aid stringlengths 9 15 | mid stringlengths 7 10 | abstract stringlengths 78 2.56k | related_work stringlengths 92 1.77k | ref_abstract dict |
|---|---|---|---|---|
1708.09006 | 2753371338 | An efficient, fully automatic method for 3D face shape and pose estimation in unconstrained 2D imagery is presented. The proposed method jointly estimates a dense set of 3D landmarks and facial geometry using a single pass of a modified version of the popular "U-Net" neural network architecture. Additionally, we propose a method for directly estimating a set of 3D Morphable Model (3DMM) parameters, using the estimated 3D landmarks and geometry as constraints in a simple linear system. Qualitative modeling results are presented, as well as quantitative evaluation of predicted 3D face landmarks in unconstrained video sequences. | In order to avoid the problem of viewpoint-dependent 2D landmarks, @cite_15 introduced the projected normalized coordinate code (PNCC), which acts as a fully-3D set of dense facial landmarks. They leverage the PNCC as an intermediate output in a cascade of convolutional neural networks, each of which produce updates to the current 3DMM coefficient estimates given the input image and a predicted PNCC, which is rendered using the current coefficient estimates at each stage. The proposed method also leverages the idea of the PNCC image, but directly estimates the PNCC (in conjunction with an image of dense 3D offsets) using a single network given only the original image as input. 3DMM coefficients are then fit (if desired) directly to the predicted 3D output. | {
"cite_N": [
"@cite_15"
],
"mid": [
"2964014798"
],
"abstract": [
"Face alignment, which fits a face model to an image and extracts the semantic meanings of facial pixels, has been an important topic in CV community. However, most algorithms are designed for faces in small to medium poses (below 45), lacking the ability to align faces in large poses up to 90. The challenges are three-fold: Firstly, the commonly used landmark-based face model assumes that all the landmarks are visible and is therefore not suitable for profile views. Secondly, the face appearance varies more dramatically across large poses, ranging from frontal view to profile view. Thirdly, labelling landmarks in large poses is extremely challenging since the invisible landmarks have to be guessed. In this paper, we propose a solution to the three problems in an new alignment framework, called 3D Dense Face Alignment (3DDFA), in which a dense 3D face model is fitted to the image via convolutional neutral network (CNN). We also propose a method to synthesize large-scale training samples in profile views to solve the third problem of data labelling. Experiments on the challenging AFLW database show that our approach achieves significant improvements over state-of-the-art methods."
]
} |
1708.09006 | 2753371338 | An efficient, fully automatic method for 3D face shape and pose estimation in unconstrained 2D imagery is presented. The proposed method jointly estimates a dense set of 3D landmarks and facial geometry using a single pass of a modified version of the popular "U-Net" neural network architecture. Additionally, we propose a method for directly estimating a set of 3D Morphable Model (3DMM) parameters, using the estimated 3D landmarks and geometry as constraints in a simple linear system. Qualitative modeling results are presented, as well as quantitative evaluation of predicted 3D face landmarks in unconstrained video sequences. | By making some simplifying assumptions about the reflectance properties of the face, it is possible to estimate fine-scaled 3D facial geometry using shape from shading (SfS) constraints. Although recent approaches @cite_13 @cite_12 @cite_7 have shown applicability in real-world'' scenarios, we avoid making assumptions about lighting, imaging, and reflectance conditions with the aim of producing as robust a system as is possible. | {
"cite_N": [
"@cite_13",
"@cite_7",
"@cite_12"
],
"mid": [
"2134724684",
"",
"2559821077"
],
"abstract": [
"We address the problem of reconstructing 3D face models from large unstructured photo collections, e.g., obtained by Google image search or from personal photo collections in iPhoto. This problem is extremely challenging due to the high degree of variability in pose, illumination, facial expression, non-rigid changes in face shape and reflectance over time and occlusions. In light of this extreme variability, no single reconstruction can be consistent with all of the images. Instead, we define as the goal of reconstruction to recover a model that is locally consistent with the image set. I.e., each local region of the model is consistent with a large set of photos, resulting in a model that captures the dominant trends in the input data for different parts of the face. Our approach leverages multi-image shading, but unlike traditional photometric stereo approaches, allows for changes in viewpoint and shape. We optimize over pose, shape, and lighting in an iterative approach that seeks to minimize the rank of the transformed images. This approach produces high quality shape models for a wide range of celebrities from photos available on the Internet.",
