in this optimization process, which is the reason why derivative-free optimization methods such as evolutionary algorithms or reinforcement learning are omnipresent. When gradient-based methods are used, models are kept small or rely on finite difference approximations for the Jacobian. This method quickly grows expensive with increasing numbers of parameters, such as found in deep learning. We propose an implementation of a modern physics engine, which can differentiate control parameters. This engine is implemented for both CPU and GPU. Firstly, this paper shows how such an engine speeds up the optimization process, even for small problems. Furthermore, it explains why this is an alternative approach to deep Q-learning, for using deep learning in robotics. Finally, we argue that this is a big step for deep learning in robotics, as it opens up new possibilities to optimize robots, both in hardware and software.",
"This paper presents OptNet, a network architecture that integrates optimization problems (here, specifically in the form of quadratic programs) as individual layers in larger end-to-end train-able deep networks. These layers encode constraints and complex dependencies between the hidden states that traditional convolutional and fully-connected layers often cannot capture. In this paper, we explore the foundations for such an architecture: we show how techniques from sensitivity analysis, bilevel optimization, and implicit differentiation can be used to exactly differentiate through these layers and with respect to layer parameters; we develop a highly efficient solver for these layers that exploits fast GPU-based batch solves within a primal-dual interior point method, and which provides backpropagation gradients with virtually no additional cost on top of the solve; and we highlight the application of these approaches in several problems. In one notable example, we show that the method is capable of learning to play mini-Sudoku (4x4) given just input and output games, with no a priori information about the rules of the game; this highlights the ability of our architecture to learn hard constraints better than other neural architectures.",
"As computational challenges in optimization and statistical inference grow ever harder, algorithms that utilize derivatives are becoming increasingly more important. The implementation of the derivatives that make these algorithms so powerful, however, is a substantial user burden and the practicality of these algorithms depends critically on tools like automatic differentiation that remove the implementation burden entirely. The Stan Math Library is a C++, reverse-mode automatic differentiation library designed to be usable, extensive and extensible, efficient, scalable, stable, portable, and redistributable in order to facilitate the construction and utilization of such algorithms. Usability is achieved through a simple direct interface and a cleanly abstracted functional interface. The extensive built-in library includes functions for matrix operations, linear algebra, differential equation solving, and most common probability functions. Extensibility derives from a straightforward object-oriented framework for expressions, allowing users to easily create custom functions. Efficiency is achieved through a combination of custom memory management, subexpression caching, traits-based metaprogramming, and expression templates. Partial derivatives for compound functions are evaluated lazily for improved scalability. Stability is achieved by taking care with arithmetic precision in algebraic expressions and providing stable, compound functions where possible. For portability, the library is standards-compliant C++ (03) and has been tested for all major compilers for Windows, Mac OS X, and Linux.",
"Rigid body dynamics is a well-established framework in robotics. It can be used to expose the analytic form of kinematic and dynamic functions of the robot model. So far, two major algorithms, namely the recursive Newton-Euler algorithm (RNEA) and the articulated body algorithm (ABA), have been proposed to compute the inverse dynamics and the forward dynamics in a few microseconds. Evaluating their derivatives is an important challenge for various robotic applications (optimal control, estimation, co-design or reinforcement learning). However it remains time consuming, whether using finite differences or automatic differentiation. In this paper, we propose new algorithms to efficiently compute them thanks to closed-form formulations. Using the chain rule and adequate algebraic differentiation of spatial algebra, we firstly differentiate explicitly RNEA. Then, using properties about the derivative of function composition, we show that the same algorithm can also be used to compute the derivatives of ABA with a marginal additional cost. For this purpose, we introduce a new algorithm to compute the inverse of the joint-space inertia matrix, without explicitly computing the matrix itself. All the algorithms are implemented in our open-source C++ framework called Pinocchio. Benchmarks show computational costs varying between 3 microseconds (for a 7-dof arm) up to 17 microseconds (for a 36-dof humanoid), outperforming the alternative approaches of the state of the art.",
"We present a differentiable physics engine that can be integrated as a module in deep neural networks for end-to-end learning. As a result, structured physics knowledge can be embedded into larger systems, allowing them, for example, to match observations by performing precise simulations, while achieves high sample efficiency. Specifically, in this paper we demonstrate how to perform backpropagation analytically through a physical simulator defined via a linear complementarity problem. Unlike traditional finite difference methods, such gradients can be computed analytically, which allows for greater flexibility of the engine. Through experiments in diverse domains, we highlight the system's ability to learn physical parameters from data, efficiently match and simulate observed visual behavior, and readily enable control via gradient-based planning methods."
]
} |
1905.10706 | 2946544065 | Intelligent agents need a physical understanding of the world to predict the impact of their actions in the future. While learning-based models of the environment dynamics have contributed to significant improvements in sample efficiency compared to model-free reinforcement learning algorithms, they typically fail to generalize to system states beyond the training data, while often grounding their predictions on non-interpretable latent variables. We introduce Interactive Differentiable Simulation (IDS), a differentiable physics engine, that allows for efficient, accurate inference of physical properties of rigid-body systems. Integrated into deep learning architectures, our model is able to accomplish system identification using visual input, leading to an interpretable model of the world whose parameters have physical meaning. We present experiments showing automatic task-based robot design and parameter estimation for nonlinear dynamical systems by automatically calculating gradients in IDS. When integrated into an adaptive model-predictive control algorithm, our approach exhibits orders of magnitude improvements in sample efficiency over model-free reinforcement learning algorithms on challenging nonlinear control domains. | The approach of adapting the simulator to real world dynamics, which we demonstrate through our adaptive MPC algorithm in Sec. , has been less explored. While many previous works have shown to adapt simulators to the real world using system identification and state estimation @cite_8 @cite_0 , few have shown adaptive model-based control schemes that actively close the feedback loop between the real and the simulated system @cite_15 @cite_5 @cite_18 . Instead of using a simulator, model-based reinforcement learning is a broader field @cite_16 , where the system dynamics, and state-action transitions in particular, are learned to achieve higher sample efficiency compared to model-free methods. Within this framework, predominantly Gaussian Processes @cite_20 @cite_28 @cite_25 and neural networks @cite_21 @cite_3 have been proposed to learn the dynamics and optimize policies. | {
"cite_N": [
"@cite_18",
"@cite_8",
"@cite_28",
"@cite_21",
"@cite_3",
"@cite_0",
"@cite_5",
"@cite_15",
"@cite_16",
"@cite_25",
"@cite_20"
],
"mid": [
"",
"2209762605",
"",
"2410617946",
"",
"",
"",
"1989106069",
"2580909119",
"",
"2162717641"
],
"abstract": [
"",
"Successful model based control relies heavily on proper system identification and accurate state estimation. We present a framework for solving these problems in the context of robotic control applications. We are particularly interested in robotic manipulation tasks, which are especially hard due to the non-linear nature of contact phenomena. We developed a solution that solves both the problems of estimation and system identification jointly. We show that these two problems are difficult to solve separately in the presence of discontinuous phenomena such as contacts. The problem is posed as a joint optimization across both trajectory and model parameters and solved via Newton's method. We present several challenges we encountered while modeling contacts and performing state estimation and propose solutions within the MuJoCo physics engine. We present experimental results performed on our manipulation system consisting of 3-DOF Phantom Haptic Devices, turned into finger manipulators. Cross-validation between different datasets, as well as leave-one-out cross-validation show that our method is robust and is able to accurately explain sensory data.",
"",
"In this paper we present a model predictive control algorithm designed for optimizing non-linear systems subject to complex cost criteria. The algorithm is based on a stochastic optimal control framework using a fundamental relationship between the information theoretic notions of free energy and relative entropy. The optimal controls in this setting take the form of a path integral, which we approximate using an efficient importance sampling scheme. We experimentally verify the algorithm by implementing it on a Graphics Processing Unit (GPU) and apply it to the problem of controlling a fifth-scale Auto-Rally vehicle in an aggressive driving task.",
"",
"",
"",
"Computer simulation is an essential step in the design and construction of various mechanical structures, including biped robots, because it enables rapid testing and virtual prototyping during the construction phase. Although many different simulators are available, this article gives an overview and a motivation for building a new dynamic multibody simulator. The simulator is especially adapted to humanoid robot Archie, developed at the IHRT Institute at the Technical University of Vienna. In addition, it is shown how the simulator can be used not only in the controller design, but also in the online control loop to extend the available sensors: a virtual sensors principle.",
"Reinforcement learning is an appealing approach for allowing robots to learn new tasks. Relevant literature reveals a plethora of methods, but at the same time makes clear the lack of implementations for dealing with real life challenges. Current expectations raise the demand for adaptable robots. We argue that, by employing model-based reinforcement learning, the--now limited--adaptability characteristics of robotic systems can be expanded. Also, model-based reinforcement learning exhibits advantages that makes it more applicable to real life use-cases compared to model-free methods. Thus, in this survey, model-based methods that have been applied in robotics are covered. We categorize them based on the derivation of an optimal policy, the definition of the returns function, the type of the transition model and the learned task. Finally, we discuss the applicability of model-based reinforcement learning approaches in new applications, taking into consideration the state of the art in both algorithms and hardware.",
"",
"Blimps are a promising platform for aerial robotics and have been studied extensively for this purpose. Unlike other aerial vehicles, blimps are relatively safe and also possess the ability to loiter for long periods. These advantages, however, have been difficult to exploit because blimp dynamics are complex and inherently non-linear. The classical approach to system modeling represents the system as an ordinary differential equation (ODE) based on Newtonian principles. A more recent modeling approach is based on representing state transitions as a Gaussian process (GP). In this paper, we present a general technique for system identification that combines these two modeling approaches into a single formulation. This is done by training a Gaussian process on the residual between the non-linear model and ground truth training data. The result is a GP-enhanced model that provides an estimate of uncertainty in addition to giving better state predictions than either ODE or GP alone. We show how the GP-enhanced model can be used in conjunction with reinforcement learning to generate a blimp controller that is superior to those learned with ODE or GP models alone."
]
} |
1905.10695 | 2946213629 | Deep Neural Networks (DNNs) are vulnerable to adversarial attacks, especially white-box targeted attacks. One scheme of learning attacks is to design a proper adversarial objective function that leads to the imperceptible perturbation for any test image (e.g., the Carlini-Wagner (C&W) method). Most methods address targeted attacks in the Top-1 manner. In this paper, we propose to learn ordered Top-k attacks (k>= 1) for image classification tasks, that is to enforce the Top-k predicted labels of an adversarial example to be the k (randomly) selected and ordered labels (the ground-truth label is exclusive). To this end, we present an adversarial distillation framework: First, we compute an adversarial probability distribution for any given ordered Top-k targeted labels with respect to the ground-truth of a test image. Then, we learn adversarial examples by minimizing the Kullback-Leibler (KL) divergence together with the perturbation energy penalty, similar in spirit to the network distillation method. We explore how to leverage label semantic similarities in computing the targeted distributions, leading to knowledge-oriented attacks. In experiments, we thoroughly test Top-1 and Top-5 attacks in the ImageNet-1000 validation dataset using two popular DNNs trained with clean ImageNet-1000 train dataset, ResNet-50 and DenseNet-121. For both models, our proposed adversarial distillation approach outperforms the C&W method in the Top-1 setting, as well as other baseline methods. Our approach shows significant improvement in the Top-5 setting against a strong modified C&W method. | The growing ubiquity of DNNs in advanced machine learning and AI systems dramatically increases their capabilities, but also increases the potential for new vulnerabilities to attacks. This situation has become critical as many powerful approaches have been developed where imperceptible perturbations to DNN inputs could deceive a well-trained DNN, significantly altering its prediction. Please refer to @cite_28 for a comprehensive survey of attack methods in computer vision. We review some related work that motivate our work and show the difference. | {
"cite_N": [
"@cite_28"
],
"mid": [
"2962700793"
],
"abstract": [
"Deep learning is at the heart of the current rise of artificial intelligence. In the field of computer vision, it has become the workhorse for applications ranging from self-driving cars to surveillance and security. Whereas, deep neural networks have demonstrated phenomenal success (often beyond human capabilities) in solving complex problems, recent studies show that they are vulnerable to adversarial attacks in the form of subtle perturbations to inputs that lead a model to predict incorrect outputs. For images, such perturbations are often too small to be perceptible, yet they completely fool the deep learning models. Adversarial attacks pose a serious threat to the success of deep learning in practice. This fact has recently led to a large influx of contributions in this direction. This paper presents the first comprehensive survey on adversarial attacks on deep learning in computer vision. We review the works that design adversarial attacks, analyze the existence of such attacks and propose defenses against them. To emphasize that adversarial attacks are possible in practical conditions, we separately review the contributions that evaluate adversarial attacks in the real-world scenarios. Finally, drawing on the reviewed literature, we provide a broader outlook of this research direction."
]
} |
1905.10695 | 2946213629 | Deep Neural Networks (DNNs) are vulnerable to adversarial attacks, especially white-box targeted attacks. One scheme of learning attacks is to design a proper adversarial objective function that leads to the imperceptible perturbation for any test image (e.g., the Carlini-Wagner (C&W) method). Most methods address targeted attacks in the Top-1 manner. In this paper, we propose to learn ordered Top-k attacks (k>= 1) for image classification tasks, that is to enforce the Top-k predicted labels of an adversarial example to be the k (randomly) selected and ordered labels (the ground-truth label is exclusive). To this end, we present an adversarial distillation framework: First, we compute an adversarial probability distribution for any given ordered Top-k targeted labels with respect to the ground-truth of a test image. Then, we learn adversarial examples by minimizing the Kullback-Leibler (KL) divergence together with the perturbation energy penalty, similar in spirit to the network distillation method. We explore how to leverage label semantic similarities in computing the targeted distributions, leading to knowledge-oriented attacks. In experiments, we thoroughly test Top-1 and Top-5 attacks in the ImageNet-1000 validation dataset using two popular DNNs trained with clean ImageNet-1000 train dataset, ResNet-50 and DenseNet-121. For both models, our proposed adversarial distillation approach outperforms the C&W method in the Top-1 setting, as well as other baseline methods. Our approach shows significant improvement in the Top-5 setting against a strong modified C&W method. | For image classification tasks using DNNs, the discovery of the existence of visually-imperceptible adversarial attacks @cite_8 was a big shock in developing DNNs. White-box attacks provide a powerful way of evaluating model brittleness. In a plain and loose explanation, DNNs are universal function approximator @cite_29 and capable of even fitting random labels @cite_30 in large scale classification tasks as ImageNet-1000 @cite_5 . Thus, adversarial attacks are always learnable provided proper objective functions are given, especially when DNNs are trained with fully differentible back-propagation. Many white-box attack methods focus on norm-ball constrained objective functions @cite_8 @cite_3 @cite_27 @cite_24 . The C &W method investigates 7 different loss functions. The best performing loss function found by the C &W method has been appliedin many attack methods and achieved strong results @cite_21 @cite_22 @cite_26 . By introducing momentum in the MIFGSM method @cite_24 and the @math gradient projection in the PGD method @cite_17 , they usually achieve better performance in generating adversarial examples. In the meanwhile, some other attack methods such as the StrAttack @cite_23 also investigate different loss functions for better interpretability of attacks. Our proposed method leverage label semantic knowledge in the loss function design for the first time. | {
"cite_N": [
"@cite_30",
"@cite_26",
"@cite_22",
"@cite_8",
"@cite_29",
"@cite_21",
"@cite_3",
"@cite_24",
"@cite_27",
"@cite_23",
"@cite_5",
"@cite_17"
],
"mid": [
"2566079294",
"",
"",
"2964153729",
"2137983211",
"2746600820",
"2963542245",
"2774644650",
"2516574342",
"2951954400",
"2117539524",
"2964253222"
],
"abstract": [
"Despite their massive size, successful deep artificial neural networks can exhibit a remarkably small difference between training and test performance. Conventional wisdom attributes small generalization error either to properties of the model family, or to the regularization techniques used during training. @PARASPLIT Through extensive systematic experiments, we show how these traditional approaches fail to explain why large neural networks generalize well in practice. Specifically, our experiments establish that state-of-the-art convolutional networks for image classification trained with stochastic gradient methods easily fit a random labeling of the training data. This phenomenon is qualitatively unaffected by explicit regularization, and occurs even if we replace the true images by completely unstructured random noise. We corroborate these experimental findings with a theoretical construction showing that simple depth two neural networks already have perfect finite sample expressivity as soon as the number of parameters exceeds the number of data points as it usually does in practice. @PARASPLIT We interpret our experimental findings by comparison with traditional models.",
"",
"",
"Abstract: Deep neural networks are highly expressive models that have recently achieved state of the art performance on speech and visual recognition tasks. While their expressiveness is the reason they succeed, it also causes them to learn uninterpretable solutions that could have counter-intuitive properties. In this paper we report two such properties. First, we find that there is no distinction between individual high level units and random linear combinations of high level units, according to various methods of unit analysis. It suggests that it is the space, rather than the individual units, that contains of the semantic information in the high layers of neural networks. Second, we find that deep neural networks learn input-output mappings that are fairly discontinuous to a significant extend. We can cause the network to misclassify an image by applying a certain imperceptible perturbation, which is found by maximizing the network's prediction error. In addition, the specific nature of these perturbations is not a random artifact of learning: the same perturbation can cause a different network, that was trained on a different subset of the dataset, to misclassify the same input.",
"Abstract This paper rigorously establishes that standard multilayer feedforward networks with as few as one hidden layer using arbitrary squashing functions are capable of approximating any Borel measurable function from one finite dimensional space to another to any desired degree of accuracy, provided sufficiently many hidden units are available. In this sense, multilayer feedforward networks are a class of universal approximators.",
"Deep neural networks (DNNs) are one of the most prominent technologies of our time, as they achieve state-of-the-art performance in many machine learning tasks, including but not limited to image classification, text mining, and speech processing. However, recent research on DNNs has indicated ever-increasing concern on the robustness to adversarial examples, especially for security-critical tasks such as traffic sign identification for autonomous driving. Studies have unveiled the vulnerability of a well-trained DNN by demonstrating the ability of generating barely noticeable (to both human and machines) adversarial images that lead to misclassification. Furthermore, researchers have shown that these adversarial images are highly transferable by simply training and attacking a substitute model built upon the target model, known as a black-box attack to DNNs. Similar to the setting of training substitute models, in this paper we propose an effective black-box attack that also only has access to the input (images) and the output (confidence scores) of a targeted DNN. However, different from leveraging attack transferability from substitute models, we propose zeroth order optimization (ZOO) based attacks to directly estimate the gradients of the targeted DNN for generating adversarial examples. We use zeroth order stochastic coordinate descent along with dimension reduction, hierarchical attack and importance sampling techniques to efficiently attack black-box models. By exploiting zeroth order optimization, improved attacks to the targeted DNN can be accomplished, sparing the need for training substitute models and avoiding the loss in attack transferability. Experimental results on MNIST, CIFAR10 and ImageNet show that the proposed ZOO attack is as effective as the state-of-the-art white-box attack (e.g., Carlini and Wagner's attack) and significantly outperforms existing black-box attacks via substitute models.",
"Most existing machine learning classifiers are highly vulnerable to adversarial examples. An adversarial example is a sample of input data which has been modified very slightly in a way that is intended to cause a machine learning classifier to misclassify it. In many cases, these modifications can be so subtle that a human observer does not even notice the modification at all, yet the classifier still makes a mistake. Adversarial examples pose security concerns because they could be used to perform an attack on machine learning systems, even if the adversary has no access to the underlying model. Up to now, all previous work has assumed a threat model in which the adversary can feed data directly into the machine learning classifier. This is not always the case for systems operating in the physical world, for example those which are using signals from cameras and other sensors as input. This paper shows that even in such physical world scenarios, machine learning systems are vulnerable to adversarial examples. We demonstrate this by feeding adversarial images obtained from a cell-phone camera to an ImageNet Inception classifier and measuring the classification accuracy of the system. We find that a large fraction of adversarial examples are classified incorrectly even when perceived through the camera.",
"Deep neural networks are vulnerable to adversarial examples, which poses security concerns on these algorithms due to the potentially severe consequences. Adversarial attacks serve as an important surrogate to evaluate the robustness of deep learning models before they are deployed. However, most of existing adversarial attacks can only fool a black-box model with a low success rate. To address this issue, we propose a broad class of momentum-based iterative algorithms to boost adversarial attacks. By integrating the momentum term into the iterative process for attacks, our methods can stabilize update directions and escape from poor local maxima during the iterations, resulting in more transferable adversarial examples. To further improve the success rates for black-box attacks, we apply momentum iterative algorithms to an ensemble of models, and show that the adversarially trained models with a strong defense ability are also vulnerable to our black-box attacks. We hope that the proposed methods will serve as a benchmark for evaluating the robustness of various deep models and defense methods. With this method, we won the first places in NIPS 2017 Non-targeted Adversarial Attack and Targeted Adversarial Attack competitions.",
"Neural networks provide state-of-the-art results for most machine learning tasks. Unfortunately, neural networks are vulnerable to adversarial examples: given an input @math and any target classification @math , it is possible to find a new input @math that is similar to @math but classified as @math . This makes it difficult to apply neural networks in security-critical areas. Defensive distillation is a recently proposed approach that can take an arbitrary neural network, and increase its robustness, reducing the success rate of current attacks' ability to find adversarial examples from @math to @math . In this paper, we demonstrate that defensive distillation does not significantly increase the robustness of neural networks by introducing three new attack algorithms that are successful on both distilled and undistilled neural networks with @math probability. Our attacks are tailored to three distance metrics used previously in the literature, and when compared to previous adversarial example generation algorithms, our attacks are often much more effective (and never worse). Furthermore, we propose using high-confidence adversarial examples in a simple transferability test we show can also be used to break defensive distillation. We hope our attacks will be used as a benchmark in future defense attempts to create neural networks that resist adversarial examples.",
"When generating adversarial examples to attack deep neural networks (DNNs), Lp norm of the added perturbation is usually used to measure the similarity between original image and adversarial example. However, such adversarial attacks perturbing the raw input spaces may fail to capture structural information hidden in the input. This work develops a more general attack model, i.e., the structured attack (StrAttack), which explores group sparsity in adversarial perturbations by sliding a mask through images aiming for extracting key spatial structures. An ADMM (alternating direction method of multipliers)-based framework is proposed that can split the original problem into a sequence of analytically solvable subproblems and can be generalized to implement other attacking methods. Strong group sparsity is achieved in adversarial perturbations even with the same level of Lp norm distortion as the state-of-the-art attacks. We demonstrate the effectiveness of StrAttack by extensive experimental results onMNIST, CIFAR-10, and ImageNet. We also show that StrAttack provides better interpretability (i.e., better correspondence with discriminative image regions)through adversarial saliency map (, 2016b) and class activation map(, 2016).",
"The ImageNet Large Scale Visual Recognition Challenge is a benchmark in object category classification and detection on hundreds of object categories and millions of images. The challenge has been run annually from 2010 to present, attracting participation from more than fifty institutions. This paper describes the creation of this benchmark dataset and the advances in object recognition that have been possible as a result. We discuss the challenges of collecting large-scale ground truth annotation, highlight key breakthroughs in categorical object recognition, provide a detailed analysis of the current state of the field of large-scale image classification and object detection, and compare the state-of-the-art computer vision accuracy with human accuracy. We conclude with lessons learned in the 5 years of the challenge, and propose future directions and improvements.",
"Recent work has demonstrated that neural networks are vulnerable to adversarial examples, i.e., inputs that are almost indistinguishable from natural data and yet classified incorrectly by the network. To address this problem, we study the adversarial robustness of neural networks through the lens of robust optimization. This approach provides us with a broad and unifying view on much prior work on this topic. Its principled nature also enables us to identify methods for both training and attacking neural networks that are reliable and, in a certain sense, universal. In particular, they specify a concrete security guarantee that would protect against a well-defined class of adversaries. These methods let us train networks with significantly improved resistance to a wide range of adversarial attacks. They also suggest robustness against a first-order adversary as a natural and broad security guarantee. We believe that robustness against such well-defined classes of adversaries is an important stepping stone towards fully resistant deep learning models."
]
} |
1905.10702 | 2944878684 | Over the past decade, knowledge graphs became popular for capturing structured domain knowledge. Relational learning models enable the prediction of missing links inside knowledge graphs. More specifically, latent distance approaches model the relationships among entities via a distance between latent representations. Translating embedding models (e.g., TransE) are among the most popular latent distance approaches which use one distance function to learn multiple relation patterns. However, they are not capable of capturing symmetric relations. They also force relations with reflexive patterns to become symmetric and transitive. In order to improve distance based embedding, we propose multi-distance embeddings (MDE). Our solution is based on the idea that by learning independent embedding vectors for each entity and relation one can aggregate contrasting distance functions. Benefiting from MDE, we also develop supplementary distances resolving the above-mentioned limitations of TransE. We further propose an extended loss function for distance based embeddings and show that MDE and TransE are fully expressive using this loss function. Furthermore, we obtain a bound on the size of their embeddings for full expressivity. Our empirical results show that MDE significantly improves the translating embeddings and outperforms several state-of-the-art embedding models on benchmark datasets. | define the score of triples via pairwise multiplication of embeddings. Dismult @cite_9 simply multiplies the embedding vectors of a triple element by element @math as the score function. Since multiplication of real numbers is symmetric, Dismult can not distinguish displacement of head relation and tail entities and therefore it can not model anti-symmetric relations. To solve this limitation SimplE @cite_29 collaborates the reverse of relations to Dismult and ties a relation and its inverse. ComplEx @cite_5 solves the same issue of DistMult by the idea that multiplication of complex values is not symmetric. By introducing complex-valued embeddings instead of real-valued vectors to dismult, the score of a triple in ComplEx is @math with @math the conjugate of t and @math is the real part of a complex value. In RESCAL @cite_21 instead of a vector, a matrix represents @math , and performs outer products of @math and @math vectors to this matrix so that its score function becomes @math . A simplified version of RESCAL is HolE @cite_24 that defines a vector for @math and performs circular correlation of @math and @math has been found equivalent @cite_30 to ComplEx. | {
"cite_N": [
"@cite_30",
"@cite_9",
"@cite_29",
"@cite_21",
"@cite_24",
"@cite_5"
],
"mid": [
"2593682006",
"",
"2962850650",
"",
"2145544171",
"2432356473"
],
"abstract": [
"We show the equivalence of two state-of-the-art link prediction knowledge graph completion methods: 's holographic embedding and 's complex embedding. We first consider a spectral version of the holographic embedding, exploiting the frequency domain in the Fourier transform for efficient computation. The analysis of the resulting method reveals that it can be viewed as an instance of the complex embedding with certain constraints cast on the initial vectors upon training. Conversely, any complex embedding can be converted to an equivalent holographic embedding.",
"",
"Knowledge graphs contain knowledge about the world and provide a structured representation of this knowledge. Current knowledge graphs contain only a small subset of what is true in the world. Link prediction approaches aim at predicting new links for a knowledge graph given the existing links between the entities. Tensor factorization approaches have proved promising for such link prediction problems. Proposed in 1927, Canonical Polyadic (CP) decomposition is among the first tensor factorization approaches. CP generally performs poorly for link prediction as it learns two independent embedding vectors for each entity, whereas they are really tied. We present a simple enhancement (which we call SimplE) of CP to allow the two embeddings of each entity to be learned dependently. The complexity of SimplE grows linearly with the size of embeddings. The embeddings learned through SimplE are interpretable, and certain types of background knowledge in terms of logical rules can be incorporated into these embeddings through weight tying. We prove SimplE is fully-expressive and derive a bound on the size of its embeddings for full expressivity. We show empirically that, despite its simplicity, SimplE outperforms several state-of-the-art tensor factorization techniques.",
"",
"Learning embeddings of entities and relations is an efficient and versatile method to perform machine learning on relational data such as knowledge graphs. In this work, we propose holographic embeddings (HOLE) to learn compositional vector space representations of entire knowledge graphs. The proposed method is related to holographic models of associative memory in that it employs circular correlation to create compositional representations. By using correlation as the compositional operator, HOLE can capture rich interactions but simultaneously remains efficient to compute, easy to train, and scalable to very large datasets. Experimentally, we show that holographic embeddings are able to outperform state-of-the-art methods for link prediction on knowledge graphs and relational learning benchmark datasets.",
"In statistical relational learning, the link prediction problem is key to automatically understand the structure of large knowledge bases. As in previous studies, we propose to solve this problem through latent factorization. However, here we make use of complex valued embeddings. The composition of complex embeddings can handle a large variety of binary relations, among them symmetric and antisymmetric relations. Compared to state-of-the-art models such as Neural Tensor Network and Holographic Embeddings, our approach based on complex embeddings is arguably simpler, as it only uses the Hermitian dot product, the complex counterpart of the standard dot product between real vectors. Our approach is scalable to large datasets as it remains linear in both space and time, while consistently outperforming alternative approaches on standard link prediction benchmarks."
]
} |
1905.10702 | 2944878684 | Over the past decade, knowledge graphs became popular for capturing structured domain knowledge. Relational learning models enable the prediction of missing links inside knowledge graphs. More specifically, latent distance approaches model the relationships among entities via a distance between latent representations. Translating embedding models (e.g., TransE) are among the most popular latent distance approaches which use one distance function to learn multiple relation patterns. However, they are not capable of capturing symmetric relations. They also force relations with reflexive patterns to become symmetric and transitive. In order to improve distance based embedding, we propose multi-distance embeddings (MDE). Our solution is based on the idea that by learning independent embedding vectors for each entity and relation one can aggregate contrasting distance functions. Benefiting from MDE, we also develop supplementary distances resolving the above-mentioned limitations of TransE. We further propose an extended loss function for distance based embeddings and show that MDE and TransE are fully expressive using this loss function. Furthermore, we obtain a bound on the size of their embeddings for full expressivity. Our empirical results show that MDE significantly improves the translating embeddings and outperforms several state-of-the-art embedding models on benchmark datasets. | train a neural network to learn the interaction of the , and . ER-MLP @cite_18 is a two layer feedforward neural network considering @math , @math and @math vectors in the input. | {
"cite_N": [
"@cite_18"
],
"mid": [
"2016753842"
],
"abstract": [
"Recent years have witnessed a proliferation of large-scale knowledge bases, including Wikipedia, Freebase, YAGO, Microsoft's Satori, and Google's Knowledge Graph. To increase the scale even further, we need to explore automatic methods for constructing knowledge bases. Previous approaches have primarily focused on text-based extraction, which can be very noisy. Here we introduce Knowledge Vault, a Web-scale probabilistic knowledge base that combines extractions from Web content (obtained via analysis of text, tabular data, page structure, and human annotations) with prior knowledge derived from existing knowledge repositories. We employ supervised machine learning methods for fusing these distinct information sources. The Knowledge Vault is substantially bigger than any previously published structured knowledge repository, and features a probabilistic inference system that computes calibrated probabilities of fact correctness. We report the results of multiple studies that explore the relative utility of the different information sources and extraction methods."
]
} |
1905.10478 | 2946624195 | Tensor decomposition is an effective approach to compress over-parameterized neural networks and to enable their deployment on resource-constrained hardware platforms. However, directly applying tensor compression in the training process is a challenging task due to the difficulty of choosing a proper tensor rank. In order to achieve this goal, this paper proposes a Bayesian tensorized neural network. Our Bayesian method performs automatic model compression via an adaptive tensor rank determination. We also present approaches for posterior density calculation and maximum a posteriori (MAP) estimation for the end-to-end training of our tensorized neural network. We provide experimental validation on a fully connected neural network, a CNN and a residual neural network where our work produces @math to @math more compact neural networks directly from the training. | Tensor Compression of Neural Networks. CP-format tensor decomposition @cite_12 @cite_18 was first employed to compress the fully connected (FC) layers of pre-trained models by . Following work then compressed both FC and convolutional layers by tensor train decomposition @cite_32 and by Tucker decomposition @cite_26 , respectively. The results in @cite_32 @cite_26 show that the FC layers can be compressed more significantly than the convolution layers. In order to avoid the expensive pre-training, and trained FC layers in low-rank tensor-train and Tucker formats, respectively, with the tensor ranks fixed in advance. In practice determining a proper tensor rank a priori is hard, and a bad rank estimation can result in low accuracy or high training cost. This challenge is the main motivation of our work. | {
"cite_N": [
"@cite_18",
"@cite_26",
"@cite_32",
"@cite_12"
],
"mid": [
"1974403130",
"2177847924",
"2559813832",
"2000215628"
],
"abstract": [
"We review the method of Parallel Factor Analysis, which simultaneously fits multiple two-way arrays or ‘slices’ of a three-way array in terms of a common set of factors with differing relative weights in each ‘slice’. Mathematically, it is a straightforward generalization of the bilinear model of factor (or component) analysis (xij = ΣRr = 1airbjr) to a trilinear model (xijk = ΣRr = 1airbjrckr). Despite this simplicity, it has an important property not possessed by the two-way model: if the latent factors show adequately distinct patterns of three-way variation, the model is fully identified; the orientation of factors is uniquely determined by minimizing residual error, eliminating the need for a separate ‘rotation’ phase of analysis. The model can be used several ways. It can be directly fit to a three-way array of observations with (possibly incomplete) factorial structure, or it can be indirectly fit to the original observations by fitting a set of covariance matrices computed from the observations, with each matrix corresponding to a two-way subset of the data. Even more generally, one can simultaneously analyze covariance matrices computed from different samples, perhaps corresponding to different treatment groups, different kinds of cases, data from different studies, etc. To demonstrate the method we analyze data from an experiment on right vs. left cerebral hemispheric control of the hands during various tasks. The factors found appear to correspond to the causal influences manipulated in the experiment, revealing their patterns of influence in all three ways of the data. Several generalizations of the parallel factor analysis model are currently under development, including ones that combine parallel factors with Tucker-like factor ‘interactions’. Of key importance is the need to increase the method's robustness against nonstationary factor structures and qualitative (nonproportional) factor change.",
"Although the latest high-end smartphone has powerful CPU and GPU, running deeper convolutional neural networks (CNNs) for complex tasks such as ImageNet classification on mobile devices is challenging. To deploy deep CNNs on mobile devices, we present a simple and effective scheme to compress the entire CNN, which we call one-shot whole network compression. The proposed scheme consists of three steps: (1) rank selection with variational Bayesian matrix factorization, (2) Tucker decomposition on kernel tensor, and (3) fine-tuning to recover accumulated loss of accuracy, and each step can be easily implemented using publicly available tools. We demonstrate the effectiveness of the proposed scheme by testing the performance of various compressed CNNs (AlexNet, VGGS, GoogLeNet, and VGG-16) on the smartphone. Significant reductions in model size, runtime, and energy consumption are obtained, at the cost of small loss in accuracy. In addition, we address the important implementation level issue on 1?1 convolution, which is a key operation of inception module of GoogLeNet as well as CNNs compressed by our proposed scheme.",
"Convolutional neural networks excel in image recognition tasks, but this comes at the cost of high computational and memory complexity. To tackle this problem, [1] developed a tensor factorization framework to compress fully-connected layers. In this paper, we focus on compressing convolutional layers. We show that while the direct application of the tensor framework [1] to the 4-dimensional kernel of convolution does compress the layer, we can do better. We reshape the convolutional kernel into a tensor of higher order and factorize it. We combine the proposed approach with the previous work to compress both convolutional and fully-connected layers of a network and achieve 80x network compression rate with 1.1 accuracy drop on the CIFAR-10 dataset.",
"An individual differences model for multidimensional scaling is outlined in which individuals are assumed differentially to weight the several dimensions of a common “psychological space”. A corresponding method of analyzing similarities data is proposed, involving a generalization of “Eckart-Young analysis” to decomposition of three-way (or higher-way) tables. In the present case this decomposition is applied to a derived three-way table of scalar products between stimuli for individuals. This analysis yields a stimulus by dimensions coordinate matrix and a subjects by dimensions matrix of weights. This method is illustrated with data on auditory stimuli and on perception of nations."
]
} |
1905.10478 | 2946624195 | Tensor decomposition is an effective approach to compress over-parameterized neural networks and to enable their deployment on resource-constrained hardware platforms. However, directly applying tensor compression in the training process is a challenging task due to the difficulty of choosing a proper tensor rank. In order to achieve this goal, this paper proposes a Bayesian tensorized neural network. Our Bayesian method performs automatic model compression via an adaptive tensor rank determination. We also present approaches for posterior density calculation and maximum a posteriori (MAP) estimation for the end-to-end training of our tensorized neural network. We provide experimental validation on a fully connected neural network, a CNN and a residual neural network where our work produces @math to @math more compact neural networks directly from the training. | Tensor Rank Determination. Two main approaches are used for rank determination in tensor completion: low-rank optimization and Bayesian inference. Optimization methods mainly rely on some generalization of the matrix nuclear norm @cite_28 to tensors. The most popular approaches place a low-rank objective on tensor unfoldings @cite_11 @cite_27 @cite_9 , but the computation is expensive for high-order tensors. The Frobenius norms of CP factors are used as a regularization for 3-way tensor completion @cite_15 , but this does not generalize to high-order tensors either. Bayesian methods can directly infer the tensor rank in CP or Tucker tensor completion through low-rank priors @cite_25 @cite_2 @cite_23 @cite_15 @cite_13 . In the CP and Tucker formats, the ranks of different tensor factors do not couple with each other. This is not true in the tensor-train format (see ). Furthermore, in tensor completion the observed data is a linear mapping of a tensor, whereas the mapping is nonlinear in neural networks. Therefore, previous work on tensor rank determination cannot be directly applied to tensorized neural networks. | {
"cite_N": [
"@cite_28",
"@cite_9",
"@cite_27",
"@cite_23",
"@cite_2",
"@cite_15",
"@cite_13",
"@cite_25",
"@cite_11"
],
"mid": [
"2118550318",
"2962881496",
"2091449379",
"2963404198",
"2112292531",
"2082600181",
"2890009591",
"2147512299",
"2078677240"
],
"abstract": [
"The affine rank minimization problem consists of finding a matrix of minimum rank that satisfies a given system of linear equality constraints. Such problems have appeared in the literature of a diverse set of fields including system identification and control, Euclidean embedding, and collaborative filtering. Although specific instances can often be solved with specialized algorithms, the general affine rank minimization problem is NP-hard because it contains vector cardinality minimization as a special case. In this paper, we show that if a certain restricted isometry property holds for the linear transformation defining the constraints, the minimum-rank solution can be recovered by solving a convex optimization problem, namely, the minimization of the nuclear norm over the given affine space. We present several random ensembles of equations where the restricted isometry property holds with overwhelming probability, provided the codimension of the subspace is sufficiently large. The techniques used in our analysis have strong parallels in the compressed sensing framework. We discuss how affine rank minimization generalizes this preexisting concept and outline a dictionary relating concepts from cardinality minimization to those of rank minimization. We also discuss several algorithmic approaches to minimizing the nuclear norm and illustrate our results with numerical examples.",
"Tensor train (TT) decomposition provides a space-efficient representation for higher-order tensors. Despite its advantage, we face two crucial limitations when we apply the TT decomposition to machine learning problems: the lack of statistical theory and of scalable algorithms. In this paper, we address the limitations. First, we introduce a convex relaxation of the TT decomposition problem and derive its error bound for the tensor completion task. Next, we develop a randomized optimization method, in which the time complexity is as efficient as the space complexity is. In experiments, we numerically confirm the derived bounds and empirically demonstrate the performance of our method with a real higher-order tensor.",
"In this paper, we propose an algorithm to estimate missing values in tensors of visual data. The values can be missing due to problems in the acquisition process or because the user manually identified unwanted outliers. Our algorithm works even with a small amount of samples and it can propagate structure to fill larger missing regions. Our methodology is built on recent studies about matrix completion using the matrix trace norm. The contribution of our paper is to extend the matrix case to the tensor case by proposing the first definition of the trace norm for tensors and then by building a working algorithm. First, we propose a definition for the tensor trace norm that generalizes the established definition of the matrix trace norm. Second, similarly to matrix completion, the tensor completion is formulated as a convex optimization problem. Unfortunately, the straightforward problem extension is significantly harder to solve than the matrix case because of the dependency among multiple constraints. To tackle this problem, we developed three algorithms: simple low rank tensor completion (SiLRTC), fast low rank tensor completion (FaLRTC), and high accuracy low rank tensor completion (HaLRTC). The SiLRTC algorithm is simple to implement and employs a relaxation technique to separate the dependant relationships and uses the block coordinate descent (BCD) method to achieve a globally optimal solution; the FaLRTC algorithm utilizes a smoothing scheme to transform the original nonsmooth problem into a smooth one and can be used to solve a general tensor trace norm minimization problem; the HaLRTC algorithm applies the alternating direction method of multipliers (ADMMs) to our problem. Our experiments show potential applications of our algorithms and the quantitative evaluation indicates that our methods are more accurate and robust than heuristic approaches. The efficiency comparison indicates that FaLTRC and HaLRTC are more efficient than SiLRTC and between FaLRTC and HaLRTC the former is more efficient to obtain a low accuracy solution and the latter is preferred if a high-accuracy solution is desired.",
"",
"We present a scalable Bayesian framework for low-rank decomposition of multiway tensor data with missing observations. The key issue of pre-specifying the rank of the decomposition is sidestepped in a principled manner using a multiplicative gamma process prior. Both continuous and binary data can be analyzed under the framework, in a coherent way using fully conjugate Bayesian analysis. In particular, the analysis in the non-conjugate binary case is facilitated via the use of the Polya-Gamma sampling strategy which elicits closed-form Gibbs sampling updates. The resulting samplers are efficient and enable us to apply our framework to large-scale problems, with time-complexity that is linear in the number of observed entries in the tensor. This is especially attractive in analyzing very large but sparsely observed tensors with very few known entries. Moreover, our method admits easy extension to the supervised setting where entities in one or more tensor modes have labels. Our method outperforms several state-of-the-art tensor decomposition methods on various synthetic and benchmark real-world datasets.",
"A novel regularizer of the PARAFAC decomposition factors capturing the tensor's rank is proposed in this paper, as the key enabler for completion of three-way data arrays with missing entries. Set in a Bayesian framework, the tensor completion method incorporates prior information to enhance its smoothing and prediction capabilities. This probabilistic approach can naturally accommodate general models for the data distribution, lending itself to various fitting criteria that yield optimum estimates in the maximum-a-posteriori sense. In particular, two algorithms are devised for Gaussian- and Poisson-distributed data, that minimize the rank-regularized least-squares error and Kullback-Leibler divergence, respectively. The proposed technique is able to recover the “ground-truth” tensor rank when tested on synthetic data, and to complete brain imaging and yeast gene expression datasets with 50 and 15 of missing entries respectively, resulting in recovery errors at -11 dB and -15 dB.",
"Streaming tensor factorization is a powerful tool for processing high-volume and multi-way temporal data in Internet networks, recommender systems and image video data analysis. In many applications the full tensor is not known, but instead received in a slice-by-slice manner over time. Streaming factorizations aim to take advantage of inherent temporal relationships in data analytics. Existing streaming tensor factorization algorithms rely on least-squares data fitting and they do not possess a mechanism for tensor rank determination. This leaves them susceptible to outliers and vulnerable to over-fitting. This paper presents the first Bayesian robust streaming tensor factorization model. Our model successfully identifies sparse outliers, automatically determines the underlying tensor rank and accurately fits low-rank structure. We implement our model in Matlab and compare it to existing algorithms. Our algorithm is applied to factorize and complete various streaming tensors including synthetic data, dynamic MRI, video sequences, and Internet traffic data.",
"CANDECOMP PARAFAC (CP) tensor factorization of incomplete data is a powerful technique for tensor completion through explicitly capturing the multilinear latent factors. The existing CP algorithms require the tensor rank to be manually specified, however, the determination of tensor rank remains a challenging problem especially for CP rank . In addition, existing approaches do not take into account uncertainty information of latent factors, as well as missing entries. To address these issues, we formulate CP factorization using a hierarchical probabilistic model and employ a fully Bayesian treatment by incorporating a sparsity-inducing prior over multiple latent factors and the appropriate hyperpriors over all hyperparameters, resulting in automatic rank determination. To learn the model, we develop an efficient deterministic Bayesian inference algorithm, which scales linearly with data size. Our method is characterized as a tuning parameter-free approach, which can effectively infer underlying multilinear factors with a low-rank constraint, while also providing predictive distributions over missing entries. Extensive simulations on synthetic data illustrate the intrinsic capability of our method to recover the ground-truth of CP rank and prevent the overfitting problem, even when a large amount of entries are missing. Moreover, the results from real-world applications, including image inpainting and facial image synthesis, demonstrate that our method outperforms state-of-the-art approaches for both tensor factorization and tensor completion in terms of predictive performance.",
"In this paper we consider sparsity on a tensor level, as given by the n-rank of a tensor. In an important sparse-vector approximation problem (compressed sensing) and the low-rank matrix recovery problem, using a convex relaxation technique proved to be a valuable solution strategy. Here, we will adapt these techniques to the tensor setting. We use the n-rank of a tensor as a sparsity measure and consider the low-n-rank tensor recovery problem, i.e. the problem of finding the tensor of the lowest n-rank that fulfills some linear constraints. We introduce a tractable convex relaxation of the n-rank and propose efficient algorithms to solve the low-n-rank tensor recovery problem numerically. The algorithms are based on the Douglas–Rachford splitting technique and its dual variant, the alternating direction method of multipliers."
]
} |
1905.10478 | 2946624195 | Tensor decomposition is an effective approach to compress over-parameterized neural networks and to enable their deployment on resource-constrained hardware platforms. However, directly applying tensor compression in the training process is a challenging task due to the difficulty of choosing a proper tensor rank. In order to achieve this goal, this paper proposes a Bayesian tensorized neural network. Our Bayesian method performs automatic model compression via an adaptive tensor rank determination. We also present approaches for posterior density calculation and maximum a posteriori (MAP) estimation for the end-to-end training of our tensorized neural network. We provide experimental validation on a fully connected neural network, a CNN and a residual neural network where our work produces @math to @math more compact neural networks directly from the training. | Bayesian Neural Networks. Bayesian neural networks were developed to quantify the model uncertainty @cite_7 . Both and explored sparsity-inducing shrinkage priors for relevance determination and weight removal in neural networks. Another popular prior for Bayesian neural network compression is the Minimum Description Length (MDL) framework for quantization @cite_1 . MDL has seen modern implementations by and , who also investigated Bayesian pruning. To our best knowledge, no Bayesian approaches have been reported to train a low-rank tensorized neural network. A main challenge of training a Bayesian neural network is the high computational cost caused by sampling-based posterior density estimations. We will address this issue by the recently developed Stein variational gradient descent @cite_31 . | {
"cite_N": [
"@cite_31",
"@cite_1",
"@cite_7"
],
"mid": [
"2963956018",
"2047229728",
"1567512734"
],
"abstract": [
"We propose a general purpose variational inference algorithm that forms a natural counterpart of gradient descent for optimization. Our method iteratively transports a set of particles to match the target distribution, by applying a form of functional gradient descent that minimizes the KL divergence. Empirical studies are performed on various real world models and datasets, on which our method is competitive with existing state-of-the-art methods. The derivation of our method is based on a new theoretical result that connects the derivative of KL divergence under smooth transforms with Stein’s identity and a recently proposed kernelized Stein discrepancy, which is of independent interest.",
"",
"From the Publisher: Artificial \"neural networks\" are now widely used as flexible models for regression classification applications, but questions remain regarding what these models mean, and how they can safely be used when training data is limited. Bayesian Learning for Neural Networks shows that Bayesian methods allow complex neural network models to be used without fear of the \"overfitting\" that can occur with traditional neural network learning methods. Insight into the nature of these complex Bayesian models is provided by a theoretical investigation of the priors over functions that underlie them. Use of these models in practice is made possible using Markov chain Monte Carlo techniques. Both the theoretical and computational aspects of this work are of wider statistical interest, as they contribute to a better understanding of how Bayesian methods can be applied to complex problems. Presupposing only the basic knowledge of probability and statistics, this book should be of interest to many researchers in statistics, engineering, and artificial intelligence. Software for Unix systems that implements the methods described is freely available over the Internet."
]
} |
1905.10705 | 2944908540 | Ability to quantify and predict progression of a disease is fundamental for selecting an appropriate treatment. Many clinical metrics cannot be acquired frequently either because of their cost (e.g. MRI, gait analysis) or because they are inconvenient or harmful to a patient (e.g. biopsy, x-ray). In such scenarios, in order to estimate individual trajectories of disease progression, it is advantageous to leverage similarities between patients, i.e. the covariance of trajectories, and find a latent representation of progression. Most of existing methods for estimating trajectories do not account for events in-between observations, what dramatically decreases their adequacy for clinical practice. In this study, we develop a machine learning framework named Coordinatewise-Soft-Impute (CSI) for analyzing disease progression from sparse observations in the presence of confounding events. CSI is guaranteed to converge to the global minimum of the corresponding optimization problem. Experimental results also demonstrates the effectiveness of CSI using both simulated and real dataset. | In mixed-effect models, every trajectory @math is assumed to be composed of two parts: the @math for some @math that remains the same among all patients and a @math that differs for each @math . In its simplest form, we assume where @math is the @math covariance matrix, @math is the standard deviation and @math , @math are functions @math and @math evaluated at the times @math , respectively. Estimations of model parameters @math can be made via expectation maximization (EM) algorithm @cite_1 . Individual coefficients @math can be estimated using the best unbiased linear predictor (BLUP) @cite_4 . | {
"cite_N": [
"@cite_1",
"@cite_4"
],
"mid": [
"2082246284",
"2013866489"
],
"abstract": [
"SUMMARY Models for the analysis of longitudinal data must recogrlize the relationship between serial observations on the same unit. Multivariate models with general covariance structure are often difficult to apply to highly unbalanced data, whereas two-stage random-effects models can be used easily. In two-stage models, the probability distributions for the response vectors of different individuals belong to a single family, but some random-effects parameters vary across individuals, with a distribution specified at the second stage. A general family of models is discussed, which includes both growth models and repeated-measures models as special cases. A unified approach to fitting these models, based on a combination of empirical Bayes and maximum likelihood estimation of model parameters and using the EM algorithm, is discussed. Two examples are taken from a current epidemiological study of the health effects of air pollution. Many longitudinal studies are designed to investigate changes over time in a characteristic which is measured repeatedly for each study participant. In medical studies, the measurement might be blood pressure, cholesterol level, lung volume, or serum glucose. Multiple measurements are obtained from each individual, at different times and possibly under changing experimental conditions. Often, we cannot fully control the circumstances under which the measurements are taken, and there may be considerable variation among individuals in the number and timing of observations. The resulting unbalanced data sets are typically not amenable to analysis using a general multivariate model with unrestricted covariance structure. Statisticians have often analyzed data of this form using some variant of a two-stage model. In this formulation, the probability distribution for the multiple measurements has the same form for each individual, but the parameters of that distribution vary over individuals. The distribution of these parameters, or 'random effects', in the population constitutes the second stage of the model. In a study of changes in lung volume during childhood, for instance, it may be reasonable to assume that the relationship between lung volume and the cube of height is linear for each child, but with linear regression parameters that vary among children. If we assume that the usual linear regression model applies for each child, conditional on the child's individual parameters, and that the regression parameters have a bivariate normal distribution in the population, the marginal distribution of the serial measurements is multivariate normal with a special covariance structure.",
"Mixed linear models are assumed in most animal breeding applications. Convenient methods for computing BLUE of the estimable linear functions of the fixed elements of the model and for computing best linear unbiased predictions of the random elements of the model have been available. Most data available to animal breeders, however, do not meet the usual requirements of random sampling, the problem being that the data arise either from selection experiments or from breeders' herds which are undergoing selection. Consequently, the usual methods are likely to yield biased estimates and predictions. Methods for dealing with such data are presented in this paper."