"",
"Given a photo collection of “unconstrained” face images of one individual captured under a variety of unknown pose, expression, and illumination conditions, this paper presents a method for reconstructing a 3D face surface model of the individual along with albedo information. Unlike prior work on face reconstruction that requires large photo collections, we formulate an approach to adapt to photo collections with a high diversity in both the number of images and the image quality. To achieve this, we incorporate prior knowledge about face shape by fitting a 3D morphable model to form a personalized template, following by using a novel photometric stereo formulation to complete the fine details, under a coarse-to-fine scheme. Our scheme incorporates a structural similarity-based local selection step to help identify a common expression for reconstruction while discarding occluded portions of faces. The evaluation of reconstruction performance is through a novel quality measure, in the absence of ground truth 3D scans. Superior large-scale experimental results are reported on synthetic, Internet, and personal photo collections."
]
} |
1708.08739 | 2751546962 | Graphs are an important tool to model data in different domains, including social networks, bioinformatics and the world wide web. Most of the networks formed in these domains are directed graphs, where all the edges have a direction and they are not symmetric. Betweenness centrality is an important index widely used to analyze networks. In this paper, first given a directed network @math and a vertex @math , we propose a new exact algorithm to compute betweenness score of @math . Our algorithm pre-computes a set @math , which is used to prune a huge amount of computations that do not contribute in the betweenness score of @math . Time complexity of our exact algorithm depends on @math and it is respectively @math and @math for unweighted graphs and weighted graphs with positive weights. @math is bounded from above by @math and in most cases, it is a small constant. Then, for the cases where @math is large, we present a simple randomized algorithm that samples from @math and performs computations for only the sampled elements. We show that this algorithm provides an @math -approximation of the betweenness score of @math . Finally, we perform extensive experiments over several real-world datasets from different domains for several randomly chosen vertices as well as for the vertices with the highest betweenness scores. Our experiments reveal that in most cases, our algorithm significantly outperforms the most efficient existing randomized algorithms, in terms of both running time and accuracy. Our experiments also show that our proposed algorithm computes betweenness scores of all vertices in the sets of sizes 5, 10 and 15, much faster and more accurate than the most efficient existing algorithms. | Everett and Borgatti @cite_14 defined as a natural extension of betweenness centrality for sets of vertices. Group betweenness centrality of a set is defined as the number of shortest paths passing through at least one of the vertices in the set @cite_14 . The other natural extension of betweenness centrality is . Co-betweenness centrality is defined as the number of shortest paths passing through all vertices in the set. Kolaczyk et. al. @cite_7 presented an @math time algorithm for co-betweenness centrality computation of sets of size 2. Chehreghani @cite_4 presented efficient algorithms for co-betweenness centrality computation of any set or sequence of vertices in weighted and unweighted networks. Puzis et. al. @cite_32 proposed an @math time algorithm for computing successive group betweenness centrality, where @math is the size of the set. The same authors in @cite_15 presented two algorithms for finding . A of a network is a set vertices of minimum size, so that every shortest path in the network passes through at least one of the vertices in the set. The first algorithm is based on a heuristic search and the second one is based on iterative greedy choice of vertices. | {
"cite_N": [
"@cite_14",
"@cite_4",
"@cite_7",
"@cite_32",
"@cite_15"
],
"mid": [
"2089370556",
"",
"1992229444",
"2003408562",
"1790404493"
],
"abstract": [