]
} |
1905.10705 | 2944908540 | Ability to quantify and predict progression of a disease is fundamental for selecting an appropriate treatment. Many clinical metrics cannot be acquired frequently either because of their cost (e.g. MRI, gait analysis) or because they are inconvenient or harmful to a patient (e.g. biopsy, x-ray). In such scenarios, in order to estimate individual trajectories of disease progression, it is advantageous to leverage similarities between patients, i.e. the covariance of trajectories, and find a latent representation of progression. Most of existing methods for estimating trajectories do not account for events in-between observations, what dramatically decreases their adequacy for clinical practice. In this study, we develop a machine learning framework named Coordinatewise-Soft-Impute (CSI) for analyzing disease progression from sparse observations in the presence of confounding events. CSI is guaranteed to converge to the global minimum of the corresponding optimization problem. Experimental results also demonstrates the effectiveness of CSI using both simulated and real dataset. | In linear mixed-effect model, each trajectory is estimated with @math degrees of freedom, which can still be too complex when observations are sparse. One typical solution is to assume a low-rank structure of the covariance matrix @math by introducing a contraction mapping @math from the functional basis to a low-dimensional latent space. More specifically, one may rewrite the LMM model as where @math is a @math matrix with @math and @math is the new, shorter random effect to be estimated. Methods based on low-rank approximations are widely adopted and applied in practice and different algorithms on fitting the model have been proposed @cite_2 @cite_5 @cite_6 . In the later sections, we will compare our algorithm with one specific implementation named functional-Principle-Component-Analysis (fPCA) @cite_2 , which uses EM algorithm for estimating model parameters and latent variables @math . | {
"cite_N": [
"@cite_5",
"@cite_6",
"@cite_2"
],
"mid": [
"",
"2962732961",
"2091083714"
],
"abstract": [
"",
"Medical researchers are coming to appreciate that many diseases are in fact complex, heterogeneous syndromes composed of subpopulations that express different variants of a related complication. Longitudinal data extracted from individual electronic health records (EHR) offer an exciting new way to study subtle differences in the way these diseases progress over time. In this paper, we focus on answering two questions that can be asked using these databases of longitudinal EHR data. First, we want to understand whether there are individuals with similar disease trajectories and whether there are a small number of degrees of freedom that account for differences in trajectories across the population. Second, we want to understand how important clinical outcomes are associated with disease trajectories. To answer these questions, we propose the Disease Trajectory Map (DTM), a novel probabilistic model that learns low-dimensional representations of sparse and irregularly sampled longitudinal data. We propose a stochastic variational inference algorithm for learning the DTM that allows the model to scale to large modern medical datasets. To demonstrate the DTM, we analyze data collected on patients with the complex autoimmune disease, scleroderma. We find that DTM learns meaningful representations of disease trajectories and that the representations are significantly associated with important clinical outcomes.",
"SUMMARY The elements of a multivariate data set are often curves rather than single points. Functional principal components can be used to describe the modes of variation of such curves. If one has complete measurements for each individual curve or, as is more common, one has measurements on a fine grid taken at the same time points for all curves, then many standard techniques may be applied. However, curves are often measured at an irregular and sparse set of time points which can differ widely across individuals. We present a technique for handling this more difficult case using a reduced rank mixed effects framework."
]
} |
1905.10705 | 2944908540 | Ability to quantify and predict progression of a disease is fundamental for selecting an appropriate treatment. Many clinical metrics cannot be acquired frequently either because of their cost (e.g. MRI, gait analysis) or because they are inconvenient or harmful to a patient (e.g. biopsy, x-ray). In such scenarios, in order to estimate individual trajectories of disease progression, it is advantageous to leverage similarities between patients, i.e. the covariance of trajectories, and find a latent representation of progression. Most of existing methods for estimating trajectories do not account for events in-between observations, what dramatically decreases their adequacy for clinical practice. In this study, we develop a machine learning framework named Coordinatewise-Soft-Impute (CSI) for analyzing disease progression from sparse observations in the presence of confounding events. CSI is guaranteed to converge to the global minimum of the corresponding optimization problem. Experimental results also demonstrates the effectiveness of CSI using both simulated and real dataset. | While the probabilistic approach of mixed-effect models offers many theoretical advantages including convergence rates and inference testing, it is often sensitive to the assumptions on distributions, some of which are hard to verify in practice. To avoid the potential bias of distributional assumptions in mixed-effect models, in @cite_8 formulate the problem as a sparse matrix completion problem. We will review this approach in the current section. | {
"cite_N": [
"@cite_8"
],
"mid": [
"2891004699"
],
"abstract": [
"In clinical practice and biomedical research, measurements are often collected sparsely and irregularly in time while the data acquisition is expensive and inconvenient. Examples include measurements of spine bone mineral density, cancer growth through mammography or biopsy, a progression of defect of vision, or assessment of gait in patients with neurological disorders. Since the data collection is often costly and inconvenient, estimation of progression from sparse observations is of great interest for practitioners. From the statistical standpoint, such data is often analyzed in the context of a mixed-effect model where time is treated as both random and fixed effect. Alternatively, researchers analyze Gaussian processes or functional data where observations are assumed to be drawn from a certain distribution of processes. These models are flexible but rely on probabilistic assumptions and require very careful implementation. In this study, we propose an alternative elementary framework for analyzing longitudinal data, relying on matrix completion. Our method yields point estimates of progression curves by iterative application of the SVD. Our framework covers multivariate longitudinal data, regression and can be easily extended to other settings. We apply our methods to understand trends of progression of motor impairment in children with Cerebral Palsy. Our model approximates individual progression curves and explains 30 of the variability. Low-rank representation of progression trends enables discovering that subtypes of Cerebral Palsy exhibit different progression trends."
]
} |
1905.10705 | 2944908540 | Ability to quantify and predict progression of a disease is fundamental for selecting an appropriate treatment. Many clinical metrics cannot be acquired frequently either because of their cost (e.g. MRI, gait analysis) or because they are inconvenient or harmful to a patient (e.g. biopsy, x-ray). In such scenarios, in order to estimate individual trajectories of disease progression, it is advantageous to leverage similarities between patients, i.e. the covariance of trajectories, and find a latent representation of progression. Most of existing methods for estimating trajectories do not account for events in-between observations, what dramatically decreases their adequacy for clinical practice. In this study, we develop a machine learning framework named Coordinatewise-Soft-Impute (CSI) for analyzing disease progression from sparse observations in the presence of confounding events. CSI is guaranteed to converge to the global minimum of the corresponding optimization problem. Experimental results also demonstrates the effectiveness of CSI using both simulated and real dataset. | The direct way of estimating @math is to solve the optimization problem where @math is the Fr "obenius norm. Again, if @math is larger than the number of observations for some subject we will overfit. To avoid this problem we need some additional constraints on @math . A typical approach in the matrix completion community is to introduce a nuclear norm penalty---a relaxed version of the rank penalty while preserving convexity @cite_11 @cite_12 . The optimization problem with the nuclear norm penalty takes form where @math is the regularization parameter, @math is the Fr "obenius norm, and @math is the nuclear norm, i.e. the sum of singular values. In @cite_8 , a Soft-Longitudinal-Impute (SLI) algorithm is proposed to solve efficiently. For completeness of the paper, we include the SLI algorithm in Appendix , while noting that it is also a special case of our algorithm defined in the next section with @math fixed to be @math . | {
"cite_N": [
"@cite_8",
"@cite_12",
"@cite_11"
],
"mid": [
"2891004699",
"",
"1976618413"
],
"abstract": [
"In clinical practice and biomedical research, measurements are often collected sparsely and irregularly in time while the data acquisition is expensive and inconvenient. Examples include measurements of spine bone mineral density, cancer growth through mammography or biopsy, a progression of defect of vision, or assessment of gait in patients with neurological disorders. Since the data collection is often costly and inconvenient, estimation of progression from sparse observations is of great interest for practitioners. From the statistical standpoint, such data is often analyzed in the context of a mixed-effect model where time is treated as both random and fixed effect. Alternatively, researchers analyze Gaussian processes or functional data where observations are assumed to be drawn from a certain distribution of processes. These models are flexible but rely on probabilistic assumptions and require very careful implementation. In this study, we propose an alternative elementary framework for analyzing longitudinal data, relying on matrix completion. Our method yields point estimates of progression curves by iterative application of the SVD. Our framework covers multivariate longitudinal data, regression and can be easily extended to other settings. We apply our methods to understand trends of progression of motor impairment in children with Cerebral Palsy. Our model approximates individual progression curves and explains 30 of the variability. Low-rank representation of progression trends enables discovering that subtypes of Cerebral Palsy exhibit different progression trends.",
"",
"Maximum Margin Matrix Factorization (MMMF) was recently suggested (, 2005) as a convex, infinite dimensional alternative to low-rank approximations and standard factor models. MMMF can be formulated as a semi-definite programming (SDP) and learned using standard SDP solvers. However, current SDP solvers can only handle MMMF problems on matrices of dimensionality up to a few hundred. Here, we investigate a direct gradient-based optimization method for MMMF and demonstrate it on large collaborative prediction problems. We compare against results obtained by Marlin (2004) and find that MMMF substantially outperforms all nine methods he tested."
]
} |
1905.10622 | 2946281912 | Images with visual and scene text content are ubiquitous in everyday life. However current image interpretation systems are mostly limited to using only the visual features, neglecting to leverage the scene text content. In this paper we propose to jointly use scene text and visual channels for robust semantic interpretation of images. We undertake the task of matching Advertisement images against their human generated statements that describe the action that the ad prompts and the rationale it provides for taking this action. We extract the scene text and generate semantic and lexical text representations, which are used in the interpretation of the Ad Image. To deal with irrelevant or erroneous detection of scene text, we use a text attention scheme. We also learn an embedding of the visual channel, visual features based on detected symbolism and objects, into a semantic embedding space, leveraging text semantics obtained from scene text. We show how the multi channel approach, involving visual semantics and scene text, improves upon the current state of the art. | As language and vision are the two most important aspects of human communication, one of the most natural questions for researchers working in artificial intelligence is how to map vision to its corresponding text and vice versa. This gives rise to problems like image captioning @cite_20 , text based image retrieval (e.g. google image search), visual question answering https: visualqa.org etc. | {
"cite_N": [
"@cite_20"
],
"mid": [
"2803259101"
],
"abstract": [
"Abstract Image captioning means automatically generating a caption for an image. As a recently emerged research area, it is attracting more and more attention. To achieve the goal of image captioning, semantic information of images needs to be captured and expressed in natural languages. Connecting both research communities of computer vision and natural language processing, image captioning is a quite challenging task. Various approaches have been proposed to solve this problem. In this paper, we present a survey on advances in image captioning research. Based on the technique adopted, we classify image captioning approaches into different categories. Representative methods in each category are summarized, and their strengths and limitations are talked about. In this paper, we first discuss methods used in early work which are mainly retrieval and template based. Then, we focus our main attention on neural network based methods, which give state of the art results. Neural network based methods are further divided into subcategories based on the specific framework they use. Each subcategory of neural network based methods are discussed in detail. After that, state of the art methods are compared on benchmark datasets. Following that, discussions on future research directions are presented."
]
} |
1905.10622 | 2946281912 | Images with visual and scene text content are ubiquitous in everyday life. However current image interpretation systems are mostly limited to using only the visual features, neglecting to leverage the scene text content. In this paper we propose to jointly use scene text and visual channels for robust semantic interpretation of images. We undertake the task of matching Advertisement images against their human generated statements that describe the action that the ad prompts and the rationale it provides for taking this action. We extract the scene text and generate semantic and lexical text representations, which are used in the interpretation of the Ad Image. To deal with irrelevant or erroneous detection of scene text, we use a text attention scheme. We also learn an embedding of the visual channel, visual features based on detected symbolism and objects, into a semantic embedding space, leveraging text semantics obtained from scene text. We show how the multi channel approach, involving visual semantics and scene text, improves upon the current state of the art. | In @cite_24 the authors leverage textual data to establish semantic relationships between visual object category labels, thereby mapping images into a semantic embedding space. Following these lines, a number of recent works, leverage the word2vec semantic vector space @cite_15 @cite_17 , to incorporate semantics into their representation. | {
"cite_N": [
"@cite_24",
"@cite_15",
"@cite_17"
],
"mid": [
"2123024445",
"1614298861",
"2131744502"
],
"abstract": [
"Modern visual recognition systems are often limited in their ability to scale to large numbers of object categories. This limitation is in part due to the increasing difficulty of acquiring sufficient training data in the form of labeled images as the number of object categories grows. One remedy is to leverage data from other sources - such as text data - both to train visual models and to constrain their predictions. In this paper we present a new deep visual-semantic embedding model trained to identify visual objects using both labeled image data as well as semantic information gleaned from unannotated text. We demonstrate that this model matches state-of-the-art performance on the 1000-class ImageNet object recognition challenge while making more semantically reasonable errors, and also show that the semantic information can be exploited to make predictions about tens of thousands of image labels not observed during training. Semantic knowledge improves such zero-shot predictions achieving hit rates of up to 18 across thousands of novel labels never seen by the visual model.",
"",
"Many machine learning algorithms require the input to be represented as a fixed-length feature vector. When it comes to texts, one of the most common fixed-length features is bag-of-words. Despite their popularity, bag-of-words features have two major weaknesses: they lose the ordering of the words and they also ignore semantics of the words. For example, \"powerful,\" \"strong\" and \"Paris\" are equally distant. In this paper, we propose Paragraph Vector, an unsupervised algorithm that learns fixed-length feature representations from variable-length pieces of texts, such as sentences, paragraphs, and documents. Our algorithm represents each document by a dense vector which is trained to predict words in the document. Its construction gives our algorithm the potential to overcome the weaknesses of bag-of-words models. Empirical results show that Paragraph Vectors outperforms bag-of-words models as well as other techniques for text representations. Finally, we achieve new state-of-the-art results on several text classification and sentiment analysis tasks."
]
} |
1905.10622 | 2946281912 | Images with visual and scene text content are ubiquitous in everyday life. However current image interpretation systems are mostly limited to using only the visual features, neglecting to leverage the scene text content. In this paper we propose to jointly use scene text and visual channels for robust semantic interpretation of images. We undertake the task of matching Advertisement images against their human generated statements that describe the action that the ad prompts and the rationale it provides for taking this action. We extract the scene text and generate semantic and lexical text representations, which are used in the interpretation of the Ad Image. To deal with irrelevant or erroneous detection of scene text, we use a text attention scheme. We also learn an embedding of the visual channel, visual features based on detected symbolism and objects, into a semantic embedding space, leveraging text semantics obtained from scene text. We show how the multi channel approach, involving visual semantics and scene text, improves upon the current state of the art. | However, the use of text semantics is not only limited to such distributional representations, and semantics has also been inferred from hierarchically structured lexical database, like wordnet as in @cite_25 | {
"cite_N": [
"@cite_25"
],
"mid": [
"2953384527"
],
"abstract": [
"The goal of this work is to bring semantics into the tasks of text recognition and retrieval in natural images. Although text recognition and retrieval have received a lot of attention in recent years, previous works have focused on recognizing or retrieving exactly the same word used as a query, without taking the semantics into consideration. In this paper, we ask the following question: For this goal we propose a convolutional neural network (CNN) with a weighted ranking loss objective that ensures that the concepts relevant to the query image are ranked ahead of those that are not relevant. This can also be interpreted as learning a Euclidean space where word images and concepts are jointly embedded. This model is learned in an end-to-end manner, from image pixels to semantic concepts, using a dataset of synthetically generated word images and concepts mined from a lexical database (WordNet). Our results show that, despite the complexity of the task, word images and concepts can indeed be associated with a high degree of accuracy"
]
} |
1905.10622 | 2946281912 | Images with visual and scene text content are ubiquitous in everyday life. However current image interpretation systems are mostly limited to using only the visual features, neglecting to leverage the scene text content. In this paper we propose to jointly use scene text and visual channels for robust semantic interpretation of images. We undertake the task of matching Advertisement images against their human generated statements that describe the action that the ad prompts and the rationale it provides for taking this action. We extract the scene text and generate semantic and lexical text representations, which are used in the interpretation of the Ad Image. To deal with irrelevant or erroneous detection of scene text, we use a text attention scheme. We also learn an embedding of the visual channel, visual features based on detected symbolism and objects, into a semantic embedding space, leveraging text semantics obtained from scene text. We show how the multi channel approach, involving visual semantics and scene text, improves upon the current state of the art. | The ad image dataset presented in @cite_26 proposed multiple task for automated understanding and classification of ad images. In an extension of that work, @cite_4 proposed the task of matching ad images against relevant sentences (see Fig. for an example of this task). Similar to our approach, they embed the images in a joint text-image embedding space, where retrieval is feasible. The embedding scheme consists of features from salient regions proposed by symbol detectors and automated captions generated by Densecap @cite_34 . The text captions, treated as external knowledge, are projected in to word2vec space, and the resultant aggregated semantic vector is augmented with the visual features. Though generated caption gives the image a semantic description, which can be easily compared with the statements for retrieval, this scheme needs externally trained model for generating captions. | {
"cite_N": [
"@cite_26",
"@cite_4",
"@cite_34"
],
"mid": [
"2963037330",
"2769093093",
"2963758027"
],
"abstract": [
"There is more to images than their objective physical content: for example, advertisements are created to persuade a viewer to take a certain action. We propose the novel problem of automatic advertisement understanding. To enable research on this problem, we create two datasets: an image dataset of 64,832 image ads, and a video dataset of 3,477 ads. Our data contains rich annotations encompassing the topic and sentiment of the ads, questions and answers describing what actions the viewer is prompted to take and the reasoning that the ad presents to persuade the viewer (What should I do according to this ad, and why should I do it?), and symbolic references ads make (e.g. a dove symbolizes peace). We also analyze the most common persuasive strategies ads use, and the capabilities that computer vision systems should have to understand these strategies. We present baseline classification results for several prediction tasks, including automatically answering questions about the messages of the ads.",
"In order to convey the most content in their limited space, advertisements embed references to outside knowledge via symbolism. For example, a motorcycle stands for adventure (a positive property the ad wants associated with the product being sold), and a gun stands for danger (a negative property to dissuade viewers from undesirable behaviors). We show how to use symbolic references to better understand the meaning of an ad. We further show how anchoring ad understanding in general-purpose object recognition and image captioning can further improve results. We formulate the ad understanding task as matching the ad image to human-generated statements that describe the action that the ad prompts and the rationale it provides for this action. We greatly outperform the state of the art in this task. We also show additional applications of our learned representations for ranking the slogans of ads, and clustering ads according to their topic.",
"We introduce the dense captioning task, which requires a computer vision system to both localize and describe salient regions in images in natural language. The dense captioning task generalizes object detection when the descriptions consist of a single word, and Image Captioning when one predicted region covers the full image. To address the localization and description task jointly we propose a Fully Convolutional Localization Network (FCLN) architecture that processes an image with a single, efficient forward pass, requires no external regions proposals, and can be trained end-to-end with a single round of optimization. The architecture is composed of a Convolutional Network, a novel dense localization layer, and Recurrent Neural Network language model that generates the label sequences. We evaluate our network on the Visual Genome dataset, which comprises 94,000 images and 4,100,000 region-grounded captions. We observe both speed and accuracy improvements over baselines based on current state of the art approaches in both generation and retrieval settings."
]
} |
1905.10622 | 2946281912 | Images with visual and scene text content are ubiquitous in everyday life. However current image interpretation systems are mostly limited to using only the visual features, neglecting to leverage the scene text content. In this paper we propose to jointly use scene text and visual channels for robust semantic interpretation of images. We undertake the task of matching Advertisement images against their human generated statements that describe the action that the ad prompts and the rationale it provides for taking this action. We extract the scene text and generate semantic and lexical text representations, which are used in the interpretation of the Ad Image. To deal with irrelevant or erroneous detection of scene text, we use a text attention scheme. We also learn an embedding of the visual channel, visual features based on detected symbolism and objects, into a semantic embedding space, leveraging text semantics obtained from scene text. We show how the multi channel approach, involving visual semantics and scene text, improves upon the current state of the art. | While scene text has never been exploited in the context of advertisement images, it has been used by some researchers in other image understanding tasks. While the community has witnessed lots of interplay between text and image in general @cite_10 @cite_6 they are mainly to model parallel text @cite_33 present in the form of metadata, tags or annotation and not scene text contained in the image itself. While images with scene-text content are ubiquitous in our everyday life, their use has been limited to fine grained image classification @cite_27 @cite_21 @cite_9 @cite_32 . | {
"cite_N": [
"@cite_33",
"@cite_9",
"@cite_21",
"@cite_32",
"@cite_6",
"@cite_27",
"@cite_10"
],
"mid": [
"2884528675",
"2565591417",
"2619328388",
"2046519698",
"2963383024",
"2613009185",
"2963717374"
],
"abstract": [
"Images and text in advertisements interact in complex, non-literal ways. The two channels are usually complementary, with each channel telling a different part of the story. Current approaches, such as image captioning methods, only examine literal, redundant relationships, where image and text show exactly the same content. To understand more complex relationships, we first collect a dataset of advertisement interpretations for whether the image and slogan in the same visual advertisement form a parallel (conveying the same message without literally saying the same thing) or non-parallel relationship, with the help of workers recruited on Amazon Mechanical Turk. We develop a variety of features that capture the creativity of images and the specificity or ambiguity of text, as well as methods that analyze the semantics within and across channels. We show that our method outperforms standard image-text alignment approaches on predicting the parallel non-parallel relationship between image and text.",
"Text in natural images typically adds meaning to an object or scene. In particular, text specifies which business places serve drinks (e.g., cafe, teahouse) or food (e.g., restaurant, pizzeria), and what kind of service is provided (e.g., massage, repair). The mere presence of text, its words, and meaning are closely related to the semantics of the object or scene. This paper exploits textual contents in images for fine-grained business place classification and logo retrieval. There are four main contributions. First, we show that the textual cues extracted by the proposed method are effective for the two tasks. Combining the proposed textual and visual cues outperforms visual only classification and retrieval by a large margin. Second, to extract the textual cues, a generic and fully unsupervised word box proposal method is introduced. The method reaches state-of-the-art word detection recall with a limited number of proposals. Third, contrary to what is widely acknowledged in text detection literature, we demonstrate that high recall in word detection is more important than high f-score at least for both tasks considered in this work. Last, this paper provides a large annotated text detection dataset with 10 K images and 27 601 word boxes.",
"This paper focuses on fine-grained object classification using recognized scene text in natural images. While the state-of-the-art relies on visual cues only, this paper is the first work which proposes to combine textual and visual cues. Another novelty is the textual cue extraction. Unlike the state-of-the-art text detection methods, we focus more on the background instead of text regions. Once text regions are detected, they are further processed by two methods to perform text recognition, i.e., ABBYY commercial OCR engine and a state-of-the-art character recognition algorithm. Then, to perform textual cue encoding, bi- and trigrams are formed between the recognized characters by considering the proposed spatial pairwise constraints. Finally, extracted visual and textual cues are combined for fine-grained classification. The proposed method is validated on four publicly available data sets: ICDAR03, ICDAR13, Con-Text , and Flickr-logo . We improve the state-of-the-art end-to-end character recognition by a large margin of 15 on ICDAR03. We show that textual cues are useful in addition to visual cues for fine-grained classification. We show that textual cues are also useful for logo retrieval. Adding textual cues outperforms visual- and textual-only in fine-grained classification (70.7 to 60.3 ) and logo retrieval (57.4 to 54.8 ).",
"This paper focuses on fine-grained classification by detecting photographed text in images. We introduce a text detection method that does not try to detect all possible foreground text regions but instead aims to reconstruct the scene background to eliminate non-text regions. Object cues such as color, contrast, and objectiveness are used in corporation with a random forest classifier to detect background pixels in the scene. Results on two publicly available datasets ICDAR03 and a fine-grained Building subcategories of ImageNet shows the effectiveness of the proposed method.",
"",
"",
"Bilinear models provide an appealing framework for mixing and merging information in Visual Question Answering (VQA) tasks. They help to learn high level associations between question meaning and visual concepts in the image, but they suffer from huge dimensionality issues.,,We introduce MUTAN, a multimodal tensor-based Tucker decomposition to efficiently parametrize bilinear interactions between visual and textual representations. Additionally to the Tucker framework, we design a low-rank matrix-based decomposition to explicitly constrain the interaction rank. With MUTAN, we control the complexity of the merging scheme while keeping nice interpretable fusion relations. We show how the Tucker decomposition framework generalizes some of the latest VQA architectures, providing state-of-the-art results."
]
} |
1905.10622 | 2946281912 | Images with visual and scene text content are ubiquitous in everyday life. However current image interpretation systems are mostly limited to using only the visual features, neglecting to leverage the scene text content. In this paper we propose to jointly use scene text and visual channels for robust semantic interpretation of images. We undertake the task of matching Advertisement images against their human generated statements that describe the action that the ad prompts and the rationale it provides for taking this action. We extract the scene text and generate semantic and lexical text representations, which are used in the interpretation of the Ad Image. To deal with irrelevant or erroneous detection of scene text, we use a text attention scheme. We also learn an embedding of the visual channel, visual features based on detected symbolism and objects, into a semantic embedding space, leveraging text semantics obtained from scene text. We show how the multi channel approach, involving visual semantics and scene text, improves upon the current state of the art. | In @cite_21 @cite_9 , the authors explored the idea of fine grained classification of business places @cite_32 by leveraging scene text. In both the instances the visual cues were a combination of BOW and deep features, but they differed in their encoding of textual cues. In @cite_21 textual cues were encoded by BOW model using n-gram as vocabulary, whereas in @cite_9 uses a vocabulary from the words present in the dataset to extract BOW vectors. While these schemes resulted in better performance than what was possible with only image features, the encoding of the textual feature did not involve leveraging any semantics from the text domain. In @cite_27 the authors proposed a similar fine-grained classification task on the same dataset @cite_32 , where they encoded the text by averaging their word2vec projections and obtain better results. | {
"cite_N": [
"@cite_27",
"@cite_9",
"@cite_21",
"@cite_32"
],
"mid": [
"2613009185",
"2565591417",
"2619328388",
"2046519698"
],
"abstract": [
"",
"Text in natural images typically adds meaning to an object or scene. In particular, text specifies which business places serve drinks (e.g., cafe, teahouse) or food (e.g., restaurant, pizzeria), and what kind of service is provided (e.g., massage, repair). The mere presence of text, its words, and meaning are closely related to the semantics of the object or scene. This paper exploits textual contents in images for fine-grained business place classification and logo retrieval. There are four main contributions. First, we show that the textual cues extracted by the proposed method are effective for the two tasks. Combining the proposed textual and visual cues outperforms visual only classification and retrieval by a large margin. Second, to extract the textual cues, a generic and fully unsupervised word box proposal method is introduced. The method reaches state-of-the-art word detection recall with a limited number of proposals. Third, contrary to what is widely acknowledged in text detection literature, we demonstrate that high recall in word detection is more important than high f-score at least for both tasks considered in this work. Last, this paper provides a large annotated text detection dataset with 10 K images and 27 601 word boxes.",
"This paper focuses on fine-grained object classification using recognized scene text in natural images. While the state-of-the-art relies on visual cues only, this paper is the first work which proposes to combine textual and visual cues. Another novelty is the textual cue extraction. Unlike the state-of-the-art text detection methods, we focus more on the background instead of text regions. Once text regions are detected, they are further processed by two methods to perform text recognition, i.e., ABBYY commercial OCR engine and a state-of-the-art character recognition algorithm. Then, to perform textual cue encoding, bi- and trigrams are formed between the recognized characters by considering the proposed spatial pairwise constraints. Finally, extracted visual and textual cues are combined for fine-grained classification. The proposed method is validated on four publicly available data sets: ICDAR03, ICDAR13, Con-Text , and Flickr-logo . We improve the state-of-the-art end-to-end character recognition by a large margin of 15 on ICDAR03. We show that textual cues are useful in addition to visual cues for fine-grained classification. We show that textual cues are also useful for logo retrieval. Adding textual cues outperforms visual- and textual-only in fine-grained classification (70.7 to 60.3 ) and logo retrieval (57.4 to 54.8 ).",
"This paper focuses on fine-grained classification by detecting photographed text in images. We introduce a text detection method that does not try to detect all possible foreground text regions but instead aims to reconstruct the scene background to eliminate non-text regions. Object cues such as color, contrast, and objectiveness are used in corporation with a random forest classifier to detect background pixels in the scene. Results on two publicly available datasets ICDAR03 and a fine-grained Building subcategories of ImageNet shows the effectiveness of the proposed method."
]
} |
1905.10622 | 2946281912 | Images with visual and scene text content are ubiquitous in everyday life. However current image interpretation systems are mostly limited to using only the visual features, neglecting to leverage the scene text content. In this paper we propose to jointly use scene text and visual channels for robust semantic interpretation of images. We undertake the task of matching Advertisement images against their human generated statements that describe the action that the ad prompts and the rationale it provides for taking this action. We extract the scene text and generate semantic and lexical text representations, which are used in the interpretation of the Ad Image. To deal with irrelevant or erroneous detection of scene text, we use a text attention scheme. We also learn an embedding of the visual channel, visual features based on detected symbolism and objects, into a semantic embedding space, leveraging text semantics obtained from scene text. We show how the multi channel approach, involving visual semantics and scene text, improves upon the current state of the art. | As our work deals with information from two different modalities, in this section we also review works where multiple modalities are involved. A common approach adopted by many researchers is to project both modalities into a common subspace. This can be achieved by CCA (Canonical Correlation Analysis) @cite_3 and its deep learning variant (DCCA) @cite_19 . However this needs parallel annotated data for both the modalities. | {
"cite_N": [
"@cite_19",
"@cite_3"
],
"mid": [
"1523385540",
"2100235303"
],
"abstract": [
"We introduce Deep Canonical Correlation Analysis (DCCA), a method to learn complex nonlinear transformations of two views of data such that the resulting representations are highly linearly correlated. Parameters of both transformations are jointly learned to maximize the (regularized) total correlation. It can be viewed as a nonlinear extension of the linear method canonical correlation analysis (CCA). It is an alternative to the nonparametric method kernel canonical correlation analysis (KCCA) for learning correlated nonlinear transformations. Unlike KCCA, DCCA does not require an inner product, and has the advantages of a parametric method: training time scales well with data size and the training data need not be referenced when computing the representations of unseen instances. In experiments on two real-world datasets, we find that DCCA learns representations with significantly higher correlation than those learned by CCA and KCCA. We also introduce a novel non-saturating sigmoid function based on the cube root that may be useful more generally in feedforward neural networks.",
"We present a general method using kernel canonical correlation analysis to learn a semantic representation to web images and their associated text. The semantic space provides a common representation and enables a comparison between the text and images. In the experiments, we look at two approaches of retrieving images based on only their content from a text query. We compare orthogonalization approaches against a standard cross-representation retrieval technique known as the generalized vector space model."
]
} |
1905.10756 | 2946431317 | Partial domain adaptation (PDA) extends standard domain adaptation to a more realistic scenario where the target domain only has a subset of classes from the source domain. The key challenge of PDA is how to select the relevant samples in the shared classes for knowledge transfer. Previous PDA methods tackle this problem by re-weighting the source samples based on the prediction of classifier or discriminator, thus discarding the pixel-level information. In this paper, to utilize both high-level and pixel-level information, we propose a reinforced transfer network (RTNet), which is the first work to apply reinforcement learning to address the PDA problem. The RTNet simultaneously mitigates the negative transfer by adopting a reinforced data selector to filter out outlier source classes, and promotes the positive transfer by employing a domain adaptation model to minimize the distribution discrepancy in the shared label space. Extensive experiments indicate that RTNet can achieve state-of-the-art performance for partial domain adaptation tasks on several benchmark datasets. Codes and datasets will be available online. | Deep domain adaptation methods have been widely studied in recent years. These methods extend deep neural networks by embedding adaptation layers for moment matching @cite_10 @cite_4 @cite_29 @cite_16 or adding domain discriminators for adversarial training @cite_20 @cite_5 . However, these methods may be restricted by the assumption that the source and target domains share the same label space, which is not held in the partial domain adaptation scenario. Several methods have been proposed to solve the PDA problem. Selective adversarial network (SAN) @cite_17 trains a separate domain discriminator for each class with a weight mechanism to suppress the harmful influence of the outlier classes. Partial adversarial domain adaptation (PADA) @cite_30 improves SAN by adopting only one domain discriminator and obtains the weight of each class based on the predicted target probability distribution of the classifier. Example transfer network (ETN) @cite_6 automatically quantifies the weights of source examples based on their similarities to the target domain. Unlike previous PDA methods, which are based on adversarial networks, our proposed RTNet can be integrated not only into the domain adaptation method based on adversarial training but also into the domain adaptation method based on moment matching. | {
"cite_N": [
"@cite_30",
"@cite_4",
"@cite_29",
"@cite_6",
"@cite_5",
"@cite_16",
"@cite_10",
"@cite_20",
"@cite_17"
],
"mid": [
"2885722640",
"2159291411",
"2467286621",
"2926393255",
"2593768305",
"2889153461",
"1565327149",
"1731081199",
"2964109570"
],
"abstract": [
"Domain adversarial learning aligns the feature distributions across the source and target domains in a two-player minimax game. Existing domain adversarial networks generally assume identical label space across different domains. In the presence of big data, there is strong motivation of transferring deep models from existing big domains to unknown small domains. This paper introduces partial domain adaptation as a new domain adaptation scenario, which relaxes the fully shared label space assumption to that the source label space subsumes the target label space. Previous methods typically match the whole source domain to the target domain, which are vulnerable to negative transfer for the partial domain adaptation problem due to the large mismatch between label spaces. We present Partial Adversarial Domain Adaptation (PADA), which simultaneously alleviates negative transfer by down-weighing the data of outlier source classes for training both source classifier and domain adversary, and promotes positive transfer by matching the feature distributions in the shared label space. Experiments show that PADA exceeds state-of-the-art results for partial domain adaptation tasks on several datasets.",
"Recent studies reveal that a deep neural network can learn transferable features which generalize well to novel tasks for domain adaptation. However, as deep features eventually transition from general to specific along the network, the feature transferability drops significantly in higher layers with increasing domain discrepancy. Hence, it is important to formally reduce the dataset bias and enhance the transferability in task-specific layers. In this paper, we propose a new Deep Adaptation Network (DAN) architecture, which generalizes deep convolutional neural network to the domain adaptation scenario. In DAN, hidden representations of all task-specific layers are embedded in a reproducing kernel Hilbert space where the mean embeddings of different domain distributions can be explicitly matched. The domain discrepancy is further reduced using an optimal multikernel selection method for mean embedding matching. DAN can learn transferable features with statistical guarantees, and can scale linearly by unbiased estimate of kernel embedding. Extensive empirical evidence shows that the proposed architecture yields state-of-the-art image classification error rates on standard domain adaptation benchmarks.",
"Deep neural networks are able to learn powerful representations from large quantities of labeled input data, however they cannot always generalize well across changes in input distributions. Domain adaptation algorithms have been proposed to compensate for the degradation in performance due to domain shift. In this paper, we address the case when the target domain is unlabeled, requiring unsupervised adaptation. CORAL is a \"frustratingly easy\" unsupervised domain adaptation method that aligns the second-order statistics of the source and target distributions with a linear transformation. Here, we extend CORAL to learn a nonlinear transformation that aligns correlations of layer activations in deep neural networks (Deep CORAL). Experiments on standard benchmark datasets show state-of-the-art performance.",
"Domain adaptation is critical for learning in new and unseen environments. With domain adversarial training, deep networks can learn disentangled and transferable features that effectively diminish the dataset shift between the source and target domains for knowledge transfer. In the era of Big Data, the ready availability of large-scale labeled datasets has stimulated wide interest in partial domain adaptation (PDA), which transfers a recognizer from a labeled large domain to an unlabeled small domain. It extends standard domain adaptation to the scenario where target labels are only a subset of source labels. Under the condition that target labels are unknown, the key challenge of PDA is how to transfer relevant examples in the shared classes to promote positive transfer, and ignore irrelevant ones in the specific classes to mitigate negative transfer. In this work, we propose a unified approach to PDA, Example Transfer Network (ETN), which jointly learns domain-invariant representations across the source and target domains, and a progressive weighting scheme that quantifies the transferability of source examples while controlling their importance to the learning task in the target domain. A thorough evaluation on several benchmark datasets shows that our approach achieves state-of-the-art results for partial domain adaptation tasks.",
"Adversarial learning methods are a promising approach to training robust deep networks, and can generate complex samples across diverse domains. They can also improve recognition despite the presence of domain shift or dataset bias: recent adversarial approaches to unsupervised domain adaptation reduce the difference between the training and test domain distributions and thus improve generalization performance. However, while generative adversarial networks (GANs) show compelling visualizations, they are not optimal on discriminative tasks and can be limited to smaller shifts. On the other hand, discriminative approaches can handle larger domain shifts, but impose tied weights on the model and do not exploit a GAN-based loss. In this work, we first outline a novel generalized framework for adversarial adaptation, which subsumes recent state-of-the-art approaches as special cases, and use this generalized view to better relate prior approaches. We then propose a previously unexplored instance of our general framework which combines discriminative modeling, untied weight sharing, and a GAN loss, which we call Adversarial Discriminative Domain Adaptation (ADDA). We show that ADDA is more effective yet considerably simpler than competing domain-adversarial methods, and demonstrate the promise of our approach by exceeding state-of-the-art unsupervised adaptation results on standard domain adaptation tasks as well as a difficult cross-modality object classification task.",
"Recently, considerable effort has been devoted to deep domain adaptation in computer vision and machine learning communities. However, most of existing work only concentrates on learning shared feature representation by minimizing the distribution discrepancy across different domains. Due to the fact that all the domain alignment approaches can only reduce, but not remove the domain shift. Target domain samples distributed near the edge of the clusters, or far from their corresponding class centers are easily to be misclassified by the hyperplane learned from the source domain. To alleviate this issue, we propose to joint domain alignment and discriminative feature learning, which could benefit both domain alignment and final classification. Specifically, an instance-based discriminative feature learning method and a center-based discriminative feature learning method are proposed, both of which guarantee the domain invariant features with better intra-class compactness and inter-class separability. Extensive experiments show that learning the discriminative features in the shared feature space can significantly boost the performance of deep domain adaptation methods.",
"Recent reports suggest that a generic supervised deep CNN model trained on a large-scale dataset reduces, but does not remove, dataset bias on a standard benchmark. Fine-tuning deep models in a new domain can require a significant amount of data, which for many applications is simply not available. We propose a new CNN architecture which introduces an adaptation layer and an additional domain confusion loss, to learn a representation that is both semantically meaningful and domain invariant. We additionally show that a domain confusion metric can be used for model selection to determine the dimension of an adaptation layer and the best position for the layer in the CNN architecture. Our proposed adaptation method offers empirical performance which exceeds previously published results on a standard benchmark visual domain adaptation task.",
"We introduce a new representation learning approach for domain adaptation, in which data at training and test time come from similar but different distributions. Our approach is directly inspired by the theory on domain adaptation suggesting that, for effective domain transfer to be achieved, predictions must be made based on features that cannot discriminate between the training (source) and test (target) domains. The approach implements this idea in the context of neural network architectures that are trained on labeled data from the source domain and unlabeled data from the target domain (no labeled target-domain data is necessary). As the training progresses, the approach promotes the emergence of features that are (i) discriminative for the main learning task on the source domain and (ii) indiscriminate with respect to the shift between the domains. We show that this adaptation behaviour can be achieved in almost any feed-forward model by augmenting it with few standard layers and a new gradient reversal layer. The resulting augmented architecture can be trained using standard backpropagation and stochastic gradient descent, and can thus be implemented with little effort using any of the deep learning packages. We demonstrate the success of our approach for two distinct classification problems (document sentiment analysis and image classification), where state-of-the-art domain adaptation performance on standard benchmarks is achieved. We also validate the approach for descriptor learning task in the context of person re-identification application.",
"Adversarial learning has been successfully embedded into deep networks to learn transferable features, which reduce distribution discrepancy between the source and target domains. Existing domain adversarial networks assume fully shared label space across domains. In the presence of big data, there is strong motivation of transferring both classification and representation models from existing large-scale domains to unknown small-scale domains. This paper introduces partial transfer learning, which relaxes the shared label space assumption to that the target label space is only a subspace of the source label space. Previous methods typically match the whole source domain to the target domain, which are prone to negative transfer for the partial transfer problem. We present Selective Adversarial Network (SAN), which simultaneously circumvents negative transfer by selecting out the outlier source classes and promotes positive transfer by maximally matching the data distributions in the shared label space. Experiments demonstrate that our models exceed state-of-the-art results for partial transfer learning tasks on several benchmark datasets."
]
} |
1905.10497 | 2946193510 | Federated learning involves training statistical models in massive, heterogeneous networks. Naively minimizing an aggregate loss function in such a network may disproportionately advantage or disadvantage some of the devices. In this work, we propose q-Fair Federated Learning (q-FFL), a novel optimization objective inspired by resource allocation in wireless networks that encourages a more fair (i.e., lower-variance) accuracy distribution across devices in federated networks. To solve q-FFL, we devise a communication-efficient method, q-FedAvg, that is suited to federated networks. We validate both the effectiveness of q-FFL and the efficiency of q-FedAvg on a suite of federated datasets, and show that q-FFL (along with q-FedAvg) outperforms existing baselines in terms of the resulting fairness, flexibility, and efficiency. | Fair resource allocation has been extensively studied in fields such as network management @cite_20 @cite_31 @cite_16 @cite_3 and wireless communications @cite_43 @cite_1 @cite_19 @cite_47 . In these contexts, the problem is defined as allocating a scarce shared resource, e.g., communication time or power, among many users. In these cases directly maximizing utilities such as total throughput usually leads to unfair allocations where some users receive poor service. As a service provider, it is important to improve the quality of service for all users while maintaining overall throughput. For this reason several popular fairness measurements have been proposed to balance between fairness and total throughput, including Jain's index @cite_4 , entropy @cite_13 , max-min min-max fairness @cite_39 , and proportional fairness @cite_28 . A unified framework is captured through @math -fairness @cite_5 @cite_35 , in which the network manager can tune the emphasis on fairness by changing a single parameter, @math . | {
"cite_N": [
"@cite_35",
"@cite_47",
"@cite_4",
"@cite_28",
"@cite_1",
"@cite_3",
"@cite_39",
"@cite_43",
"@cite_19",
"@cite_5",
"@cite_31",
"@cite_16",
"@cite_13",
"@cite_20"
],
"mid": [
"2147524058",
"1969791930",
"2100960835",
"1987497363",
"2100671335",
"2120344179",
"2126143263",
"2098254860",
"2094083774",
"2122011919",
"2105520014",
"2159715570",
"2122882636",
"2030315897"
],
"abstract": [
"In this paper, we demonstrate the existence of fair end-to-end window-based congestion control protocols for packet-switched networks with first come-first served routers. Our definition of fairness generalizes proportional fairness and includes arbitrarily close approximations of max-min fairness. The protocols use only information that is available to end hosts and are designed to converge reasonably fast. Our study is based on a multiclass fluid model of the network. The convergence of the protocols is proved using a Lyapunov function. The technical challenge is in the practical implementation of the protocols.",
"The pervasiveness of wireless technology has indeed created massive opportunity to integrate almost everything into the Internet fabric. This can be seen with the advent of Internet of Things and Cyber Physical Systems, which involves cooperation of massive number of intelligent devices to provide intelligent services. Fairness amongst these devices is an important issue that can be analysed from several dimensions, e.g., energy usage, achieving required quality of services, spectrum sharing, and so on. This article focusses on these viewpoints while looking at fairness research. To generalize, mainly wireless networks are considered. First, we present a general view of fairness studies, and pose three core questions that help us delineate the nuances in defining fairness. Then, the existing fairness models are summarized and compared. We also look into the major fairness research domains in wireless networks such as fair energy consumption control, power control, topology control, link and flow scheduling, channel assignment, rate allocation, congestion control and routing protocols. We make a distinction amongst fairness, utility and resource allocation to begin with. Later, we present their inter-relation. At the end of this article, we list the common properties of fairness and give an example of fairness management. Several open research challenges that point to further work on fairness in wireless networks are also discussed. Indeed, the research on fairness is entangled with many other aspects such as performance, utility, optimization and throughput at the network and node levels. While consolidating the contributions in the literature, this article tries to explain the niceties of all these aspects in the domain of wireless networking.",
"We study fairness in classification, where individuals are classified, e.g., admitted to a university, and the goal is to prevent discrimination against individuals based on their membership in some group, while maintaining utility for the classifier (the university). The main conceptual contribution of this paper is a framework for fair classification comprising (1) a (hypothetical) task-specific metric for determining the degree to which individuals are similar with respect to the classification task at hand; (2) an algorithm for maximizing utility subject to the fairness constraint, that similar individuals are treated similarly. We also present an adaptation of our approach to achieve the complementary goal of \"fair affirmative action,\" which guarantees statistical parity (i.e., the demographics of the set of individuals receiving any classification are the same as the demographics of the underlying population), while treating similar individuals as similarly as possible. Finally, we discuss the relationship of fairness to privacy: when fairness implies privacy, and how tools developed in the context of differential privacy may be applied to fairness.",
"This paper addresses the issues of charging, rate control and routing for a communication network carrying elastic traffic, such as an ATM network offering an available bit rate service. A model is described from which max-min fairness of rates emerges as a limiting special case; more generally, the charges users are prepared to pay influence their allocated rates. In the preferred version of the model, a user chooses the charge per unit time that the user will pay; thereafter the user's rate is determined by the network according to a proportional fairness criterion applied to the rate per unit charge. A system optimum is achieved when users' choices of charges and the network's choice of allocated rates are in equilibrium.",
"Link-layer fairness models that have been proposed for wireline and packet cellular networks cannot be generalized for shared channel wireless networks because of the unique characteristics of the wireless channel, such as location-dependent contention, inherent conflict between optimizing channel utilization and achieving fairness, and the absence of any centralized control. In this paper, we propose a general analytical framework that captures the unique characteristics of shared wireless channels and allows the modeling of a large class of system-wide fairness models via the specification of per-flow utility functions. We show that system-wide fairness can be achieved without explicit global coordination so long as each node executes a contention resolution algorithm that is designed to optimize its local utility function. We present a general mechanism for translating a given fairness model in our framework into a corresponding contention resolution algorithm . Using this translation, we derive the backoff algorithm for achieving proportional fairness in wireless shared channels, and compare the fairness properties of this algorithm with both the ideal proportional fairness objective, and state-of-the-art backoff-based contention resolution algorithms. We believe that the two aspects of the proposed framework, i.e. the ability to specify arbitrary fairness models via local utility functions, and the ability to automatically generate local contention resolution mechanisms in response to a given utility function, together provide the path for achieving flexible service differentiation in future shared channel wireless networks.",
"We consider optimal control for general networks with both wireless and wireline components and time varying channels. A dynamic strategy is developed to support all traffic whenever possible, and to make optimally fair decisions about which data to serve when inputs exceed network capacity. The strategy is decoupled into separate algorithms for flow control, routing, and resource allocation, and allows each user to make decisions independent of the actions of others. The combined strategy is shown to yield data rates that are arbitrarily close to the optimal operating point achieved when all network controllers are coordinated and have perfect knowledge of future events. The cost of approaching this fair operating point is an end-to-end delay increase for data that is served by the network.",
"Max-min fairness is widely used in various areas of networking. In every case where it is used, there is a proof of existence and one or several algorithms for computing it; in most, but not all cases, they are based on the notion of bottlenecks. In spite of this wide applicability, there are still examples, arising in the context of wireless or peer-to-peer networks, where the existing theories do not seem to apply directly. In this paper, we give a unifying treatment of max-min fairness, which encompasses all existing results in a simplifying framework, and extend its applicability to new examples. First, we observe that the existence of max-min fairness is actually a geometric property of the set of feasible allocations. There exist sets on which max-min fairness does not exist, and we describe a large class of sets on which a max-min fair allo cation does exist. This class contains, but is not limited to the compact, convex sets of RN. Second, we give a general purpose centralized algorithm, called Max-min Programming, for computing the max-min fair allocation in all cases where it exists (whether the set of feasible allocations is in our class or not). Its complexity is of the order of N linear programming steps in RN, in the case where the feasible set is defined by linear constraints. We show that, if the set of feasible allocations has the free disposal property, then Max-min Programming reduces to a simpler algorithm, called Water Filling, whose complexity is much lower. Free disposal corresponds to the cases where a bottleneck argument can be made, andWater Filling is the general form of all previously known centralized algorithms for such cases. All our results apply mutatis mutandis to min-max fairness. Our results apply to weighted, unweighted and util-max-min and min-max fairness. Distributed algorithms for the computation of max-min fair allocations are outside the scope of this paper.",
"In this paper, we describe and analyze a joint scheduling, routing and congestion control mechanism for wireless networks, that asymptotically guarantees stability of the buffers and fair allocation of the network resources. The queue-lengths serve as common information to different layers of the network protocol stack. Our main contribution is to prove the asymptotic optimality of a primal-dual congestion controller, which is known to model different versions of transmission control protocol well",
"Consider a downlink MIMO heterogeneous wireless network with multiple cells, each containing many mobile users and a number of base stations with varying capabilities. A central task in the management of such a network is to assign each user to a base station and design a linear transmit strategy to ensure a satisfactory level of network performance. In this work, we propose a formulation for the joint base station assignment and linear transceiver design problem based on the maximization of a system wide utility. We first establish the NP-hardness of the resulting optimization problem for a large family of α-fairness utility functions. Then, we propose an efficient algorithm to approximately solve this problem for a variety of different utility functions. Our numerical experiments show that the proposed algorithm can achieve significantly higher levels of system throughput and user fairness than what is possible by optimizing the precoders alone.",
"We present five axioms for fairness measures in resource allocation. A family of fairness measures satisfying the axioms is constructed. Special cases of this family include α-fairness, Jain's index, and entropy. Properties of fairness measures satisfying the axioms are proven, including Schur-concavity. Among the engineering implications is a generalized Jain's index that tunes the resolution of fairness measure, a new understanding of α-fair utility functions, and an interpretation of larger α is more fair''. We also construct an alternative set of axioms to capture system efficiency and feasibility constraints.",
"The author studies a simple strategy, proposed independently by E.L. Hahne and R.G. Gallager (1986) and M.G.H. Katevenis (1987), for fairly allocating link capacity in a point-to-point packet network with virtual circuit routing. Each link offers its packet transmission slots to its user sessions by polling them in round-robin order. In addition, window flow control is used to prevent excessive packet queues at the network nodes. As the window size increases, the session throughput rates are shown to approach limits that are perfectly fair in the max-min sense. If each session has periodic input (perhaps with jitter) or has such heavy demand that packets are always waiting to enter the network, then a finite window size suffices to produce perfectly fair throughput rates. The results suggest that the transmission capacity not used by the small window session will be approximately fairly divided among the large window sessions. The focus is on the worst-case performance of round-robin scheduling with windows. >",
"This paper analyses the stability and fairness of two classes of rate control algorithm for communication networks. The algorithms provide natural generalisations to large-scale networks of simple additive increase multiplicative decrease schemes, and are shown to be stable about a system optimum characterised by a proportional fairness criterion. Stability is established by showing that, with an appropriate formulation of the overall optimisation problem, the network's implicit objective function provides a Lyapunov function for the dynamical system defined by the rate control algorithm. The network's optimisation problem may be cast in primal or dual form: this leads naturally to two classes of algorithm, which may be interpreted in terms of either congestion indication feedback signals or explicit rates based on shadow prices. Both classes of algorithm may be generalised to include routing control, and provide natural implementations of proportionally fair pricing.",
"",
"In this paper we propose a distributed and scalable algorithm that eliminates congestion within a sensor network, and that ensures the fair delivery of packets to a central node, or base station. We say that fairness is achieved when equal number of packets are received from each node. Since in general we have many sensors transmitting data to the base station, we consider the scenario where we have many-to-one multihop routing, noting that it can easily be extended to unicast or many-to-many routing. Such routing structures often result in the sensors closer to the base station experiencing congestion, which inevitably cause packets originating from sensors further away from the base station to have a higher probability of being dropped. Our algorithm exists in the transport layer of the traditional network stack model, and is designed to work with any MAC protocol in the data-link layer with minor modifications. Our solution is scalable, each sensor mote requires state proportional to the number of its neighbors. Finally, we demonstrate the effectiveness of our solution with both simulations and actual implementation in UC Berkeley's sensor motes."
]
} |
1905.10565 | 2946787698 | In recent years, the proliferation of smart mobile devices has lead to the gradual integration of search functionality within mobile platforms. This has created an incentive to move away from the "ten blue links" metaphor, as mobile users are less likely to click on them, expecting to get the answer directly from the snippets. In turn, this has revived the interest in Question Answering. Then, along came chatbots, conversational systems, and messaging platforms, where the user needs could be better served with the system asking follow-up questions in order to better understand the user's intent. While typically a user would expect a single response at any utterance, a system could also return multiple options for the user to select from, based on different system understandings of the user's intent. However, this possibility should not be overused, as this practice could confuse and or annoy the user. How to produce good variable-length lists, given the conflicting objectives of staying short while maximizing the likelihood of having a correct answer included in the list, is an underexplored problem. It is also unclear how to evaluate a system that tries to do that. Here we aim to bridge this gap. In particular, we define some necessary and some optional properties that an evaluation measure fit for this purpose should have. We further show that existing evaluation measures from the IR tradition are not entirely suitable for this setup, and we propose novel evaluation measures that address it satisfactorily. | As we define a set of properties that need to be satisfied by an evaluation measure for variable-length output, our work is closely related to the properties of truncated evaluation measures introduced in @cite_17 . Their property is similar to our , except that imposes strict monotonicity. The property is similar to our , but it discounts for irrelevant documents at the end of the list only. | {
"cite_N": [
"@cite_17"
],
"mid": [
"2902032968"
],
"abstract": [
"A wide range of evaluation metrics have been proposed to measure the quality of search results, including in the presence of diversification. Some of these metrics have been adapted for use in search tasks with different complexities, such as where the search system returns lists of different lengths. Given the range of requirements, it can be difficult to compare the behavior of these metrics. In this work, we examine effectiveness metrics using a simple property-based approach. In particular, we present a case-analysis framework to define and study fundamental properties that seem integral to any evaluation metric. An example of a simple property is that a ranking with only one non-relevant document should never score lower than a ranking with two non-relevant documents. The framework facilitates quantifying the ability of metrics to satisfy properties, both separately and simultaneously, and to identify those cases where properties are violated. Our analysis shows that the Average Cube Test and Intent-Aware Average Precision are two metrics which fail to satisfy the desirable properties, and hence should be used with caution."