"This paper extends the standard network centrality measures of degree, closeness and betweenness to apply to groups and classes as well as individuals. The group centrality measures will enable researchers to answer such questions as ‘how central is the engineering department in the informal influence network of this company?’ or ‘among middle managers in a given organization, which are more central, the men or the women?’ With these measures we can also solve the inverse problem: given the network of ties among organization members, how can we form a team that is maximally central? The measures are illustrated using two classic network data sets. We also formalize a measure of group centrality efficiency, which indicates the extent to which a group's centrality is principally due to a small subset of its members.",
"",
"Abstract Vertex betweenness centrality is a metric that seeks to quantify a sense of the importance of a vertex in a network in terms of its ‘control’ on the flow of information along geodesic paths throughout the network. Two natural ways to extend vertex betweenness centrality to sets of vertices are (i) in terms of geodesic paths that pass through at least one of the vertices in the set, and (ii) in terms of geodesic paths that pass through all vertices in the set. The former was introduced by Everett and Borgatti [Everett, M., Borgatti, S., 1999. The centrality of groups and classes. Journal of Mathematical Sociology 23 (3), 181–201], and called group betweenness centrality. The latter, which we call co-betweenness centrality here, has not been considered formally in the literature until now, to the best of our knowledge. In this paper, we show that these two notions of centrality are in fact intimately related and, furthermore, that this relationship may be exploited to obtain deeper insight into both. In particular, we provide an expansion for group betweenness in terms of increasingly higher orders of co-betweenness, in a manner analogous to the Taylor series expansion of a mathematical function in calculus. We then demonstrate the utility of this expansion by using it to construct analytic lower and upper bounds for group betweenness that involve only simple combinations of (i) the betweenness of individual vertices in the group, and (ii) the co-betweenness of pairs of these vertices. Accordingly, we argue that the latter quantity, i.e., pairwise co-betweenness, is itself a fundamental quantity of some independent interest, and we present a computationally efficient algorithm for its calculation, which extends the algorithm of Brandes [Brandes, U., 2001. A faster algorithm for betweenness centrality. Journal of Mathematical Sociology 25, 163] in a natural manner. Applications are provided throughout, using a handful of different communication networks, which serve to illustrate the way in which our mathematical contributions allow for insight to be gained into the interaction of network structure, coalitions, and information flow in social networks.",
"In this paper, we propose a method for rapid computation of group betweenness centrality whose running time (after preprocessing) does not depend on network size. The calculation of group betweenness centrality is computationally demanding and, therefore, it is not suitable for applications that compute the centrality of many groups in order to identify new properties. Our method is based on the concept of path betweenness centrality defined in this paper. We demonstrate how the method can be used to find the most prominent group. Then, we apply the method for epidemic control in communication networks. We also show how the method can be used to evaluate distributions of group betweenness centrality and its correlation with group degree. The method may assist in finding further properties of complex networks and may open a wide range of research opportunities.",
"In many applications we are required to locate the most prominent group of vertices in a complex network. Group Betweenness Centrality can be used to evaluate the prominence of a group of vertices. Evaluating the Betweenness of every possible group in order to find the most prominent is not computationally feasible for large networks. In this paper we present two algorithms for finding the most prominent group. The first algorithm is based on heuristic search and the second is based on iterative greedy choice of vertices. The algorithms were evaluated on random and scale-free networks. Empirical evaluation suggests that the greedy algorithm results were negligibly below the optimal result. In addition, both algorithms performed better on scale-free networks: heuristic search was faster and the greedy algorithm produced more accurate results. The greedy algorithm was applied for optimizing deployment of intrusion detection devices on network service provider infrastructure."