]
} |
1905.10565 | 2946787698 | In recent years, the proliferation of smart mobile devices has lead to the gradual integration of search functionality within mobile platforms. This has created an incentive to move away from the "ten blue links" metaphor, as mobile users are less likely to click on them, expecting to get the answer directly from the snippets. In turn, this has revived the interest in Question Answering. Then, along came chatbots, conversational systems, and messaging platforms, where the user needs could be better served with the system asking follow-up questions in order to better understand the user's intent. While typically a user would expect a single response at any utterance, a system could also return multiple options for the user to select from, based on different system understandings of the user's intent. However, this possibility should not be overused, as this practice could confuse and or annoy the user. How to produce good variable-length lists, given the conflicting objectives of staying short while maximizing the likelihood of having a correct answer included in the list, is an underexplored problem. It is also unclear how to evaluate a system that tries to do that. Here we aim to bridge this gap. In particular, we define some necessary and some optional properties that an evaluation measure fit for this purpose should have. We further show that existing evaluation measures from the IR tradition are not entirely suitable for this setup, and we propose novel evaluation measures that address it satisfactorily. | @cite_19 defined seven properties for effectiveness measures. The relevant properties, which our new evaluation measures also satisfy are , , and . However, their property contradicts our property because it states that adding documents, which are not relevant, to the end of the list increases the score. | {
"cite_N": [
"@cite_19"
],
"mid": [
"2231711798"
],
"abstract": [
"Search effectiveness metrics quantify the relevance of the ranked document lists returned by retrieval systems. In this paper we characterize metrics according to seven numeric properties – boundedness, monotonicity, convergence, top-weightedness, localization, completeness, and realizability. We demonstrate that these properties partition the commonly-used evaluation metrics, and hence provide a framework in which the relationships between effectiveness metrics can be better understood, including their relative merits for different applications."
]
} |
1905.10565 | 2946787698 | In recent years, the proliferation of smart mobile devices has lead to the gradual integration of search functionality within mobile platforms. This has created an incentive to move away from the "ten blue links" metaphor, as mobile users are less likely to click on them, expecting to get the answer directly from the snippets. In turn, this has revived the interest in Question Answering. Then, along came chatbots, conversational systems, and messaging platforms, where the user needs could be better served with the system asking follow-up questions in order to better understand the user's intent. While typically a user would expect a single response at any utterance, a system could also return multiple options for the user to select from, based on different system understandings of the user's intent. However, this possibility should not be overused, as this practice could confuse and or annoy the user. How to produce good variable-length lists, given the conflicting objectives of staying short while maximizing the likelihood of having a correct answer included in the list, is an underexplored problem. It is also unclear how to evaluate a system that tries to do that. Here we aim to bridge this gap. In particular, we define some necessary and some optional properties that an evaluation measure fit for this purpose should have. We further show that existing evaluation measures from the IR tradition are not entirely suitable for this setup, and we propose novel evaluation measures that address it satisfactorily. | @cite_8 @cite_7 also conducted an axiomatic'' analysis of IR-related evaluation measures. Our properties and are akin to the ones discussed in @cite_2 , but the latter are used in a different setup, i.e., for document retrieval, clustering, and filtering. | {
"cite_N": [
"@cite_2",
"@cite_7",
"@cite_8"
],
"mid": [
"2051442094",
"2014364160",
"2799104453"
],
"abstract": [
"A number of key Information Access tasks -- Document Retrieval, Clustering, Filtering, and their combinations -- can be seen as instances of a generic document organization problem that establishes priority and relatedness relationships between documents (in other words, a problem of forming and ranking clusters). As far as we know, no analysis has been made yet on the evaluation of these tasks from a global perspective. In this paper we propose two complementary evaluation measures -- Reliability and Sensitivity -- for the generic Document Organization task which are derived from a proposed set of formal constraints (properties that any suitable measure must satisfy). In addition to be the first measures that can be applied to any mixture of ranking, clustering and filtering tasks, Reliability and Sensitivity satisfy more formal constraints than previously existing evaluation metrics for each of the subsumed tasks. Besides their formal properties, its most salient feature from an empirical point of view is their strictness: a high score according to the harmonic mean of Reliability and Sensitivity ensures a high score with any of the most popular evaluation metrics in all the Document Retrieval, Clustering and Filtering datasets used in our experiments.",
"We address the general problem of finding suitable evaluation measures for classification systems. To this end, we adopt an axiomatic approach, i.e., we discuss a number of properties (\"axioms\") that an evaluation measure for classification should arguably satisfy. We start our analysis by addressing binary classification. We show that F1, nowadays considered a standard measure for the evaluation of binary classification systems, does not comply with a number of them, and should thus be considered unsatisfactory. We go on to discuss an alternative, simple evaluation measure for binary classification, that we call K, and show that it instead satisfies all the previously proposed axioms. We thus argue that researchers and practitioners should replace F1 with K in their everyday binary classification practice. We carry on our analysis by showing that K can be smoothly extended to deal with single-label multi-class classification, cost-sensitive classification, and ordinal classification.",
"Many evaluation metrics have been defined to evaluate the effectiveness ad-hoc retrieval and search result diversification systems. However, it is often unclear which evaluation metric should be used to analyze the performance of retrieval systems given a specific task. Axiomatic analysis is an informative mechanism to understand the fundamentals of metrics and their suitability for particular scenarios. In this paper, we define a constraint-based axiomatic framework to study the suitability of existing metrics in search result diversification scenarios. The analysis informed the definition of Rank-Biased Utility (RBU) -- an adaptation of the well-known Rank-Biased Precision metric -- that takes into account redundancy and the user effort associated to the inspection of documents in the ranking. Our experiments over standard diversity evaluation campaigns show that the proposed metric captures quality criteria reflected by different metrics, being suitable in the absence of knowledge about particular features of the scenario under study."
]
} |
1905.10565 | 2946787698 | In recent years, the proliferation of smart mobile devices has lead to the gradual integration of search functionality within mobile platforms. This has created an incentive to move away from the "ten blue links" metaphor, as mobile users are less likely to click on them, expecting to get the answer directly from the snippets. In turn, this has revived the interest in Question Answering. Then, along came chatbots, conversational systems, and messaging platforms, where the user needs could be better served with the system asking follow-up questions in order to better understand the user's intent. While typically a user would expect a single response at any utterance, a system could also return multiple options for the user to select from, based on different system understandings of the user's intent. However, this possibility should not be overused, as this practice could confuse and or annoy the user. How to produce good variable-length lists, given the conflicting objectives of staying short while maximizing the likelihood of having a correct answer included in the list, is an underexplored problem. It is also unclear how to evaluate a system that tries to do that. Here we aim to bridge this gap. In particular, we define some necessary and some optional properties that an evaluation measure fit for this purpose should have. We further show that existing evaluation measures from the IR tradition are not entirely suitable for this setup, and we propose novel evaluation measures that address it satisfactorily. | In , we discussed and evaluated the most relevant evaluation measures for both unranked (precision, recall, @math , and smoothed @math ) and ranked retrieval ( @cite_11 , , , smoothed , @cite_10 ). We found that they were unable to penalize the wrong responses according to the gold order,'' thus violating . | {
"cite_N": [
"@cite_10",
"@cite_11"
],
"mid": [
"2470681930",
"1985554184"
],
"abstract": [
"Many types of search tasks are answered through the computation of a ranked list of suggested answers. We re-examine the usual assumption that answer lists should be as long as possible, and suggest that when the number of matching items is potentially small -- perhaps even zero -- it may be more helpful to \"quit while ahead\", that is, to truncate the answer ranking earlier rather than later. To capture this effect, metrics are required which are attuned to the length of the ranking, and can handle cases in which there are no relevant documents. In this work we explore a generalized approach for representing truncated result sets, and propose modifications to a number of popular evaluation metrics.",
"This paper proposes evaluation methods based on the use of non-dichotomous relevance judgements in IR experiments. It is argued that evaluation methods should credit IR methods for their ability to retrieve highly relevant documents. This is desirable from the user point of view in modem large IR environments. The proposed methods are (1) a novel application of P-R curves and average precision computations based on separate recall bases for documents of different degrees of relevance, and (2) two novel measures computing the cumulative gain the user obtains by examining the retrieval result up to a given ranked position. We then demonstrate the use of these evaluation methods in a case study on the effectiveness of query types, based on combinations of query structures and expansion, in retrieving documents of various degrees of relevance. The test was run with a best match retrieval system (In- Query I) in a text database consisting of newspaper articles. The results indicate that the tested strong query structures are most effective in retrieving highly relevant documents. The differences between the query types are practically essential and statistically significant. More generally, the novel evaluation methods and the case demonstrate that non-dichotomous relevance assessments are applicable in IR experiments, may reveal interesting phenomena, and allow harder testing of IR methods."
]
} |
1905.10565 | 2946787698 | In recent years, the proliferation of smart mobile devices has lead to the gradual integration of search functionality within mobile platforms. This has created an incentive to move away from the "ten blue links" metaphor, as mobile users are less likely to click on them, expecting to get the answer directly from the snippets. In turn, this has revived the interest in Question Answering. Then, along came chatbots, conversational systems, and messaging platforms, where the user needs could be better served with the system asking follow-up questions in order to better understand the user's intent. While typically a user would expect a single response at any utterance, a system could also return multiple options for the user to select from, based on different system understandings of the user's intent. However, this possibility should not be overused, as this practice could confuse and or annoy the user. How to produce good variable-length lists, given the conflicting objectives of staying short while maximizing the likelihood of having a correct answer included in the list, is an underexplored problem. It is also unclear how to evaluate a system that tries to do that. Here we aim to bridge this gap. In particular, we define some necessary and some optional properties that an evaluation measure fit for this purpose should have. We further show that existing evaluation measures from the IR tradition are not entirely suitable for this setup, and we propose novel evaluation measures that address it satisfactorily. | Another relevant field of research analyzes the likelihood that the user will continue exploring the response list based on different signals - time-biased gain @cite_13 , length of the snippet @cite_9 , and information foraging @cite_5 . However, in mobile search and chatbot platforms, the presented information is already minimized, and thus we aim to reduce the length of the returned results instead. | {
"cite_N": [
"@cite_5",
"@cite_9",
"@cite_13"
],
"mid": [
"2808234013",
"2106404250",
"1970977883"
],
"abstract": [
"Web Search Engine Result Pages (SERPs) are complex responses to queries, containing many heterogeneous result elements (web results, advertisements, and specialised \"answers'') positioned in a variety of layouts. This poses numerous challenges when trying to measure the quality of a SERP because standard measures were designed for homogeneous ranked lists. In this paper, we aim to measure the utility and cost of SERPs. To ground this work we adopt the framework which enables a direct comparison between different measures in the same units of measurement, i.e. expected (total) utility and cost. Within this framework, we propose a new measure based on information foraging theory, which can account for the heterogeneity of elements, through different costs, and which naturally motivates the development of a user stopping model that adapts behaviour depending on the rate of gain. This directly connects models of how people search with how we measure search, providing a number of new dimensions in which to investigate and evaluate user behaviour and performance. We perform an analysis over 1000 popular queries issued to a major search engine, and report the aggregate utility experienced by users over time. Then in an comparison against common measures, we show that the proposed foraging based measure provides a more accurate reflection of the utility and of observed behaviours (stopping rank and time spent).",
"We introduce a general information access evaluation framework that can potentially handle summaries, ranked document lists and even multi query sessions seamlessly. Our framework first builds a trailtext which represents a concatenation of all the texts read by the user during a search session, and then computes an evaluation metric called U-measure over the trailtext. Instead of discounting the value of a retrieved piece of information based on ranks, U-measure discounts it based on its position within the trailtext. U-measure takes the document length into account just like Time-Biased Gain (TBG), and has the diminishing return property. It is therefore more realistic than rank-based metrics. Furthermore, it is arguably more flexible than TBG, as it is free from the linear traversal assumption (i.e., that the user scans the ranked list from top to bottom), and can handle information access tasks other than ad hoc retrieval. This paper demonstrates the validity and versatility of the U-measure framework. Our main conclusions are: (a) For ad hoc retrieval, U-measure is at least as reliable as TBG in terms of rank correlations with traditional metrics and discriminative power; (b) For diversified search, our diversity versions of U-measure are highly correlated with state-of-the-art diversity metrics; (c) For multi-query sessions, U-measure is highly correlated with Session nDCG; and (d) Unlike rank-based metrics such as DCG, U-measure can quantify the differences between linear and nonlinear traversals in sessions. We argue that our new framework is useful for understanding the user's search behaviour and for comparison across different information access styles (e.g. examining a direct answer vs. examining a ranked list of web pages).",
"Cranfield-style information retrieval evaluation considers variance in user information needs by evaluating retrieval systems over a set of search topics. For each search topic, traditional metrics model all users searching ranked lists in exactly the same manner and thus have zero variance in their per-topic estimate of effectiveness. Metrics that fail to model user variance overestimate the effect size of differences between retrieval systems. The modeling of user variance is critical to understanding the impact of effectiveness differences on the actual user experience. If the variance of a difference is high, the effect on user experience will be low. Time-biased gain is an evaluation metric that models user interaction with ranked lists that are displayed using document surrogates. In this paper, we extend the stochastic simulation of time-biased gain to model the variation between users. We validate this new version of time-biased gain by showing that it produces distributions of gain that agree well with actual distributions produced by real users. With a per-topic variance in its effectiveness measure, time-biased gain allows for the measurement of the effect size of differences, which allows researchers to understand the extent to which predicted performance improvements matter to real users."
]
} |
1905.10529 | 2946310364 | Person re-identification (Re-ID) across multiple datasets is a challenging yet important task due to the possibly large distinctions between different datasets and the lack of training samples in practical applications. This work proposes a novel unsupervised domain adaption framework which transfers discriminative representations from the labeled source domain (dataset) to the unlabeled target domain (dataset). We propose to formulate the domain adaption task as an one-class classification problem with a novel domain similarity loss. Given the feature map of any image from a backbone network, a novel domain adaptive attention model (DAAM) first automatically learns to separate the feature map of an image to a domain-shared feature (DSH) map and a domain-specific feature (DSP) map simultaneously. Specially, the residual attention mechanism is designed to model DSP feature map for avoiding negative transfer. Then, a DSH branch and a DSP branch are introduced to learn DSH and DSP feature maps respectively. To reduce domain divergence caused by that the source and target datasets are collected from different environments, we force to project the DSH feature maps from different domains to a new nominal domain, and a novel domain similarity loss is proposed based on one-class classification. In addition, a novel unsupervised person Re-ID loss is proposed to take full use of unlabeled target data. Extensive experiments on the Market-1501 and DukeMTMC-reID benchmarks demonstrate state-of-the-art performance of the proposed method. Code will be released to facilitate further studies on the cross-domain person re-identification task. | Person Re-ID has been one of the most studied problems due to its important application, and most of the existing works are designed based on supervised learning frameworks @cite_21 @cite_47 @cite_14 @cite_6 @cite_25 @cite_24 . These methods require sufficient labeled images across cameras to learn a discriminative representation or matching function. However, it is often difficult to label person Re-ID data in many real-world applications, which severely limits the scalability of these supervised learning based methods. | {
"cite_N": [
"@cite_14",
"@cite_21",
"@cite_6",
"@cite_24",
"@cite_47",
"@cite_25"
],
"mid": [
"2963842104",
"1920259731",
"",
"2606377603",
"2598634450",
"2964163358"
],
"abstract": [
"Employing part-level features offers fine-grained information for pedestrian image description. A prerequisite of part discovery is that each part should be well located. Instead of using external resources like pose estimator, we consider content consistency within each part for precise part location. Specifically, we target at learning discriminative part-informed features for person retrieval and make two contributions. (i) A network named Part-based Convolutional Baseline (PCB). Given an image input, it outputs a convolutional descriptor consisting of several part-level features. With a uniform partition strategy, PCB achieves competitive results with the state-of-the-art methods, proving itself as a strong convolutional baseline for person retrieval. (ii) A refined part pooling (RPP) method. Uniform partition inevitably incurs outliers in each part, which are in fact more similar to other parts. RPP re-assigns these outliers to the parts they are closest to, resulting in refined parts with enhanced within-part consistency. Experiment confirms that RPP allows PCB to gain another round of performance boost. For instance, on the Market-1501 dataset, we achieve (77.4+4.2) mAP and (92.3+1.5) rank-1 accuracy, surpassing the state of the art by a large margin. Code is available at: https: github.com syfafterzy PCB_RPP",
"We propose an effective structured learning based approach to the problem of person re-identification which outperforms the current state-of-the-art on most benchmark data sets evaluated. Our framework is built on the basis of multiple low-level hand-crafted and high-level visual features. We then formulate two optimization algorithms, which directly optimize evaluation measures commonly used in person re-identification, also known as the Cumulative Matching Characteristic (CMC) curve. Our new approach is practical to many real-world surveillance applications as the re-identification performance can be concentrated in the range of most practical importance. The combination of these factors leads to a person re-identification system which outperforms most existing algorithms. More importantly, we advance state-of-the-art results on person re-identification by improving the rank-1 recognition rates from 40 to 50 on the iLIDS benchmark, 16 to 18 on the PRID2011 benchmark, 43 to 46 on the VIPeR benchmark, 34 to 53 on the CUHK01 benchmark and 21 to 62 on the CUHK03 benchmark.",
"",
"Person re-identification (ReID) is an important task in wide area video surveillance which focuses on identifying people across different cameras. Recently, deep learning networks with a triplet loss become a common framework for person ReID. However, the triplet loss pays main attentions on obtaining correct orders on the training set. It still suffers from a weaker generalization capability from the training set to the testing set, thus resulting in inferior performance. In this paper, we design a quadruplet loss, which can lead to the model output with a larger inter-class variation and a smaller intra-class variation compared to the triplet loss. As a result, our model has a better generalization ability and can achieve a higher performance on the testing set. In particular, a quadruplet deep network using a margin-based online hard negative mining is proposed based on the quadruplet loss for the person ReID. In extensive experiments, the proposed network outperforms most of the state-of-the-art algorithms on representative datasets which clearly demonstrates the effectiveness of our proposed method.",
"In the past few years, the field of computer vision has gone through a revolution fueled mainly by the advent of large datasets and the adoption of deep convolutional neural networks for end-to-end learning. The person re-identification subfield is no exception to this. Unfortunately, a prevailing belief in the community seems to be that the triplet loss is inferior to using surrogate losses (classification, verification) followed by a separate metric learning step. We show that, for models trained from scratch as well as pretrained ones, using a variant of the triplet loss to perform end-to-end deep metric learning outperforms most other published methods by a large margin.",
"Person Re-identification (ReID) is to identify the same person across different cameras. It is a challenging task due to the large variations in person pose, occlusion, background clutter, etc. How to extract powerful features is a fundamental problem in ReID and is still an open problem today. In this paper, we design a Multi-Scale Context-Aware Network (MSCAN) to learn powerful features over full body and body parts, which can well capture the local context knowledge by stacking multi-scale convolutions in each layer. Moreover, instead of using predefined rigid parts, we propose to learn and localize deformable pedestrian parts using Spatial Transformer Networks (STN) with novel spatial constraints. The learned body parts can release some difficulties, e.g. pose variations and background clutters, in part-based representation. Finally, we integrate the representation learning processes of full body and body parts into a unified framework for person ReID through multi-class person identification tasks. Extensive evaluations on current challenging large-scale person ReID datasets, including the image-based Market1501, CUHK03 and sequence-based MARS datasets, show that the proposed method achieves the state-of-the-art results."
]
} |
1905.10529 | 2946310364 | Person re-identification (Re-ID) across multiple datasets is a challenging yet important task due to the possibly large distinctions between different datasets and the lack of training samples in practical applications. This work proposes a novel unsupervised domain adaption framework which transfers discriminative representations from the labeled source domain (dataset) to the unlabeled target domain (dataset). We propose to formulate the domain adaption task as an one-class classification problem with a novel domain similarity loss. Given the feature map of any image from a backbone network, a novel domain adaptive attention model (DAAM) first automatically learns to separate the feature map of an image to a domain-shared feature (DSH) map and a domain-specific feature (DSP) map simultaneously. Specially, the residual attention mechanism is designed to model DSP feature map for avoiding negative transfer. Then, a DSH branch and a DSP branch are introduced to learn DSH and DSP feature maps respectively. To reduce domain divergence caused by that the source and target datasets are collected from different environments, we force to project the DSH feature maps from different domains to a new nominal domain, and a novel domain similarity loss is proposed based on one-class classification. In addition, a novel unsupervised person Re-ID loss is proposed to take full use of unlabeled target data. Extensive experiments on the Market-1501 and DukeMTMC-reID benchmarks demonstrate state-of-the-art performance of the proposed method. Code will be released to facilitate further studies on the cross-domain person re-identification task. | To solve the above scalability issue, a natural solution is to utilize unsupervised domain adaption method which aims to transfer useful Re-ID information from the labeled source domain (dataset) to unlabeled target domain (dataset). However, most of existing unsupervised domain adaptation methods @cite_16 @cite_5 @cite_4 @cite_22 @cite_34 @cite_23 @cite_8 are designed to the cases where the source and target domains have the same recognition tasks ( having the same set of classes), which is invalid to person Re-ID problem as different datasets contain totally different person identities. | {
"cite_N": [
"@cite_4",
"@cite_22",
"@cite_8",
"@cite_34",
"@cite_23",
"@cite_5",
"@cite_16"
],
"mid": [
"2279034837",
"1882958252",
"2593768305",
"2214409633",
"2511131004",
"2951670162",
"1565327149"
],
"abstract": [
"The recent success of deep neural networks relies on massive amounts of labeled data. For a target task where labeled data is unavailable, domain adaptation can transfer a learner from a different source domain. In this paper, we propose a new approach to domain adaptation in deep networks that can jointly learn adaptive classifiers and transferable features from labeled data in the source domain and unlabeled data in the target domain. We relax a shared-classifier assumption made by previous methods and assume that the source classifier and target classifier differ by a residual function. We enable classifier adaptation by plugging several layers into deep network to explicitly learn the residual function with reference to the target classifier. We fuse features of multiple layers with tensor product and embed them into reproducing kernel Hilbert spaces to match distributions for feature adaptation. The adaptation can be achieved in most feed-forward models by extending them with new residual layers and loss functions, which can be trained efficiently via back-propagation. Empirical evidence shows that the new approach outperforms state of the art methods on standard domain adaptation benchmarks.",
"Top-performing deep architectures are trained on massive amounts of labeled data. In the absence of labeled data for a certain task, domain adaptation often provides an attractive option given that labeled data of similar nature but from a different domain (e.g. synthetic images) are available. Here, we propose a new approach to domain adaptation in deep architectures that can be trained on large amount of labeled data from the source domain and large amount of unlabeled data from the target domain (no labeled target-domain data is necessary). As the training progresses, the approach promotes the emergence of \"deep\" features that are (i) discriminative for the main learning task on the source domain and (ii) invariant with respect to the shift between the domains. We show that this adaptation behaviour can be achieved in almost any feed-forward model by augmenting it with few standard layers and a simple new gradient reversal layer. The resulting augmented architecture can be trained using standard backpropagation. Overall, the approach can be implemented with little effort using any of the deep-learning packages. The method performs very well in a series of image classification experiments, achieving adaptation effect in the presence of big domain shifts and outperforming previous state-of-the-art on Office datasets.",
"Adversarial learning methods are a promising approach to training robust deep networks, and can generate complex samples across diverse domains. They can also improve recognition despite the presence of domain shift or dataset bias: recent adversarial approaches to unsupervised domain adaptation reduce the difference between the training and test domain distributions and thus improve generalization performance. However, while generative adversarial networks (GANs) show compelling visualizations, they are not optimal on discriminative tasks and can be limited to smaller shifts. On the other hand, discriminative approaches can handle larger domain shifts, but impose tied weights on the model and do not exploit a GAN-based loss. In this work, we first outline a novel generalized framework for adversarial adaptation, which subsumes recent state-of-the-art approaches as special cases, and use this generalized view to better relate prior approaches. We then propose a previously unexplored instance of our general framework which combines discriminative modeling, untied weight sharing, and a GAN loss, which we call Adversarial Discriminative Domain Adaptation (ADDA). We show that ADDA is more effective yet considerably simpler than competing domain-adversarial methods, and demonstrate the promise of our approach by exceeding state-of-the-art unsupervised adaptation results on standard domain adaptation tasks as well as a difficult cross-modality object classification task.",
"Recent reports suggest that a generic supervised deep CNN model trained on a large-scale dataset reduces, but does not remove, dataset bias. Fine-tuning deep models in a new domain can require a significant amount of labeled data, which for many applications is simply not available. We propose a new CNN architecture to exploit unlabeled and sparsely labeled target domain data. Our approach simultaneously optimizes for domain invariance to facilitate domain transfer and uses a soft label distribution matching loss to transfer information between tasks. Our proposed adaptation method offers empirical performance which exceeds previously published results on two standard benchmark visual domain adaptation tasks, evaluated across supervised and semi-supervised adaptation settings.",
"The cost of large scale data collection and annotation often makes the application of machine learning algorithms to new tasks or datasets prohibitively expensive. One approach circumventing this cost is training models on synthetic data where annotations are provided automatically. Despite their appeal, such models often fail to generalize from synthetic to real images, necessitating domain adaptation algorithms to manipulate these models before they can be successfully applied. Existing approaches focus either on mapping representations from one domain to the other, or on learning to extract features that are invariant to the domain from which they were extracted. However, by focusing only on creating a mapping or shared representation between the two domains, they ignore the individual characteristics of each domain. We hypothesize that explicitly modeling what is unique to each domain can improve a model's ability to extract domain-invariant features. Inspired by work on private-shared component analysis, we explicitly learn to extract image representations that are partitioned into two subspaces: one component which is private to each domain and one which is shared across domains. Our model is trained to not only perform the task we care about in the source domain, but also to use the partitioned representation to reconstruct the images from both domains. Our novel architecture results in a model that outperforms the state-of-the-art on a range of unsupervised domain adaptation scenarios and additionally produces visualizations of the private and shared representations enabling interpretation of the domain adaptation process.",
"Recent studies reveal that a deep neural network can learn transferable features which generalize well to novel tasks for domain adaptation. However, as deep features eventually transition from general to specific along the network, the feature transferability drops significantly in higher layers with increasing domain discrepancy. Hence, it is important to formally reduce the dataset bias and enhance the transferability in task-specific layers. In this paper, we propose a new Deep Adaptation Network (DAN) architecture, which generalizes deep convolutional neural network to the domain adaptation scenario. In DAN, hidden representations of all task-specific layers are embedded in a reproducing kernel Hilbert space where the mean embeddings of different domain distributions can be explicitly matched. The domain discrepancy is further reduced using an optimal multi-kernel selection method for mean embedding matching. DAN can learn transferable features with statistical guarantees, and can scale linearly by unbiased estimate of kernel embedding. Extensive empirical evidence shows that the proposed architecture yields state-of-the-art image classification error rates on standard domain adaptation benchmarks.",
"Recent reports suggest that a generic supervised deep CNN model trained on a large-scale dataset reduces, but does not remove, dataset bias on a standard benchmark. Fine-tuning deep models in a new domain can require a significant amount of data, which for many applications is simply not available. We propose a new CNN architecture which introduces an adaptation layer and an additional domain confusion loss, to learn a representation that is both semantically meaningful and domain invariant. We additionally show that a domain confusion metric can be used for model selection to determine the dimension of an adaptation layer and the best position for the layer in the CNN architecture. Our proposed adaptation method offers empirical performance which exceeds previously published results on a standard benchmark visual domain adaptation task."
]
} |
1905.10693 | 2946520073 | This paper presents a conceptually simple and effective Deep Audio-Visual Eembedding for dynamic saliency prediction dubbed DAVE". Several behavioral studies have shown a strong relation between auditory and visual cues for guiding gaze during scene free viewing. The existing video saliency models, however, only consider visual cues for predicting saliency over videos and neglect the auditory information that is ubiquitous in dynamic scenes. We propose a multimodal saliency model that utilizes audio and visual information for predicting saliency in videos. Our model consists of a two-stream encoder and a decoder. First, auditory and visual information are mapped into a feature space using 3D Convolutional Neural Networks (3D CNNs). Then, a decoder combines the features and maps them to a final saliency map. To train such model, data from various eye tracking datasets containing video and audio are pulled together. We further categorised videos into social', nature', and miscellaneous' classes to analyze the models over different content types. Several analyses show that our audio-visual model outperforms video-based models significantly over all scores; overall and over individual categories. Contextual analysis of the model performance over the location of sound source reveals that the audio-visual model behaves similar to humans in attending to the location of sound source. Our endeavour demonstrates that audio is an important signal that can boost video saliency prediction and help getting closer to human performance. | Humans are intelligent multi-sensory creatures, capable of spotting and focusing on a specific audio or visual stimulus in a cluttered environment ( , have attentional behavior). It is, hence, unsurprising that inspired by such observations psychologists and neuroscientists have been studying attentional mechanisms underlying auditory, visual, and auditory-visual attention. The recorded seminal ideas and works on attention can be traced back to the 19th century @cite_39 . Several decades of research on attention mechanisms have amassed a rich literature on the topic. Covering the whole literature is, thus, beyond the scope of this paper. Instead, to stress the need for multi-modal attention models, the following subsections provide a brief account of relevant studies along four axes. We refer the reader to @cite_40 @cite_8 @cite_9 @cite_5 for further information on these topics. | {
"cite_N": [
"@cite_8",
"@cite_9",
"@cite_39",
"@cite_40",
"@cite_5"
],
"mid": [
"119228284",
"",
"2751379301",
"2112845172",
"2563356726"
],
"abstract": [
"Although William James declared in 1890, \"Everyone knows what attention is,\" today there are many different and sometimes opposing views on the subject. This fragmented theoretical landscape may be because most of the theories and models of attention offer explanations in natural language or in a pictorial manner rather than providing a quantitative and unambiguous statement of the theory. They focus on the manifestations of attention instead of its rationale. In this book, John Tsotsos develops a formal model of visual attention with the goal of providing a theoretical explanation for why humans (and animals) must have the capacity to attend. He takes a unique approach to the theory, using the full breadth of the language of computation--rather than simply the language of mathematics--as the formal means of description. The result, the Selective Tuning model of vision and attention, explains attentive behavior in humans and provides a foundation for building computer systems that see with human-like characteristics. The overarching conclusion is that human vision is based on a general purpose processor that can be dynamically tuned to the task and the scene viewed on a moment-by-moment basis. Tsotsos offers a comprehensive, up-to-date overview of attention theories and models and a full description of the Selective Tuning model, confining the formal elements to two chapters and two appendixes. The text is accompanied by more than 100 illustrations in black and white and color; additional color illustrations and movies are available on the book's Web site",
"",
"Arguably the greatest single work in the history of psychology. James's analyses of habit, the nature of emotion, the phenomenology of attention, the stream of thought, the perception of space, and the multiplicity of the consciousness of self are still widely cited and incorporated into contemporary theoretical accounts of these phenomena.",
"This review focuses on covert attention and how it alters early vision. I explain why attention is considered a selective process, the constructs of covert attention, spatial endogenous and exogenous attention, and feature-based attention. I explain how in the last 25 years research on attention has characterized the effects of covert attention on spatial filters and how attention influences the selection of stimuli of interest. This review includes the effects of spatial attention on discriminability and appearance in tasks mediated by contrast sensitivity and spatial resolution; the effects of feature-based attention on basic visual processes, and a comparison of the effects of spatial and feature-based attention. The emphasis of this review is on psychophysical studies, but relevant electrophysiological and neuroimaging studies and models regarding how and where neuronal responses are modulated are also discussed.",
"Sounds in everyday life seldom appear in isolation. Both humans and machines are constantly flooded with a cacophony of sounds that need to be sorted through and scoured for relevant information—a phenomenon referred to as the ‘cocktail party problem’. A key component in parsing acoustic scenes is the role of attention, which mediates perception and behaviour by focusing both sensory and cognitive resources on pertinent information in the stimulus space. The current article provides a review of modelling studies of auditory attention. The review highlights how the term attention refers to a multitude of behavioural and cognitive processes that can shape sensory processing. Attention can be modulated by ‘bottom-up’ sensory-driven factors, as well as ‘top-down’ task-specific goals, expectations and learned schemas. Essentially, it acts as a selection process or processes that focus both sensory and cognitive resources on the most relevant events in the soundscape; with relevance being dictated by the stimulus itself (e.g. a loud explosion) or by a task at hand (e.g. listen to announcements in a busy airport). Recent computational models of auditory attention provide key insights into its role in facilitating perception in cluttered auditory scenes. This article is part of the themed issue ‘Auditory and visual scene analysis’."
]
} |
1905.10693 | 2946520073 | This paper presents a conceptually simple and effective Deep Audio-Visual Eembedding for dynamic saliency prediction dubbed DAVE". Several behavioral studies have shown a strong relation between auditory and visual cues for guiding gaze during scene free viewing. The existing video saliency models, however, only consider visual cues for predicting saliency over videos and neglect the auditory information that is ubiquitous in dynamic scenes. We propose a multimodal saliency model that utilizes audio and visual information for predicting saliency in videos. Our model consists of a two-stream encoder and a decoder. First, auditory and visual information are mapped into a feature space using 3D Convolutional Neural Networks (3D CNNs). Then, a decoder combines the features and maps them to a final saliency map. To train such model, data from various eye tracking datasets containing video and audio are pulled together. We further categorised videos into social', nature', and miscellaneous' classes to analyze the models over different content types. Several analyses show that our audio-visual model outperforms video-based models significantly over all scores; overall and over individual categories. Contextual analysis of the model performance over the location of sound source reveals that the audio-visual model behaves similar to humans in attending to the location of sound source. Our endeavour demonstrates that audio is an important signal that can boost video saliency prediction and help getting closer to human performance. | Auditory attention regards mechanisms of attention in the auditory system. A good example of such mechanisms is the famous cocktail party problem @cite_22 . The behavioral studies on the formation of auditory attention have been interested in the subject's response to an auditory stimulus, @cite_41 @cite_10 . From neurophysiological point of view, mechanisms of auditory attention and its influence on auditory cortical representations have been of interest, @cite_3 . | {
"cite_N": [
"@cite_41",
"@cite_10",
"@cite_22",
"@cite_3"
],
"mid": [
"1979753320",
"1986001465",
"1991139021",
"2082183045"
],
"abstract": [
"Abstract Groups of 32 children in each of three grades (kindergarten, second, and fourth) listened to a man's voice and a woman's voice, speaking words simultaneously. On one run through the 23 pairs of words, Ss were instructed to report what the man's voice was saying; on another, the words spoken by the woman. The stimulus words were systematically varied with respect to the number of syllables they contained. For cach S, the stimulus words were presented on one occasion binaurally (both words in both cars), on another occasion, dichotically (two different words in each ear). The number of correct reports of the word spoken by the desired voice increased with age; the number of intrusive errors (reports of words spoken by the unwanted voice) declined with age. At all age levels, scores were higher under dichotic presentation, but there was no change with age in the advantage derived from this presentation. Children at each age level made more errors on single-syllable than on multi-syllable words, but this difference was greater for older children. An analysis of errors revealed systematic differences among age groups in the nature of wrong guesses. Performanec improved through blocks of trials, at all age levels.",
"Abstract In 2 experiments, the development of auditory selective attention in children was assessed. The participants, aged 5, 7, and 9 years, responded to target stimuli in the left ear for 1 series (ignoring the standard stimuli) and in the right ear for the other series. In Experiment 1, event-related potentials (ERPs) evoked by the auditory stimuli were recorded from frontal, central, and pariet al sites. The 9-year-olds showed a greater processing negativity (Nd) to the attended channel compared with the 5-year-olds. Both 7-and 9-year-olds showed significantly larger amplitudes for the P3 component of ERPs in the attended vs. ignored condition. Behaviorally, the 5-year-olds made fewer hits and more false alarms than did older children, and the 9-year-olds made significantly more false alarms than did the 7-year-olds. The results of Experiment 2 showed that the detriment in the performance of the 9-year-olds was a result of task parameters. The inability of 5-year-olds to attend selectively appears to ...",
"This paper describes a number of objective experiments on recognition, concerning particularly the relation between the messages received by the two ears. Rather than use steady tones or clicks (frequency or time‐point signals) continuous speech is used, and the results interpreted in the main statistically. Two types of test are reported: (a) the behavior of a listener when presented with two speech signals simultaneously (statistical filtering problem) and (b) behavior when different speech signals are presented to his two ears.",
"The neural correlates of how attended speech is internally represented are described, shedding light on the ‘cocktail party problem’."
]
} |
1905.10693 | 2946520073 | This paper presents a conceptually simple and effective Deep Audio-Visual Eembedding for dynamic saliency prediction dubbed DAVE". Several behavioral studies have shown a strong relation between auditory and visual cues for guiding gaze during scene free viewing. The existing video saliency models, however, only consider visual cues for predicting saliency over videos and neglect the auditory information that is ubiquitous in dynamic scenes. We propose a multimodal saliency model that utilizes audio and visual information for predicting saliency in videos. Our model consists of a two-stream encoder and a decoder. First, auditory and visual information are mapped into a feature space using 3D Convolutional Neural Networks (3D CNNs). Then, a decoder combines the features and maps them to a final saliency map. To train such model, data from various eye tracking datasets containing video and audio are pulled together. We further categorised videos into social', nature', and miscellaneous' classes to analyze the models over different content types. Several analyses show that our audio-visual model outperforms video-based models significantly over all scores; overall and over individual categories. Contextual analysis of the model performance over the location of sound source reveals that the audio-visual model behaves similar to humans in attending to the location of sound source. Our endeavour demonstrates that audio is an important signal that can boost video saliency prediction and help getting closer to human performance. | Parallel to studies on auditory attention, numerous works have explored visual attention. The span of behavioral studies on visual attention is broad and covers a wide range of experiments on primates and humans. In such experiments, an observer is often presented with a stimulus and his neural and or behavioral responses are recorded ( using single unit recording, brain imaging, or eye tracking) @cite_40 . | {
"cite_N": [
"@cite_40"
],
"mid": [
"2112845172"
],
"abstract": [
"This review focuses on covert attention and how it alters early vision. I explain why attention is considered a selective process, the constructs of covert attention, spatial endogenous and exogenous attention, and feature-based attention. I explain how in the last 25 years research on attention has characterized the effects of covert attention on spatial filters and how attention influences the selection of stimuli of interest. This review includes the effects of spatial attention on discriminability and appearance in tasks mediated by contrast sensitivity and spatial resolution; the effects of feature-based attention on basic visual processes, and a comparison of the effects of spatial and feature-based attention. The emphasis of this review is on psychophysical studies, but relevant electrophysiological and neuroimaging studies and models regarding how and where neuronal responses are modulated are also discussed."
]
} |
1905.10693 | 2946520073 | This paper presents a conceptually simple and effective Deep Audio-Visual Eembedding for dynamic saliency prediction dubbed DAVE". Several behavioral studies have shown a strong relation between auditory and visual cues for guiding gaze during scene free viewing. The existing video saliency models, however, only consider visual cues for predicting saliency over videos and neglect the auditory information that is ubiquitous in dynamic scenes. We propose a multimodal saliency model that utilizes audio and visual information for predicting saliency in videos. Our model consists of a two-stream encoder and a decoder. First, auditory and visual information are mapped into a feature space using 3D Convolutional Neural Networks (3D CNNs). Then, a decoder combines the features and maps them to a final saliency map. To train such model, data from various eye tracking datasets containing video and audio are pulled together. We further categorised videos into social', nature', and miscellaneous' classes to analyze the models over different content types. Several analyses show that our audio-visual model outperforms video-based models significantly over all scores; overall and over individual categories. Contextual analysis of the model performance over the location of sound source reveals that the audio-visual model behaves similar to humans in attending to the location of sound source. Our endeavour demonstrates that audio is an important signal that can boost video saliency prediction and help getting closer to human performance. | With respect to multimodal attention, some behavioral studies have investigated the role of audio-visual components during sensory development in infants @cite_49 . Richards @cite_38 studied attention engagement to compound auditory-visual stimuli and showed an increase in sustained attention development with age progression in infants. Burg al @cite_37 conducted experiments to understand the effect of audiovisual events on guiding attention and concluded that audiovisual synchrony guides attention in an exogenous manner in adults. | {
"cite_N": [
"@cite_38",
"@cite_37",
"@cite_49"
],
"mid": [
"2128779052",
"1988712925",
"11812203"
],
"abstract": [
"This study examined the effect of attention engagement to compound auditory-visual stimuli on the modification of the startle blink reflex in infants. Infants at 8, 14, 20, or 26 weeks of age were presented with interesting audiovisual stimuli. After stimulus onset, at delays defined by heart rate changes known to be associated with sustained attention or attention disengagement, blink reflexes were elicited by visual or auditory stimuli. Blink amplitude to either visual or auditory stimuli was enhanced when the infants were engaged in attention to the foreground auditory-visual stimuli relative to control trials with no foreground patterns. This enhancement of the blink amplitude increased from 8 to 26 weeks of age. In contrast to selective modality enhancement for single-modality foreground stimuli, these results show that these multimodal stimuli engage both visual and auditory attention systems in this age range. Descriptors: Infants, Attention, Heart rate, Blink reflex, Multimodal stimuli",
"De volgorde waarin we 2 vrijwel tegelijkertijd weergegeven stipjes zien hangt af van de plek waarop onze aandacht gericht is: het stipje dat het dichtst bij de focus van de aandacht ligt wordt iets eerder waargenomen. In deze experimenten wordt dit effect gebruikt om aan te tonen dat een irrelevante stimulus onwillekeurig de aandacht trekt wanneer een geluid wordt weergegeven op het moment dat die stimulus van kleur verandert. Het bijzondere hierbij is dat het niet uitmaakt waar het geluid vandaan komt.",
"The present invention is used to make an excellent recording of color pictures and writings by a simple unit and easy operation wherein each set of ball-pen system recording part is provided for each restored primary color signal at receiving part for a color recording apparatus which is composed of transmitting and receiving parts, and a ball-pen is held pressed on a recording paper with some pressure corresponding to the strength of said primary color signal related. The present invention is able to apply to a facsimile."
]
} |
1905.10693 | 2946520073 | This paper presents a conceptually simple and effective Deep Audio-Visual Eembedding for dynamic saliency prediction dubbed DAVE". Several behavioral studies have shown a strong relation between auditory and visual cues for guiding gaze during scene free viewing. The existing video saliency models, however, only consider visual cues for predicting saliency over videos and neglect the auditory information that is ubiquitous in dynamic scenes. We propose a multimodal saliency model that utilizes audio and visual information for predicting saliency in videos. Our model consists of a two-stream encoder and a decoder. First, auditory and visual information are mapped into a feature space using 3D Convolutional Neural Networks (3D CNNs). Then, a decoder combines the features and maps them to a final saliency map. To train such model, data from various eye tracking datasets containing video and audio are pulled together. We further categorised videos into social', nature', and miscellaneous' classes to analyze the models over different content types. Several analyses show that our audio-visual model outperforms video-based models significantly over all scores; overall and over individual categories. Contextual analysis of the model performance over the location of sound source reveals that the audio-visual model behaves similar to humans in attending to the location of sound source. Our endeavour demonstrates that audio is an important signal that can boost video saliency prediction and help getting closer to human performance. | The computational modeling of auditory saliency is a relatively younger field than visual salience modeling. Some works have been inspired by visual saliency techniques. For example, @cite_27 adopts the model of @cite_26 to audio domain and produces auditory saliency maps. An auditory salience map identifies the presence of a salient sound over time. Some models are, however, rooted in the biology of the auditory system or extend saliency map idea for audio-based tasks. | {
"cite_N": [
"@cite_27",
"@cite_26"
],
"mid": [
"2156232986",
"2054802006"
],
"abstract": [
"Summary Our nervous system is confronted with a barrage of sensory stimuli, but neural resources are limited and not all stimuli can be processed to the same extent. Mechanisms exist to bias attention toward the particularly salient events, thereby providing a weighted representation of our environment [1]. Our understanding of these mechanisms is still limited, but theoretical models can replicate such a weighting of sensory inputs and provide a basis for understanding the underlying principles [2, 3]. Here, we describe such a model for the auditory system—an auditory saliency map. We experimentally validate the model on natural acoustical scenarios, demonstrating that it reproduces human judgments of auditory saliency and predicts the detectability of salient sounds embedded in noisy backgrounds. In addition, it also predicts the natural orienting behavior of naive macaque monkeys to the same salient stimuli. The structure of the suggested model is identical to that of successfully used visual saliency maps. Hence, we conclude that saliency is determined either by implementing similar mechanisms in different unisensory pathways or by the same mechanism in multisensory areas. In any case, our results demonstrate that different primate sensory systems rely on common principles for extracting relevant sensory events.",
"Most models of visual search, whether involving overt eye movements or covert shifts of attention, are based on the concept of a saliency map, that is, an explicit two-dimensional map that encodes the saliency or conspicuity of objects in the visual environment. Competition among neurons in this map gives rise to a single winning location that corresponds to the next attended target. Inhibiting this location automatically allows the system to attend to the next most salient location. We describe a detailed computer implementation of such a scheme, focusing on the problem of combining information across modalities, here orientation, intensity and color information, in a purely stimulus-driven manner. The model is applied to common psychophysical stimuli as well as to a very demanding visual search task. Its successful performance is used to address the extent to which the primate visual system carries out visual search via one or more such saliency maps and how this can be tested."
]
} |
1905.10693 | 2946520073 | This paper presents a conceptually simple and effective Deep Audio-Visual Eembedding for dynamic saliency prediction dubbed DAVE". Several behavioral studies have shown a strong relation between auditory and visual cues for guiding gaze during scene free viewing. The existing video saliency models, however, only consider visual cues for predicting saliency over videos and neglect the auditory information that is ubiquitous in dynamic scenes. We propose a multimodal saliency model that utilizes audio and visual information for predicting saliency in videos. Our model consists of a two-stream encoder and a decoder. First, auditory and visual information are mapped into a feature space using 3D Convolutional Neural Networks (3D CNNs). Then, a decoder combines the features and maps them to a final saliency map. To train such model, data from various eye tracking datasets containing video and audio are pulled together. We further categorised videos into social', nature', and miscellaneous' classes to analyze the models over different content types. Several analyses show that our audio-visual model outperforms video-based models significantly over all scores; overall and over individual categories. Contextual analysis of the model performance over the location of sound source reveals that the audio-visual model behaves similar to humans in attending to the location of sound source. Our endeavour demonstrates that audio is an important signal that can boost video saliency prediction and help getting closer to human performance. | Wrigley al @cite_25 proposed a model in which a network of neural oscillators performs auditory stream segregation on the basis of oscillatory correlation. To be concise, the model processes the audio input to simulate auditory nerve encoding. Periodicity information is extracted using a correlogram and noise segments are identified. This information is passed through an array of neural oscillators in which each oscillator corresponds to a particular frequency channel. The output from each oscillator is connected via excitatory links to a leaky integrator that decides which oscillator and in consequence input channel should be attended. | {
"cite_N": [
"@cite_25"
],
"mid": [
"2130788908"
],
"abstract": [
"The human auditory system is able to separate acoustic mixtures in order to create a perceptual description of each sound source. It has been proposed that this is achieved by an auditory scene analysis (ASA) in which a mixture of sounds is parsed to give a number of perceptual streams, each of which describes a single sound source. It is widely assumed that ASA is a precursor of attentional mechanisms, which select a stream for attentional focus. However, recent studies suggest that attention plays a key role in the formation of auditory streams. Motivated by these findings, this paper presents a conceptual framework for auditory selective attention in which the formation of groups and streams is heavily influenced by conscious and subconscious attention. This framework is implemented as a computational model comprising a network of neural oscillators, which perform stream segregation on the basis of oscillatory correlation. Within the network, attentional interest is modeled as a Gaussian distribution in frequency. This determines the connection weights between oscillators and the attentional process, which is modeled as an attentional leaky integrator (ALI). Acoustic features are held to be the subject of attention if their oscillatory activity coincides temporally with a peak in the ALI activity. The output of the model is an \"attentional stream,\" which encodes the frequency bands in the attentional focus at each epoch. The model successfully simulates a range of psychophysical phenomena."
]
} |
1905.10693 | 2946520073 | This paper presents a conceptually simple and effective Deep Audio-Visual Eembedding for dynamic saliency prediction dubbed DAVE". Several behavioral studies have shown a strong relation between auditory and visual cues for guiding gaze during scene free viewing. The existing video saliency models, however, only consider visual cues for predicting saliency over videos and neglect the auditory information that is ubiquitous in dynamic scenes. We propose a multimodal saliency model that utilizes audio and visual information for predicting saliency in videos. Our model consists of a two-stream encoder and a decoder. First, auditory and visual information are mapped into a feature space using 3D Convolutional Neural Networks (3D CNNs). Then, a decoder combines the features and maps them to a final saliency map. To train such model, data from various eye tracking datasets containing video and audio are pulled together. We further categorised videos into social', nature', and miscellaneous' classes to analyze the models over different content types. Several analyses show that our audio-visual model outperforms video-based models significantly over all scores; overall and over individual categories. Contextual analysis of the model performance over the location of sound source reveals that the audio-visual model behaves similar to humans in attending to the location of sound source. Our endeavour demonstrates that audio is an important signal that can boost video saliency prediction and help getting closer to human performance. | Oldoni al @cite_48 proposed an auditory attention model for designing better urban soundscapes. Their model is largely inspired by the FIT @cite_14 . The audio signal is converted to a 1 3-octave band spectrogram. Intensity, spectral contrast, and temporal contrast are computed from the spectrogram. This process corresponds to the peripheral auditory processing and results in a feature vector that forms a saliency map, which will be used to identify the sounds that will be heard. While having some commonalities with vision based models of attention in the first layer, this model goes beyond saliency map approaches and extends such models to task-based top-down driven attention. | {
"cite_N": [
"@cite_48",
"@cite_14"
],
"mid": [
"2054750616",
"2149095485"
],
"abstract": [
"Urban soundscape design involves creating outdoor spaces that are pleasing to the ear. One way to achieve this goal is to add or accentuate sounds that are considered to be desired by most users of the space, such that the desired sounds mask undesired sounds, or at least distract attention away from undesired sounds. In view of removing the need for a listening panel to assess the effectiveness of such soundscape measures, the interest for new models and techniques is growing. In this paper, a model of auditory attention to environmental sound is presented, which balances computational complexity and biological plausibility. Once the model is trained for a particular location, it classifies the sounds that are present in the soundscape and simulates how a typical listener would switch attention over time between different sounds. The model provides an acoustic summary, giving the soundscape designer a quick overview of the typical sounds at a particular location, and allows assessment of the perceptual e...",
"A new hypothesis about the role of focused attention is proposed. The feature-integration theory of attention suggests that attention must be directed serially to each stimulus in a display whenever conjunctions of more than one separable feature are needed to characterize or distinguish the possible objects presented. A number of predictions were tested in a variety of paradigms including visual search, texture segregation, identification and localization, and using both separable dimensions (shape and color) and local elements or parts of figures (lines, curves, etc. in letters) as the features to be integrated into complex wholes. The results were in general consistent with the hypothesis. They offer a new set of criteria for distinguishing separable from integral features and a new rationale for predicting which tasks will show attention limits and which will not."