]
} |
1708.08739 | 2751546962 | Graphs are an important tool to model data in different domains, including social networks, bioinformatics and the world wide web. Most of the networks formed in these domains are directed graphs, where all the edges have a direction and they are not symmetric. Betweenness centrality is an important index widely used to analyze networks. In this paper, first given a directed network @math and a vertex @math , we propose a new exact algorithm to compute betweenness score of @math . Our algorithm pre-computes a set @math , which is used to prune a huge amount of computations that do not contribute in the betweenness score of @math . Time complexity of our exact algorithm depends on @math and it is respectively @math and @math for unweighted graphs and weighted graphs with positive weights. @math is bounded from above by @math and in most cases, it is a small constant. Then, for the cases where @math is large, we present a simple randomized algorithm that samples from @math and performs computations for only the sampled elements. We show that this algorithm provides an @math -approximation of the betweenness score of @math . Finally, we perform extensive experiments over several real-world datasets from different domains for several randomly chosen vertices as well as for the vertices with the highest betweenness scores. Our experiments reveal that in most cases, our algorithm significantly outperforms the most efficient existing randomized algorithms, in terms of both running time and accuracy. Our experiments also show that our proposed algorithm computes betweenness scores of all vertices in the sets of sizes 5, 10 and 15, much faster and more accurate than the most efficient existing algorithms. | Lee et. al. @cite_20 proposed an algorithm to efficiently update betweenness centralities of vertices when the graph obtains a new edge. They reduced the search space by finding a candidate set of vertices whose betweenness centralities can be updated. Bergamini et. al. @cite_27 presented approximate algorithms that update betweenness scores of all vertices when an edge is inserted or the weight of an edge decreases. They used the algorithm of @cite_29 as the building block. Hayashi et. al. @cite_30 proposed a fully dynamic algorithm for estimating betweenness centrality of all vertices in a large dynamic network. Their algorithm is based on two data structures: hypergraph sketch that keeps track of SPDs, and two-ball index that helps to identify the parts of hypergraph sketches that require updates. | {
"cite_N": [
"@cite_30",
"@cite_27",
"@cite_29",
"@cite_20"
],
"mid": [
"",
"2963957156",
"1029409534",
"2036977244"
],
"abstract": [
"",
"Betweenness centrality ranks the importance of nodes by their participation in all shortest paths of the network. Therefore computing exact betweenness values is impractical in large networks. For static networks, approximation based on randomly sampled paths has been shown to be significantly faster in practice. However, for dynamic networks, no approximation algorithm for betweenness centrality is known that improves on static recomputation. We address this deficit by proposing two incremental approximation algorithms (for weighted and unweighted connected graphs) which provide a provable guarantee on the absolute approximation error. Processing batches of edge insertions, our algorithms yield significant speedups up to a factor of 104 compared to restarting the approximation. This is enabled by investing memory to store and efficiently update shortest paths. As a building block, we also propose an asymptotically faster algorithm for updating the SSSP problem in unweighted graphs. Our experimental study shows that our algorithms are the first to make in-memory computation of a betweenness ranking practical for million-edge semi-dynamic networks. Moreover, our results show that the accuracy is even better than the theoretical guarantees in terms of absolute errors and the rank of nodes is well preserved, in particular for those with high betweenness.",
"Betweenness centrality is a fundamental measure in social network analysis, expressing the importance or influence of individual vertices (or edges) in a network in terms of the fraction of shortest paths that pass through them. Since exact computation in large networks is prohibitively expensive, we present two efficient randomized algorithms for betweenness estimation. The algorithms are based on random sampling of shortest paths and offer probabilistic guarantees on the quality of the approximation. The first algorithm estimates the betweenness of all vertices (or edges): all approximate values are within an additive factor @math ??(0,1) from the real values, with probability at least @math 1-?. The second algorithm focuses on the top-K vertices (or edges) with highest betweenness and estimate their betweenness value to within a multiplicative factor @math ?, with probability at least @math 1-?. This is the first algorithm that can compute such approximation for the top-K vertices (or edges). By proving upper and lower bounds to the VC-dimension of a range set associated with the problem at hand, we can bound the sample size needed to achieve the desired approximations. We obtain sample sizes that are independent from the number of vertices in the network and only depend on a characteristic quantity that we call the vertex-diameter, that is the maximum number of vertices in a shortest path. In some cases, the sample size is completely independent from any quantitative property of the graph. An extensive experimental evaluation on real and artificial networks shows that our algorithms are significantly faster and much more scalable as the number of vertices grows than other algorithms with similar approximation guarantees.",
"The betweenness centrality of a vertex in a graph is a measure for the participation of the vertex in the shortest paths in the graph. The Betweenness centrality is widely used in network analyses. Especially in a social network, the recursive computation of the betweenness centralities of vertices is performed for the community detection and finding the influential user in the network. Since a social network graph is frequently updated, it is necessary to update the betweenness centrality efficiently. When a graph is changed, the betweenness centralities of all the vertices should be recomputed from scratch using all the vertices in the graph. To the best of our knowledge, this is the first work that proposes an efficient algorithm which handles the update of the betweenness centralities of vertices in a graph. In this paper, we propose a method that efficiently reduces the search space by finding a candidate set of vertices whose betweenness centralities can be updated and computes their betweenness centeralities using candidate vertices only. As the cost of calculating the betweenness centrality mainly depends on the number of vertices to be considered, the proposed algorithm significantly reduces the cost of calculation. The proposed algorithm allows the transformation of an existing algorithm which does not consider the graph update. Experimental results on large real datasets show that the proposed algorithm speeds up the existing algorithm 2 to 2418 times depending on the dataset."