]
} |
1905.10693 | 2946520073 | This paper presents a conceptually simple and effective Deep Audio-Visual Eembedding for dynamic saliency prediction dubbed DAVE". Several behavioral studies have shown a strong relation between auditory and visual cues for guiding gaze during scene free viewing. The existing video saliency models, however, only consider visual cues for predicting saliency over videos and neglect the auditory information that is ubiquitous in dynamic scenes. We propose a multimodal saliency model that utilizes audio and visual information for predicting saliency in videos. Our model consists of a two-stream encoder and a decoder. First, auditory and visual information are mapped into a feature space using 3D Convolutional Neural Networks (3D CNNs). Then, a decoder combines the features and maps them to a final saliency map. To train such model, data from various eye tracking datasets containing video and audio are pulled together. We further categorised videos into social', nature', and miscellaneous' classes to analyze the models over different content types. Several analyses show that our audio-visual model outperforms video-based models significantly over all scores; overall and over individual categories. Contextual analysis of the model performance over the location of sound source reveals that the audio-visual model behaves similar to humans in attending to the location of sound source. Our endeavour demonstrates that audio is an important signal that can boost video saliency prediction and help getting closer to human performance. | Boccignone al @cite_24 follow a similar path for automating saliency prediction in social interaction scenarios. They extract several priority maps, including (1) spatio-temporal saliency features using @cite_2 , (2) face maps from a convolutional neural network (CNN) face detector, and (3) the active speaker map from automatic lip sync algorithm of @cite_12 . Once the maps are extracted, a sampling scheme is employed to find attention attractors. Instead of relying on a complicated sampling scheme and multiple feature maps, we directly learn the mapping using a deep neural network. | {
"cite_N": [
"@cite_24",
"@cite_12",
"@cite_2"
],
"mid": [
"2913121856",
"2604379605",
"2034436892"
],
"abstract": [
"We address the deployment of perceptual attention to social interactions as displayed in conversational clips, when relying on multimodal information (audio and video). A probabilistic modelling framework is proposed that goes beyond the classic saliency paradigm while integrating multiple information cues. Attentional allocation is determined not just by stimulus-driven selection but, importantly, by social value as modulating the selection history of relevant multimodal items. Thus, the construction of attentional priority is the result of a sampling procedure conditioned on the potential value dynamics of socially relevant objects emerging moment to moment within the scene. Preliminary experiments on a publicly available dataset are presented.",
"The goal of this work is to determine the audio-video synchronisation between mouth motion and speech in a video.",
"We present a novel unified framework for both static and space -time saliency detection. Our method is a bottom-up approach and computes so-called local regression kernels (i.e., local descriptors) from the given image (or a video), which measure the likeness of a pixel (or voxel) to its surroundings. Visual saliency is then computed using the said “self-resemblance” measure. The framework results in a saliency map where each pixel (or voxel) indicates the statistical li kelihood of saliency of a feature matrix given its surrounding feature matrices. As a similarity measure, matrix cosine similarity (a generalization of cosine similarity) is employed. State of the art performance is demonstrated on commonly used human eye fixation data (static scenes [5] and dynamic scenes [16]) and some psychological patterns."
]
} |
1905.10729 | 2946455873 | Machine learning models are vulnerable to adversarial examples. Iterative adversarial training has shown promising results against strong white-box attacks. However, adversarial training is very expensive, and every time a model needs to be protected, such expensive training scheme needs to be performed. In this paper, we propose to apply iterative adversarial training scheme to an external auto-encoder, which once trained can be used to protect other models directly. We empirically show that our model outperforms other purifying-based methods against white-box attacks, and transfers well to directly protect other base models with different architectures. | Our work builds upon iterative adversarial training. Adversarial training @cite_15 protects a model against adversarial perturbation by augmenting training data with adversarial examples. Original adversarial training generates the adversarial examples once before training. As a result the trained model is still vulnerable to white-box attacks. Iterative adversarial training @cite_13 @cite_5 generates adversarial examples during each iteration of training, and is shown to be successful against white-box attacks. However, such iterative training scheme introduces much computational overhead. To make things worse, current adversarial training cannot protect a pre-trained models without modifying it, thus such computational overhead needs to be paid whenever a model needs to be hardened. Ensemble adversarial training @cite_9 augments the training data with adversarial examples generated on other pre-trained models. Like adversarial training, this technique trains the base classifier, thus cannot be transferred to protect other models. Feature denoising @cite_19 also uses adversarial training to enhance white-box robustness. But they integrate denoise blocks between the convolutional layers in the base model, and thus is model specific and lack of transferability. | {
"cite_N": [
"@cite_9",
"@cite_19",
"@cite_5",
"@cite_15",
"@cite_13"
],
"mid": [
"2620038827",
"2903785932",
"2640329709",
"1945616565",
"2552767274"
],
"abstract": [
"Adversarial examples are perturbed inputs designed to fool machine learning models. Adversarial training injects such examples into training data to increase robustness. To scale this technique to large datasets, perturbations are crafted using fast single-step methods that maximize a linear approximation of the model's loss. We show that this form of adversarial training converges to a degenerate global minimum, wherein small curvature artifacts near the data points obfuscate a linear approximation of the loss. The model thus learns to generate weak perturbations, rather than defend against strong ones. As a result, we find that adversarial training remains vulnerable to black-box attacks, where we transfer perturbations computed on undefended models, as well as to a powerful novel single-step attack that escapes the non-smooth vicinity of the input data via a small random step. We further introduce Ensemble Adversarial Training, a technique that augments training data with perturbations transferred from other models. On ImageNet, Ensemble Adversarial Training yields models with strong robustness to black-box attacks. In particular, our most robust model won the first round of the NIPS 2017 competition on Defenses against Adversarial Attacks.",
"Adversarial attacks to image classification systems present challenges to convolutional networks and opportunities for understanding them. This study suggests that adversarial perturbations on images lead to noise in the features constructed by these networks. Motivated by this observation, we develop new network architectures that increase adversarial robustness by performing feature denoising. Specifically, our networks contain blocks that denoise the features using non-local means or other filters; the entire networks are trained end-to-end. When combined with adversarial training, our feature denoising networks substantially improve the state-of-the-art in adversarial robustness in both white-box and black-box attack settings. On ImageNet, under 10-iteration PGD white-box attacks where prior art has 27.9 accuracy, our method achieves 55.7 ; even under extreme 2000-iteration PGD white-box attacks, our method secures 42.6 accuracy. Our method was ranked first in Competition on Adversarial Attacks and Defenses (CAAD) 2018 --- it achieved 50.6 classification accuracy on a secret, ImageNet-like test dataset against 48 unknown attackers, surpassing the runner-up approach by 10 . Code is available at this https URL.",
"Recent work has demonstrated that neural networks are vulnerable to adversarial examples, i.e., inputs that are almost indistinguishable from natural data and yet classified incorrectly by the network. In fact, some of the latest findings suggest that the existence of adversarial attacks may be an inherent weakness of deep learning models. To address this problem, we study the adversarial robustness of neural networks through the lens of robust optimization. This approach provides us with a broad and unifying view on much of the prior work on this topic. Its principled nature also enables us to identify methods for both training and attacking neural networks that are reliable and, in a certain sense, universal. In particular, they specify a concrete security guarantee that would protect against any adversary. These methods let us train networks with significantly improved resistance to a wide range of adversarial attacks. They also suggest the notion of security against a first-order adversary as a natural and broad security guarantee. We believe that robustness against such well-defined classes of adversaries is an important stepping stone towards fully resistant deep learning models.",
"Several machine learning models, including neural networks, consistently misclassify adversarial examples---inputs formed by applying small but intentionally worst-case perturbations to examples from the dataset, such that the perturbed input results in the model outputting an incorrect answer with high confidence. Early attempts at explaining this phenomenon focused on nonlinearity and overfitting. We argue instead that the primary cause of neural networks' vulnerability to adversarial perturbation is their linear nature. This explanation is supported by new quantitative results while giving the first explanation of the most intriguing fact about them: their generalization across architectures and training sets. Moreover, this view yields a simple and fast method of generating adversarial examples. Using this approach to provide examples for adversarial training, we reduce the test set error of a maxout network on the MNIST dataset.",
"Adversarial examples are malicious inputs designed to fool machine learning models. They often transfer from one model to another, allowing attackers to mount black box attacks without knowledge of the target model's parameters. Adversarial training is the process of explicitly training a model on adversarial examples, in order to make it more robust to attack or to reduce its test error on clean inputs. So far, adversarial training has primarily been applied to small problems. In this research, we apply adversarial training to ImageNet. Our contributions include: (1) recommendations for how to succesfully scale adversarial training to large models and datasets, (2) the observation that adversarial training confers robustness to single-step attack methods, (3) the finding that multi-step attack methods are somewhat less transferable than single-step attack methods, so single-step attacks are the best for mounting black-box attacks, and (4) resolution of a \"label leaking\" effect that causes adversarially trained models to perform better on adversarial examples than on clean examples, because the adversarial example construction process uses the true label and the model can learn to exploit regularities in the construction process."
]
} |
1905.10729 | 2946455873 | Machine learning models are vulnerable to adversarial examples. Iterative adversarial training has shown promising results against strong white-box attacks. However, adversarial training is very expensive, and every time a model needs to be protected, such expensive training scheme needs to be performed. In this paper, we propose to apply iterative adversarial training scheme to an external auto-encoder, which once trained can be used to protect other models directly. We empirically show that our model outperforms other purifying-based methods against white-box attacks, and transfers well to directly protect other base models with different architectures. | There has been a line of work of purifying-based approach where the defense mechanism purifies the adversarial examples into benign ones, without modifying the underlying models being protected. One kind of such work is by adding denoisers, e.g. MagNet @cite_11 and HGD @cite_6 . However, as shown by @cite_36 @cite_40 , none of them achieve white-box security. In particular, HGD by @cite_6 is very similar to us, in the sense that they also add auto-encoders that are trained using adversarial examples. However, same as original adversarial training, the adversarial examples are pre-generated against the base model. As a result, the model is still vulnerable to white-box attacks @cite_20 . Defense-VAE @cite_27 uses VAE instead of denoising auto-encoders. Same as HGD, Defense-VAE pre-generates adversarial examples, and does not claim white-box security, but oblivious one. | {
"cite_N": [
"@cite_36",
"@cite_6",
"@cite_40",
"@cite_27",
"@cite_20",
"@cite_11"
],
"mid": [
"2768899812",
"2774018344",
"",
"2903580494",
"2797455600",
"2618043096"
],
"abstract": [
"MagNet and \"Efficient Defenses...\" were recently proposed as a defense to adversarial examples. We find that we can construct adversarial examples that defeat these defenses with only a slight increase in distortion.",
"Neural networks are vulnerable to adversarial examples, which poses a threat to their application in security sensitive systems. We propose high-level representation guided denoiser (HGD) as a defense for image classification. Standard denoiser suffers from the error amplification effect, in which small residual adversarial noise is progressively amplified and leads to wrong classifications. HGD overcomes this problem by using a loss function defined as the difference between the target model's outputs activated by the clean image and denoised image. Compared with ensemble adversarial training which is the state-of-the-art defending method on large images, HGD has three advantages. First, with HGD as a defense, the target model is more robust to either white-box or black-box adversarial attacks. Second, HGD can be trained on a small subset of the images and generalizes well to other images and unseen classes. Third, HGD can be transferred to defend models other than the one guiding it. In NIPS competition on defense against adversarial attacks, our HGD solution won the first place and outperformed other models by a large margin.1",
"",
"Deep neural networks (DNNs) have been enormously successful across a variety of prediction tasks. However, recent research shows that DNNs are particularly vulnerable to adversarial attacks, which poses a serous threat to their applications in security-sensitive systems. In this paper, we propose a simple yet effective defense algorithm Defense-VAE that uses variational autoencoder (VAE) to purge adversarial perturbations from contaminated images. The proposed method is generic and can defend white-box and black-box attacks without the need of retraining the original CNN classifiers, and can further strengthen the defense by retraining CNN or end-to-end finetuning the whole pipeline. In addition, the proposed method is very efficient compared to the optimization-based alternatives, such as Defense-GAN, since no iterative optimization is needed for online prediction. Extensive experiments on MNIST, Fashion-MNIST, CelebA and CIFAR-10 demonstrate the superior defense accuracy of Defense-VAE compared to Defense-GAN, while being 50x faster than the latter. This makes Defense-VAE widely deployable in real-time security-sensitive systems. We plan to open source our implementation to facilitate the research in this area.",
"Neural networks are known to be vulnerable to adversarial examples. In this note, we evaluate the two white-box defenses that appeared at CVPR 2018 and find they are ineffective: when applying existing techniques, we can reduce the accuracy of the defended models to 0 .",
"Deep learning has shown impressive performance on hard perceptual problems. However, researchers found deep learning systems to be vulnerable to small, specially crafted perturbations that are imperceptible to humans. Such perturbations cause deep learning systems to mis-classify adversarial examples, with potentially disastrous consequences where safety or security is crucial. Prior defenses against adversarial examples either targeted specific attacks or were shown to be ineffective. We propose MagNet, a framework for defending neural network classifiers against adversarial examples. MagNet neither modifies the protected classifier nor requires knowledge of the process for generating adversarial examples. MagNet includes one or more separate detector networks and a reformer network. The detector networks learn to differentiate between normal and adversarial examples by approximating the manifold of normal examples. Since they assume no specific process for generating adversarial examples, they generalize well. The reformer network moves adversarial examples towards the manifold of normal examples, which is effective for correctly classifying adversarial examples with small perturbation. We discuss the intrinsic difficulties in defending against whitebox attack and propose a mechanism to defend against graybox attack. Inspired by the use of randomness in cryptography, we use diversity to strengthen MagNet. We show empirically that MagNet is effective against the most advanced state-of-the-art attacks in blackbox and graybox scenarios without sacrificing false positive rate on normal examples."
]
} |
1905.10828 | 2944848738 | This article describes an absolutely stable, first-order constraint solverfor multi-rigid body systems that calculates (predicts) constraint forces for typical bilateral and unilateral constraints, contact constraints with friction, and many other constraint types. Redundant constraints do not pose numerical problems or require regularization. Coulomb friction for contact is modeled using a true friction cone, rather than a linearized approximation. The computational expense of the solver is dependent upon the types of constraints present in the input. The hardest (in a computational complexity sense) inputs are reducible to solving convex optimization problems, i.e., polynomial time solvable. The simplest inputs require only solving a linear system. The solver is L-stable, which will imply that the forces due to constraints induce no computational stiffness into the multi-body dynamics differential equations. This approach is targeted to multibodies simulated with coarse accuracy, subject to computational stiffness arising from constraints, and where the number of constraint equations is not large compared to the number of multibody position and velocity state variables. For such applications, the approach should prove far faster than using other implicit integration approaches. I assess the approach on some fundamental multibody dynamics problems. | The contact aspects of the constraint model introduced in this article are effectively that of the @cite_25 contact model, though I introduce numerous computational and modeling advantages over the approach they describe. Mine does not require categorizing contact into sticking and sliding. My approach permits using realistic stiffness values derived via engineering tables and without needing to consider whether the choice of modeling parameters might make the simulation unstable. | {
"cite_N": [
"@cite_25"
],
"mid": [
"2040973163"
],
"abstract": [
"The use of Coulomb's friction law with the principles of classical rigid-body dynamics introduces mathematical inconsistencies. Specifically, the forward dynamics problem can have no solutions or multiple solutions. In these situations, compliant contact models, while increasing the dimensionality of the state vector, can resolve these problems. The simplicity and efficiency of rigid-body models, however, provide strong motivation for their use during those portions of a simulation when the rigid-body solution is unique and stable. In this paper, we use singular perturbation analysis in conjunction with linear complementarity theory to establish conditions under which the solution predicted by the rigid-body dynamic model is stable. We employ a general model of contact compliance to derive stability criteria for planar mechanical systems. In particular, we show that for cases with one sliding contact, there is always at most one stable solution. Our approach is not directly applicable to transitions between rolling and sliding where the Coulomb friction law is discontinuous. To overcome this difficulty, we introduce a smooth nonlinear friction law, which approximates Coulomb friction. Such a friction model can also increase the efficiency of both rigid-body and compliant contact simulation. Numerical simulations for the different models and comparison with experimental results are also presented."
]
} |
1905.10828 | 2944848738 | This article describes an absolutely stable, first-order constraint solverfor multi-rigid body systems that calculates (predicts) constraint forces for typical bilateral and unilateral constraints, contact constraints with friction, and many other constraint types. Redundant constraints do not pose numerical problems or require regularization. Coulomb friction for contact is modeled using a true friction cone, rather than a linearized approximation. The computational expense of the solver is dependent upon the types of constraints present in the input. The hardest (in a computational complexity sense) inputs are reducible to solving convex optimization problems, i.e., polynomial time solvable. The simplest inputs require only solving a linear system. The solver is L-stable, which will imply that the forces due to constraints induce no computational stiffness into the multi-body dynamics differential equations. This approach is targeted to multibodies simulated with coarse accuracy, subject to computational stiffness arising from constraints, and where the number of constraint equations is not large compared to the number of multibody position and velocity state variables. For such applications, the approach should prove far faster than using other implicit integration approaches. I assess the approach on some fundamental multibody dynamics problems. | A rigid [point] contact'', i.e., a point of contact between two completely rigid bodies, corresponds to a unilateral rigid constraint and several more equations that describe the frictional interaction at that point (see, e.g., @cite_19 ). Rigid contact makes reasoning about the contact geometry simple, but the problems from redundant forces described above (also called indeterminacy'') remain. The conceptual problem of inconsistent configurations'', like Painlev ' e 's famous paradox (described in detail in @cite_11 ) emerges with rigid contact also. And rigid contact is non-smooth: force is discontinuous when two bodies first contact, for example. In addition to these challenges that arise from the mathematics (and likely the real physics as well, see, e.g. @cite_20 ), the computational demands of implementing such models are significant. | {
"cite_N": [
"@cite_19",
"@cite_20",
"@cite_11"
],
"mid": [
"2007677977",
"",
"2163862895"
],
"abstract": [
"This paper is a summary of a comprehensive study of the problem of predicting the possible acceleration(s) of a set of rigid, three-dimensional bodies in contact in the presence of Coulomb friction. We begin with a brief introduction to this problem and a survey of related work and previous approaches. This is followed by the introduction of two novel complementarity formulations for the contact problem under two friction laws: Coulomb''s Law and an analogous law in which Coulomb''s quadratic friction cone is approximated by a pyramid. Under a full co umn rank assumption on the system Jacobian matrix, we establish the existence and uniqueness of a solution to our new models in the case where the friction coefficients are nonnegative and sufficiently small. For the model based on the friction pyramid law, we also show that the classical Lemke almost-complementary pivot algorithm and our new feasible interior point method are guaranteed to compute a solution. Extensive computational results are presented to demonstrate the effectiveness and characteristics of these solution methods for solving the model with friction pyramids. Finally, results are summarized and conclusions are drawn.",
"",
"Rigid-body dynamics with unilateral contact is a good approximation for a wide range of everyday phenomena, from the operation of car brakes to walking to rock slides. It is also of vital importance for simulating robots, virtual reality, and realistic animation. However, correctly modeling rigid-body dynamics with friction is difficult due to a number of discontinuities in the behavior of rigid bodies and the discontinuities inherent in the Coulomb friction law. This is particularly crucial for handling situations with large coefficients of friction, which can result in paradoxical results known at least since Painleve [C. R. Acad. Sci. Paris, 121 (1895), pp. 112--115]. This single example has been a counterexample and cause of controversy ever since, and only recently have there been rigorous mathematical results that show the existence of solutions to his example. The new mathematical developments in rigid-body dynamics have come from several sources: \"sweeping processes\" and the measure differential inclusions of Moreau in the 1970s and 1980s, the variational inequality approaches of Duvaut and J.-L. Lions in the 1970s, and the use of complementarity problems to formulate frictional contact problems by Lotstedt in the early 1980s. However, it wasn't until much more recently that these tools were finally able to produce rigorous results about rigid-body dynamics with Coulomb friction and impulses."
]
} |
1905.10479 | 2946660403 | In this effort we propose a new deep architecture utilizing residual blocks inspired by implicit discretization schemes. As opposed to the standard feed-forward networks, the outputs of the proposed implicit residual blocks are defined as the fixed points of the appropriately chosen nonlinear transformations. We show that this choice leads to improved stability of both forward and backward propagations, has a favorable impact on the generalization power of the network and allows for higher learning rates. In addition, we consider a reformulation of ResNet which does not introduce new parameters and can potentially lead to a reduction in the number of required layers due to improved forward stability and robustness. Finally, we derive the memory efficient reversible training algorithm and provide numerical results in support of our findings. | The authors of @cite_13 concentrated on the well-posedness of the learning problem for ODE-constrained control and emphasized the importance of stability in the design of deep architectures. For instance, the solution of a homogeneous linear ODE with constant coefficients is given by where @math is the eigen-decomposition of a matrix @math , and @math is the diagonal matrix with corresponding eigenvalues. Similar equation holds for the backpropagation of gradients. To guarantee the efficient propagation of information through the network, one must ensure that the elements of @math have magnitudes close to one. This condition, of course, is satisfied when all eigenvalues of the matrix @math are imaginary with real parts close to zero. In order to preserve this property, the authors of @cite_13 proposed several time continuous architectures of the form When @math , @math , the equations above provide an example of a conservative Hamiltonian system with the total energy @math . | {
"cite_N": [
"@cite_13"
],
"mid": [
"2612252966"
],
"abstract": [
"Deep neural networks have become invaluable tools for supervised machine learning, e.g. classification of text or images. While often offering superior results over traditional techniques and successfully expressing complicated patterns in data, deep architectures are known to be challenging to design and train such that they generalize well to new data. Critical issues with deep architectures are numerical instabilities in derivative-based learning algorithms commonly called exploding or vanishing gradients. In this paper, we propose new forward propagation techniques inspired by systems of ordinary differential equations (ODE) that overcome this challenge and lead to well-posed learning problems for arbitrarily deep networks. The backbone of our approach is our interpretation of deep learning as a parameter estimation problem of nonlinear dynamical systems. Given this formulation, we analyze stability and well-posedness of deep learning and use this new understanding to develop new network architectures. We relate the exploding and vanishing gradient phenomenon to the stability of the discrete ODE and present several strategies for stabilizing deep learning for very deep networks. While our new architectures restrict the solution space, several numerical experiments show their competitiveness with state-of-the-art networks."
]
} |
1905.10479 | 2946660403 | In this effort we propose a new deep architecture utilizing residual blocks inspired by implicit discretization schemes. As opposed to the standard feed-forward networks, the outputs of the proposed implicit residual blocks are defined as the fixed points of the appropriately chosen nonlinear transformations. We show that this choice leads to improved stability of both forward and backward propagations, has a favorable impact on the generalization power of the network and allows for higher learning rates. In addition, we consider a reformulation of ResNet which does not introduce new parameters and can potentially lead to a reduction in the number of required layers due to improved forward stability and robustness. Finally, we derive the memory efficient reversible training algorithm and provide numerical results in support of our findings. | Memory efficient explicit reversible architectures can be obtained by considering time discretization of the partitioned system of ODEs in . The reversibility property allows to recover the internal states of the system by propagating through the network in both directions and thus does not require one to cache these values for the evaluation of the gradients. First, such architecture (RevNet) was proposed in @cite_10 , and without using a connection to discrete solutions of ODEs, it has the form . . It was later recognized as the Verlet method applied to the particular form of the system in , see @cite_13 @cite_11 . The leapfrog and midpoint networks are two other examples of reversible architectures proposed in @cite_11 . | {
"cite_N": [
"@cite_13",
"@cite_10",
"@cite_11"
],
"mid": [
"2612252966",
"2963684275",
"2963359731"
],
"abstract": [
"Deep neural networks have become invaluable tools for supervised machine learning, e.g. classification of text or images. While often offering superior results over traditional techniques and successfully expressing complicated patterns in data, deep architectures are known to be challenging to design and train such that they generalize well to new data. Critical issues with deep architectures are numerical instabilities in derivative-based learning algorithms commonly called exploding or vanishing gradients. In this paper, we propose new forward propagation techniques inspired by systems of ordinary differential equations (ODE) that overcome this challenge and lead to well-posed learning problems for arbitrarily deep networks. The backbone of our approach is our interpretation of deep learning as a parameter estimation problem of nonlinear dynamical systems. Given this formulation, we analyze stability and well-posedness of deep learning and use this new understanding to develop new network architectures. We relate the exploding and vanishing gradient phenomenon to the stability of the discrete ODE and present several strategies for stabilizing deep learning for very deep networks. While our new architectures restrict the solution space, several numerical experiments show their competitiveness with state-of-the-art networks.",
"Residual Networks (ResNets) have demonstrated significant improvement over traditional Convolutional Neural Networks (CNNs) on image classification, increasing in performance as networks grow both deeper and wider. However, memory consumption becomes a bottleneck as one needs to store all the intermediate activations for calculating gradients using backpropagation. In this work, we present the Reversible Residual Network (RevNet), a variant of ResNets where each layer's activations can be reconstructed exactly from the next layer's. Therefore, the activations for most layers need not be stored in memory during backprop. We demonstrate the effectiveness of RevNets on CIFAR and ImageNet, establishing nearly identical performance to equally-sized ResNets, with activation storage requirements independent of depth.",
""
]
} |
1905.10479 | 2946660403 | In this effort we propose a new deep architecture utilizing residual blocks inspired by implicit discretization schemes. As opposed to the standard feed-forward networks, the outputs of the proposed implicit residual blocks are defined as the fixed points of the appropriately chosen nonlinear transformations. We show that this choice leads to improved stability of both forward and backward propagations, has a favorable impact on the generalization power of the network and allows for higher learning rates. In addition, we consider a reformulation of ResNet which does not introduce new parameters and can potentially lead to a reduction in the number of required layers due to improved forward stability and robustness. Finally, we derive the memory efficient reversible training algorithm and provide numerical results in support of our findings. | Other residual architectures can be also found in the literature including Resnet in Resnet (RiR) @cite_17 , Dense Convolutional Network (DenseNet) @cite_15 and linearly implicit network (IMEXNet) @cite_0 . For some problems, all of these networks show a substantial improvement over the classical ResNet but still have an explicit structure, which has limited robustness to the perturbations of the input data and parameters of the network. Instead, in this effort we propose new fully implicit residual architecture which, unlike the above mentioned examples, is unconditionally stable and robust. As opposed to the standard feed-forward networks, the outputs of the proposed implicit residual blocks are defined as the fixed points of the appropriately chosen nonlinear transformations as follows: The right part of Figure provides a graphical illustration of the proposed layer. The choice of the nonlinear transformation @math and the design of the learning algorithm are discussed in the next section. | {
"cite_N": [
"@cite_0",
"@cite_15",
"@cite_17"
],
"mid": [
"2920348561",
"2963446712",
"2333796428"
],
"abstract": [
"Deep convolutional neural networks have revolutionized many machine learning and computer vision tasks, however, some remaining key challenges limit their wider use. These challenges include improving the network's robustness to perturbations of the input image and the limited field of view'' of convolution operators. We introduce the IMEXnet that addresses these challenges by adapting semi-implicit methods for partial differential equations. Compared to similar explicit networks, such as residual networks, our network is more stable, which has recently shown to reduce the sensitivity to small changes in the input features and improve generalization. The addition of an implicit step connects all pixels in each channel of the image and therefore addresses the field of view problem while still being comparable to standard convolutions in terms of the number of parameters and computational complexity. We also present a new dataset for semantic segmentation and demonstrate the effectiveness of our architecture using the NYU Depth dataset.",
"Recent work has shown that convolutional networks can be substantially deeper, more accurate, and efficient to train if they contain shorter connections between layers close to the input and those close to the output. In this paper, we embrace this observation and introduce the Dense Convolutional Network (DenseNet), which connects each layer to every other layer in a feed-forward fashion. Whereas traditional convolutional networks with L layers have L connections—one between each layer and its subsequent layer—our network has L(L+1) 2 direct connections. For each layer, the feature-maps of all preceding layers are used as inputs, and its own feature-maps are used as inputs into all subsequent layers. DenseNets have several compelling advantages: they alleviate the vanishing-gradient problem, strengthen feature propagation, encourage feature reuse, and substantially reduce the number of parameters. We evaluate our proposed architecture on four highly competitive object recognition benchmark tasks (CIFAR-10, CIFAR-100, SVHN, and ImageNet). DenseNets obtain significant improvements over the state-of-the-art on most of them, whilst requiring less memory and computation to achieve high performance. Code and pre-trained models are available at https: github.com liuzhuang13 DenseNet.",
"Residual networks (ResNets) have recently achieved state-of-the-art on challenging computer vision tasks. We introduce Resnet in Resnet (RiR): a deep dual-stream architecture that generalizes ResNets and standard CNNs and is easily implemented with no computational overhead. RiR consistently improves performance over ResNets, outperforms architectures with similar amounts of augmentation on CIFAR-10, and establishes a new state-of-the-art on CIFAR-100."