]
} |
1708.08638 | 2750677909 | Imitation learning has been studied widely due to its convenient transfer of human experiences to robots. This learning approach models human demonstrations by extracting relevant motion patterns as well as adaptation to different situations. In order to address unpredicted situations such as obstacles and external perturbations, motor skills adaptation is crucial and non-trivial, particularly in dynamic or unstructured environments. In this paper, we propose to tackle this problem using a novel kernelized movement primitive (KMP) adaptation, which not only allows the robot to adapt its motor skills and meet a variety of additional task constraints arising over the course of the task, but also renders fewer open parameters unlike methods built on basis functions. Moreover, we extend our approach by introducing the concept of local frames, which represent coordinate systems of interest for tasks and could be modulated in accordance with external task parameters, endowing KMP with reliable extrapolation abilities in a broader domain. Several examples of trajectory modulations and extrapolations verify the effectiveness of our method. | Similarly to our approach, information theory has also been exploited in different robot learning techniques. As an effective way to measure the distance between two probabilistic distributions, KL-divergence was exploited in policy search , trajectory optimization and imitation learning . In @cite_5 KL-divergence was used to measure the difference between the distributions of demonstrations and the predicted robot trajectories (obtained from a control policy and a Gaussian process forward model), and subsequently the probabilistic inference for learning control was employed to iteratively minimize the KL-divergence so as to find the optimal policy parameters. It is noted that this KL-divergence formulation makes the derivations of analytical solution intractable. In this article, we formulate the trajectory matching problem as ), which allows us to separate the mean and covariance subproblems and derive closed-form solutions for them separately. | {
"cite_N": [
"@cite_5"
],
"mid": [
"2141098361"
],
"abstract": [
"Recent advances in the field of humanoid robotics increase the complexity of the tasks that such robots can perform. This makes it increasingly difficult and inconvenient to program these tasks manually. Furthermore, humanoid robots, in contrast to industrial robots, should in the distant future behave within a social environment. Therefore, it must be possible to extend the robot's abilities in an easy and natural way. To address these requirements, this work investigates the topic of imitation learning of motor skills. The focus lies on providing a humanoid robot with the ability to learn new bi-manual tasks through the observation of object trajectories. For this, an imitation learning framework is presented, which allows the robot to learn the important elements of an observed movement task by application of probabilistic encoding with Gaussian Mixture Models. The learned information is used to initialize an attractor-based movement generation algorithm that optimizes the reproduced movement towards the fulfillment of additional criteria, such as collision avoidance. Experiments performed with the humanoid robot ASIMO show that the proposed system is suitable for transferring information from a human demonstrator to the robot. These results provide a good starting point for more complex and interactive learning tasks."