]
} |
1905.10649 | 2946500988 | We consider Markov Decision Processes (MDPs) where the rewards are unknown and may change in an adversarial manner. We provide an algorithm that achieves state-of-the-art regret bound of @math , where @math is the state space, @math is the action space, @math is the mixing time of the MDP, and @math is the number of periods. The algorithm's computational complexity is polynomial in @math and @math per period. We then consider a setting often encountered in practice, where the state space of the MDP is too large to allow for exact solutions. By approximating the state-action occupancy measures with a linear architecture of dimension @math , we propose a modified algorithm with computational complexity polynomial in @math . We also prove a regret bound for this modified algorithm, which to the best of our knowledge this is the first @math regret bound for large scale MDPs with changing rewards. | The history of MDPs goes back to the seminal work of Bellman @cite_11 and Howard @cite_32 from the 1950's. Some classic algorithms for solving MDPS include policy iteration, value iteration, policy gradient, Q-learning and their approximate versions (see @cite_33 @cite_23 @cite_24 for an excellent discussion). In this paper, we will focus on a relatively less used approach, which is based on finding the using linear programming, as done recently in @cite_0 @cite_13 @cite_34 to solve MDPs with rewards (see more details in Section ). To deal with the curse of dimensionality, @cite_0 uses bilinear functions to approximate the occupancy measures and @cite_34 uses a linear approximation. | {
"cite_N": [
"@cite_33",
"@cite_32",
"@cite_24",
"@cite_0",
"@cite_23",
"@cite_34",
"@cite_13",
"@cite_11"
],
"mid": [
"2119567691",
"2028145673",
"",
"2798543980",
"",
"2909648661",
"2765892966",
"2076337359"
],
"abstract": [
"From the Publisher: The past decade has seen considerable theoretical and applied research on Markov decision processes, as well as the growing use of these models in ecology, economics, communications engineering, and other fields where outcomes are uncertain and sequential decision-making processes are needed. A timely response to this increased activity, Martin L. Puterman's new work provides a uniquely up-to-date, unified, and rigorous treatment of the theoretical, computational, and applied research on Markov decision process models. It discusses all major research directions in the field, highlights many significant applications of Markov decision processes models, and explores numerous important topics that have previously been neglected or given cursory coverage in the literature. Markov Decision Processes focuses primarily on infinite horizon discrete time models and models with discrete time spaces while also examining models with arbitrary state spaces, finite horizon models, and continuous-time discrete state models. The book is organized around optimality criteria, using a common framework centered on the optimality (Bellman) equation for presenting results. The results are presented in a \"theorem-proof\" format and elaborated on through both discussion and examples, including results that are not available in any other book. A two-state Markov decision process model, presented in Chapter 3, is analyzed repeatedly throughout the book and demonstrates many results and algorithms. Markov Decision Processes covers recent research advances in such areas as countable state space models with average reward criterion, constrained models, and models with risk sensitive optimality criteria. It also explores several topics that have received little or no attention in other books, including modified policy iteration, multichain models with average reward criterion, and sensitive optimality. In addition, a Bibliographic Remarks section in each chapter comments on relevant historic",
"",
"",
"Approximate linear programming (ALP) represents one of the major algorithmic families to solve large-scale Markov decision processes (MDP). In this work, we study a primal-dual formulation of the ALP, and develop a scalable, model-free algorithm called bilinear @math learning for reinforcement learning when a sampling oracle is provided. This algorithm enjoys a number of advantages. First, it adopts (bi)linear models to represent the high-dimensional value function and state-action distributions, using given state and action features. Its run-time complexity depends on the number of features, not the size of the underlying MDPs. Second, it operates in a fully online fashion without having to store any sample, thus having minimal memory footprint. Third, we prove that it is sample-efficient, solving for the optimal policy to high precision with a sample complexity linear in the dimension of the parameter space.",
"",
"We consider the problem of controlling a fully specified Markov decision process (MDP), also known as the planning problem, when the state space is very large and calculating the optimal policy is intractable. Instead, we pursue the more modest goal of optimizing over some small family of policies. Specifically, we show that the family of policies associated with a low-dimensional approximation of occupancy measures yields a tractable optimization. Moreover, we propose an efficient algorithm, scaling with the size of the subspace but not the state space, that is able to find a policy with low excess loss relative to the best policy in this class. To the best of our knowledge, such results did not exist in the literature previously. We bound excess loss in the average cost and discounted cost cases, which are treated separately. Preliminary experiments show the effectiveness of the proposed algorithms in a queueing application.",
"Consider the problem of approximating the optimal policy of a Markov decision process (MDP) by sampling state transitions. In contrast to existing reinforcement learning methods that are based on successive approximations to the nonlinear Bellman equation, we propose a Primal-Dual @math Learning method in light of the linear duality between the value and policy. The @math learning method is model-free and makes primal-dual updates to the policy and value vectors as new data are revealed. For infinite-horizon undiscounted Markov decision process with finite state space @math and finite action space @math , the @math learning method finds an @math -optimal policy using the following number of sample transitions @math where @math is an upper bound of mixing times across all policies and @math is a parameter characterizing the range of stationary distributions across policies. The @math learning method also applies to the computational problem of MDP where the transition probabilities and rewards are explicitly given as the input. In the case where each state transition can be sampled in @math time, the @math learning method gives a sublinear-time algorithm for solving the averaged-reward MDP.",
"Abstract : The purpose of this paper is to discuss the asymptotic behavior of the sequence (f sub n(i)) generated by a nonlinear recurrence relation. This problem arises in connection with an equipment replacement problem, cf. S. Dreyfus, A Note on an Industrial Replacement Process."
]
} |
1905.10799 | 2946223208 | The knowledge base completion problem is the problem of inferring missing information from existing facts in knowledge bases. Path-ranking based methods use sequences of relations as general patterns of paths for prediction. However, these patterns usually lack accuracy because they are generic and can often apply to widely varying scenarios. We leverage type hierarchies of entities to create a new class of path patterns that are both discriminative and generalizable. Then we propose an attention-based RNN model, which can be trained end-to-end, to discover the new path patterns most suitable for the data. Experiments conducted on two benchmark knowledge base completion datasets demonstrate that the proposed model outperforms existing methods by a statistically significant margin. Our quantitative analysis of the path patterns shows that they balance between generalization and discrimination. | In prior work, there exist multiple approaches to knowledge based reasoning. Knowledge graph embedding methods map relations and entities to vector representations in continuous spaces @cite_17 . Methods based on inductive logic programming discover general rules from examples @cite_1 . Statistical relational learning (SRL) methods combine logics and graphical models to probabilistically reason about entities and relations @cite_3 . Reinforcement learning based models treat link prediction (predicting target entities given a source entity and a relation) as Markov decision processes @cite_28 @cite_19 . Path-ranking based models use supervised learning to discover generalizable path patterns from graphs @cite_27 @cite_14 @cite_10 @cite_0 . | {
"cite_N": [
"@cite_14",
"@cite_28",
"@cite_1",
"@cite_3",
"@cite_0",
"@cite_19",
"@cite_27",
"@cite_10",
"@cite_17"
],
"mid": [
"2250601658",
"2962886429",
"1531743498",
"1860880244",
"2466714650",
"2964172232",
"1756422141",
"2584683887",
"1529533208"
],
"abstract": [
"We explore some of the practicalities of using random walk inference methods, such as the Path Ranking Algorithm (PRA), for the task of knowledge base completion. We show that the random walk probabilities computed (at great expense) by PRA provide no discernible benefit to performance on this task, so they can safely be dropped. This allows us to define a simpler algorithm for generating feature matrices from graphs, which we call subgraph feature extraction (SFE). In addition to being conceptually simpler than PRA, SFE is much more efficient, reducing computation by an order of magnitude, and more expressive, allowing for much richer features than paths between two nodes in a graph. We show experimentally that this technique gives substantially better performance than PRA and its variants, improving mean average precision from .432 to .528 on a knowledge base completion task using the NELL KB.",
"",
"FOIL is a learning system that constructs Horn clause programs from examples. This paper summarises the development of FOIL from 1989 up to early 1993 and evaluates its effectiveness on a non-trivial sequence of learning tasks taken from a Prolog programming text. Although many of these tasks are handled reasonably well, the experiment highlights some weaknesses of the current implementation. Areas for further research are identified.",
"",
"Our goal is to combine the rich multistep inference of symbolic logical reasoning with the generalization capabilities of neural networks. We are particularly interested in complex reasoning about entities and relations in text and large-scale knowledge bases (KBs). (2015) use RNNs to compose the distributed semantics of multi-hop paths in KBs; however for multiple reasons, the approach lacks accuracy and practicality. This paper proposes three significant modeling advances: (1) we learn to jointly reason about relations, entities, and entity-types; (2) we use neural attention modeling to incorporate multiple paths; (3) we learn to share strength in a single RNN that represents logical composition across all relations. On a largescale Freebase+ClueWeb prediction task, we achieve 25 error reduction, and a 53 error reduction on sparse relations due to shared strength. On chains of reasoning in WordNet we reduce error in mean quantile by 84 versus previous state-of-the-art. The code and data are available at this https URL",
"Knowledge bases (KB), both automatically and manually constructed, are often incomplete --- many valid facts can be inferred from the KB by synthesizing existing information. A popular approach to KB completion is to infer new relations by combinatory reasoning over the information found along other paths connecting a pair of entities. Given the enormous size of KBs and the exponential number of paths, previous path-based models have considered only the problem of predicting a missing relation given two entities, or evaluating the truth of a proposed triple. Additionally, these methods have traditionally used random paths between fixed entity pairs or more recently learned to pick paths between them. We propose a new algorithm, MINERVA, which addresses the much more difficult and practical task of answering questions where the relation is known, but only one entity. Since random walks are impractical in a setting with combinatorially many destinations from a start node, we present a neural reinforcement learning approach which learns how to navigate the graph conditioned on the input query to find predictive paths. Empirically, this approach obtains state-of-the-art results on several datasets, significantly outperforming prior methods.",
"We consider the problem of performing learning and inference in a large scale knowledge base containing imperfect knowledge with incomplete coverage. We show that a soft inference procedure based on a combination of constrained, weighted, random walks through the knowledge base graph can be used to reliably infer new beliefs for the knowledge base. More specifically, we show that the system can learn to infer different target relations by tuning the weights associated with random walks that follow different paths through the graph, using a version of the Path Ranking Algorithm (Lao and Cohen, 2010b). We apply this approach to a knowledge base of approximately 500,000 beliefs extracted imperfectly from the web by NELL, a never-ending language learner (, 2010). This new system improves significantly over NELL's earlier Horn-clause learning and inference method: it obtains nearly double the precision at rank 100, and the new learning method is also applicable to many more inference tasks.",
"Traditional approaches to knowledge base completion have been based on symbolic representations. Lowdimensional vector embedding models proposed recently for this task are attractive since they generalize to possibly unlimited sets of relations. A significant drawback of previous embedding models for KB completion is that they merely support reasoning on individual relations (e.g., bornIn(X, Y ) ⇒ nationality(X, Y )). In this work, we develop models for KB completion that support chains of reasoning on paths of any length using compositional vector space models. We construct compositional vector representations for the paths in the KB graph from the semantic vector representations of the binary relations in that path and perform inference directly in the vector space. Unlike previous methods, our approach can generalize to paths that are unseen in training and, in a zero-shot setting, predict target relations without supervised training data for that relation.",
"Relational machine learning studies methods for the statistical analysis of relational, or graph-structured, data. In this paper, we provide a review of how such statistical models can be “trained” on large knowledge graphs, and then used to predict new facts about the world (which is equivalent to predicting new edges in the graph). In particular, we discuss two fundamentally different kinds of statistical relational models, both of which can scale to massive data sets. The first is based on latent feature models such as tensor factorization and multiway neural networks. The second is based on mining observable patterns in the graph. We also show how to combine these latent and observable models to get improved modeling power at decreased computational cost. Finally, we discuss how such statistical models of graphs can be combined with text-based information extraction methods for automatically constructing knowledge graphs from the Web. To this end, we also discuss Google's knowledge vault project as an example of such combination."
]
} |
1905.10799 | 2946223208 | The knowledge base completion problem is the problem of inferring missing information from existing facts in knowledge bases. Path-ranking based methods use sequences of relations as general patterns of paths for prediction. However, these patterns usually lack accuracy because they are generic and can often apply to widely varying scenarios. We leverage type hierarchies of entities to create a new class of path patterns that are both discriminative and generalizable. Then we propose an attention-based RNN model, which can be trained end-to-end, to discover the new path patterns most suitable for the data. Experiments conducted on two benchmark knowledge base completion datasets demonstrate that the proposed model outperforms existing methods by a statistically significant margin. Our quantitative analysis of the path patterns shows that they balance between generalization and discrimination. | Our work focuses on the knowledge base completion problem, specifically on fact prediction. We focus on path-ranking based models because they have the ability to model complex reasoning patterns. Path Ranking Algorithm (PRA) @cite_27 first propose to use random walk to discover generalizable sequences of relations between two entities and treat each sequence as a unique feature with a weight equal to the random walk probability. Subgraph Feature Extraction (SFE) @cite_14 uses bi-directional breadth-first search (BFS) to exhaustively search for sequences of relations and additional patterns in graphs. They treat each pattern as a binary feature because calculating the weights is computationally intensive and not profitable with respect to the performance. To generalize semantically similar path patterns, @cite_10 use recurrent neural networks (RNNs) to sequentially encode relations in path patterns to create their vector representations, which are then used as multidimensional features for prediction. @cite_0 improve the accuracy of the RNN method by making use of the additional information in entities and entity types. We also use entity types but we focus on using types at different levels of abstraction for different entities. | {
"cite_N": [
"@cite_0",
"@cite_27",
"@cite_14",
"@cite_10"
],
"mid": [
"2466714650",
"1756422141",
"2250601658",
"2584683887"
],
"abstract": [
"Our goal is to combine the rich multistep inference of symbolic logical reasoning with the generalization capabilities of neural networks. We are particularly interested in complex reasoning about entities and relations in text and large-scale knowledge bases (KBs). (2015) use RNNs to compose the distributed semantics of multi-hop paths in KBs; however for multiple reasons, the approach lacks accuracy and practicality. This paper proposes three significant modeling advances: (1) we learn to jointly reason about relations, entities, and entity-types; (2) we use neural attention modeling to incorporate multiple paths; (3) we learn to share strength in a single RNN that represents logical composition across all relations. On a largescale Freebase+ClueWeb prediction task, we achieve 25 error reduction, and a 53 error reduction on sparse relations due to shared strength. On chains of reasoning in WordNet we reduce error in mean quantile by 84 versus previous state-of-the-art. The code and data are available at this https URL",
"We consider the problem of performing learning and inference in a large scale knowledge base containing imperfect knowledge with incomplete coverage. We show that a soft inference procedure based on a combination of constrained, weighted, random walks through the knowledge base graph can be used to reliably infer new beliefs for the knowledge base. More specifically, we show that the system can learn to infer different target relations by tuning the weights associated with random walks that follow different paths through the graph, using a version of the Path Ranking Algorithm (Lao and Cohen, 2010b). We apply this approach to a knowledge base of approximately 500,000 beliefs extracted imperfectly from the web by NELL, a never-ending language learner (, 2010). This new system improves significantly over NELL's earlier Horn-clause learning and inference method: it obtains nearly double the precision at rank 100, and the new learning method is also applicable to many more inference tasks.",
"We explore some of the practicalities of using random walk inference methods, such as the Path Ranking Algorithm (PRA), for the task of knowledge base completion. We show that the random walk probabilities computed (at great expense) by PRA provide no discernible benefit to performance on this task, so they can safely be dropped. This allows us to define a simpler algorithm for generating feature matrices from graphs, which we call subgraph feature extraction (SFE). In addition to being conceptually simpler than PRA, SFE is much more efficient, reducing computation by an order of magnitude, and more expressive, allowing for much richer features than paths between two nodes in a graph. We show experimentally that this technique gives substantially better performance than PRA and its variants, improving mean average precision from .432 to .528 on a knowledge base completion task using the NELL KB.",
"Traditional approaches to knowledge base completion have been based on symbolic representations. Lowdimensional vector embedding models proposed recently for this task are attractive since they generalize to possibly unlimited sets of relations. A significant drawback of previous embedding models for KB completion is that they merely support reasoning on individual relations (e.g., bornIn(X, Y ) ⇒ nationality(X, Y )). In this work, we develop models for KB completion that support chains of reasoning on paths of any length using compositional vector space models. We construct compositional vector representations for the paths in the KB graph from the semantic vector representations of the binary relations in that path and perform inference directly in the vector space. Unlike previous methods, our approach can generalize to paths that are unseen in training and, in a zero-shot setting, predict target relations without supervised training data for that relation."
]
} |
1905.10799 | 2946223208 | The knowledge base completion problem is the problem of inferring missing information from existing facts in knowledge bases. Path-ranking based methods use sequences of relations as general patterns of paths for prediction. However, these patterns usually lack accuracy because they are generic and can often apply to widely varying scenarios. We leverage type hierarchies of entities to create a new class of path patterns that are both discriminative and generalizable. Then we propose an attention-based RNN model, which can be trained end-to-end, to discover the new path patterns most suitable for the data. Experiments conducted on two benchmark knowledge base completion datasets demonstrate that the proposed model outperforms existing methods by a statistically significant margin. Our quantitative analysis of the path patterns shows that they balance between generalization and discrimination. | Attention was first introduced in @cite_4 for machine translation. The attention mechanism helps the encoder-decoder model to condition generation of translated words on different parts of the original sentence. Later, cross-modality attention was shown to be effective at image captioning @cite_29 and speech recognition @cite_24 . Our approach uses attention to focus on contextually important information from multiple paths much like the above methods. In addition, we use attention in an original way to efficiently discover the correct levels of abstraction for entities from a large search space. | {
"cite_N": [
"@cite_24",
"@cite_29",
"@cite_4"
],
"mid": [
"",
"2950178297",
"2964308564"
],
"abstract": [
"",
"Inspired by recent work in machine translation and object detection, we introduce an attention based model that automatically learns to describe the content of images. We describe how we can train this model in a deterministic manner using standard backpropagation techniques and stochastically by maximizing a variational lower bound. We also show through visualization how the model is able to automatically learn to fix its gaze on salient objects while generating the corresponding words in the output sequence. We validate the use of attention with state-of-the-art performance on three benchmark datasets: Flickr8k, Flickr30k and MS COCO.",
"Abstract: Neural machine translation is a recently proposed approach to machine translation. Unlike the traditional statistical machine translation, the neural machine translation aims at building a single neural network that can be jointly tuned to maximize the translation performance. The models proposed recently for neural machine translation often belong to a family of encoder-decoders and consists of an encoder that encodes a source sentence into a fixed-length vector from which a decoder generates a translation. In this paper, we conjecture that the use of a fixed-length vector is a bottleneck in improving the performance of this basic encoder-decoder architecture, and propose to extend this by allowing a model to automatically (soft-)search for parts of a source sentence that are relevant to predicting a target word, without having to form these parts as a hard segment explicitly. With this new approach, we achieve a translation performance comparable to the existing state-of-the-art phrase-based system on the task of English-to-French translation. Furthermore, qualitative analysis reveals that the (soft-)alignments found by the model agree well with our intuition."
]
} |
1905.10777 | 2946059079 | Recent deep learning based face recognition methods have achieved great performance, but it still remains challenging to recognize very low-resolution query face like 28x28 pixels when CCTV camera is far from the captured subject. Such face with very low-resolution is totally out of detail information of the face identity compared to normal resolution in a gallery and hard to find corresponding faces therein. To this end, we propose a Resolution Invariant Model (RIM) for addressing such cross-resolution face recognition problems, with three distinct novelties. First, RIM is a novel and unified deep architecture, containing a Face Hallucination sub-Net (FHN) and a Heterogeneous Recognition sub-Net (HRN), which are jointly learned end to end. Second, FHN is a well-designed tri-path Generative Adversarial Network (GAN) which simultaneously perceives facial structure and geometry prior information, i.e. landmark heatmaps and parsing maps, incorporated with an unsupervised cross-domain adversarial training strategy to super-resolve very low-resolution query image to its 8x larger ones without requiring them to be well aligned. Third, HRN is a generic Convolutional Neural Network (CNN) for heterogeneous face recognition with our proposed residual knowledge distillation strategy for learning discriminative yet generalized feature representation. Quantitative and qualitative experiments on several benchmarks demonstrate the superiority of the proposed model over the state-of-the-arts. Codes and models will be released upon acceptance. | Face recognition via deep learning has achieved a series of breakthrough in these years @cite_9 @cite_23 @cite_16 @cite_1 . For instance, DeepID @cite_9 and DeepID2 @cite_23 can be effectively learned by challenging multi-class identification and verification jointly, which achieve excellent recognition performance. Wen @cite_1 propose a center loss to further enhance the capacity of discriminative feature learning. Deng @cite_16 introduce an additive angular margin loss to obtain highly discriminative features for face recognition. Generally speaking, deep learning models have achieved outstanding results on face recognition. However, these methods hardly perform satisfactory on cross-resolution face recognition because of the absence of facial details in very low-resolution face images. | {
"cite_N": [
"@cite_9",
"@cite_16",
"@cite_1",
"@cite_23"
],
"mid": [
"1998808035",
"2784874046",
"2520774990",
"2144172034"
],
"abstract": [
"This paper proposes to learn a set of high-level feature representations through deep learning, referred to as Deep hidden IDentity features (DeepID), for face verification. We argue that DeepID can be effectively learned through challenging multi-class face identification tasks, whilst they can be generalized to other tasks (such as verification) and new identities unseen in the training set. Moreover, the generalization capability of DeepID increases as more face classes are to be predicted at training. DeepID features are taken from the last hidden layer neuron activations of deep convolutional networks (ConvNets). When learned as classifiers to recognize about 10, 000 face identities in the training set and configured to keep reducing the neuron numbers along the feature extraction hierarchy, these deep ConvNets gradually form compact identity-related features in the top layers with only a small number of hidden neurons. The proposed features are extracted from various face regions to form complementary and over-complete representations. Any state-of-the-art classifiers can be learned based on these high-level representations for face verification. 97:45 verification accuracy on LFW is achieved with only weakly aligned faces.",
"One of the main challenges in feature learning using Deep Convolutional Neural Networks (DCNNs) for large-scale face recognition is the design of appropriate loss functions that enhance discriminative power. Centre loss penalises the distance between the deep features and their corresponding class centres in the Euclidean space to achieve intra-class compactness. SphereFace assumes that the linear transformation matrix in the last fully connected layer can be used as a representation of the class centres in an angular space and penalises the angles between the deep features and their corresponding weights in a multiplicative way. Recently, a popular line of research is to incorporate margins in well-established loss functions in order to maximise face class separability. In this paper, we propose an Additive Angular Margin Loss (ArcFace) to obtain highly discriminative features for face recognition. The proposed ArcFace has a clear geometric interpretation due to the exact correspondence to the geodesic distance on the hypersphere. We present arguably the most extensive experimental evaluation of all the recent state-of-the-art face recognition methods on over 10 face recognition benchmarks including a new large-scale image database with trillion level of pairs and a large-scale video dataset. We show that ArcFace consistently outperforms the state-of-the-art and can be easily implemented with negligible computational overhead. We release all refined training data, training codes, pre-trained models and training logs, which will help reproduce the results in this paper.",
"Convolutional neural networks (CNNs) have been widely used in computer vision community, significantly improving the state-of-the-art. In most of the available CNNs, the softmax loss function is used as the supervision signal to train the deep model. In order to enhance the discriminative power of the deeply learned features, this paper proposes a new supervision signal, called center loss, for face recognition task. Specifically, the center loss simultaneously learns a center for deep features of each class and penalizes the distances between the deep features and their corresponding class centers. More importantly, we prove that the proposed center loss function is trainable and easy to optimize in the CNNs. With the joint supervision of softmax loss and center loss, we can train a robust CNNs to obtain the deep features with the two key learning objectives, inter-class dispension and intra-class compactness as much as possible, which are very essential to face recognition. It is encouraging to see that our CNNs (with such joint supervision) achieve the state-of-the-art accuracy on several important face recognition benchmarks, Labeled Faces in the Wild (LFW), YouTube Faces (YTF), and MegaFace Challenge. Especially, our new approach achieves the best results on MegaFace (the largest public domain face benchmark) under the protocol of small training set (contains under 500000 images and under 20000 persons), significantly improving the previous results and setting new state-of-the-art for both face recognition and face verification tasks.",
"The key challenge of face recognition is to develop effective feature representations for reducing intra-personal variations while enlarging inter-personal differences. In this paper, we show that it can be well solved with deep learning and using both face identification and verification signals as supervision. The Deep IDentification-verification features (DeepID2) are learned with carefully designed deep convolutional networks. The face identification task increases the inter-personal variations by drawing DeepID2 features extracted from different identities apart, while the face verification task reduces the intra-personal variations by pulling DeepID2 features extracted from the same identity together, both of which are essential to face recognition. The learned DeepID2 features can be well generalized to new identities unseen in the training data. On the challenging LFW dataset [11], 99.15 face verification accuracy is achieved. Compared with the best previous deep learning result [20] on LFW, the error rate has been significantly reduced by 67 ."
]
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.