]
} |
1708.08687 | 2752077245 | Input binarization has shown to be an effective way for network acceleration. However, previous binarization scheme could be regarded as simple pixel-wise thresholding operations (i.e., order-one approximation) and suffers a big accuracy loss. In this paper, we propose a highorder binarization scheme, which achieves more accurate approximation while still possesses the advantage of binary operation. In particular, the proposed scheme recursively performs residual quantization and yields a series of binary input images with decreasing magnitude scales. Accordingly, we propose high-order binary filtering and gradient propagation operations for both forward and backward computations. Theoretical analysis shows approximation error guarantee property of proposed method. Extensive experimental results demonstrate that the proposed scheme yields great recognition accuracy while being accelerated. | Collins al @cite_16 proposed a method using sparsity-inducing regularizer during the training of CNNs. Then Han al @cite_26 proposed an approach to apply parameter pruning to a memory-efficient structure. Related approaches can be found in @cite_2 and @cite_14 . They firstly find similar neurons during the training process and then combine or remove these neurons. These methods are based on a pre-trained neural network. Our High-Order Residual Quantization method does not rely on a pre-trained network. Therefor, our method has an advantage of easy training. | {
"cite_N": [
"@cite_14",
"@cite_16",
"@cite_26",
"@cite_2"
],
"mid": [
"2520760693",
"1570197553",
"2119144962",
"992687842"
],
"abstract": [
"To attain a favorable performance on large-scale datasets, convolutional neural networks (CNNs) are usually designed to have very high capacity involving millions of parameters. In this work, we aim at optimizing the number of neurons in a network, thus the number of parameters. We show that, by incorporating sparse constraints into the objective function, it is possible to decimate the number of neurons during the training stage. As a result, the number of parameters and the memory footprint of the neural network are also reduced, which is also desirable at the test time. We evaluated our method on several well-known CNN structures including AlexNet, and VGG over different datasets including ImageNet. Extensive experimental results demonstrate that our method leads to compact networks. Taking first fully connected layer as an example, our compact CNN contains only (30 , ) of the original neurons without any degradation of the top-1 classification accuracy.",
"In this work, we investigate the use of sparsity-inducing regularizers during training of Convolution Neural Networks (CNNs). These regularizers encourage that fewer connections in the convolution and fully connected layers take non-zero values and in effect result in sparse connectivity between hidden units in the deep network. This in turn reduces the memory and runtime cost involved in deploying the learned CNNs. We show that training with such regularization can still be performed using stochastic gradient descent implying that it can be used easily in existing codebases. Experimental evaluation of our approach on MNIST, CIFAR, and ImageNet datasets shows that our regularizers can result in dramatic reductions in memory requirements. For instance, when applied on AlexNet, our method can reduce the memory consumption by a factor of four with minimal loss in accuracy.",
"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.",
"Deep Neural nets (NNs) with millions of parameters are at the heart of many state-of-the-art computer vision systems today. However, recent works have shown that much smaller models can achieve similar levels of performance. In this work, we address the problem of pruning parameters in a trained NN model. Instead of removing individual weights one at a time as done in previous works, we remove one neuron at a time. We show how similar neurons are redundant, and propose a systematic way to remove them. Our experiments in pruning the densely connected layers show that we can remove upto 85 of the total parameters in an MNIST-trained network, and about 35 for AlexNet without significantly affecting performance. Our method can be applied on top of most networks with a fully connected layer to give a smaller network."
]
} |
1708.08732 | 2748391982 | Abstract Most existing approaches address multi-view subspace clustering problem by constructing the affinity matrix on each view separately and afterwards propose how to extend spectral clustering algorithm to handle multi-view data. This paper presents an approach to multi-view subspace clustering that learns a joint subspace representation by constructing affinity matrix shared among all views. Relying on the importance of both low-rank and sparsity constraints in the construction of the affinity matrix, we introduce the objective that balances between the agreement across different views, while at the same time encourages sparsity and low-rankness of the solution. Related low-rank and sparsity constrained optimization problem is for each view solved using the alternating direction method of multipliers. Furthermore, we extend our approach to cluster data drawn from nonlinear subspaces by solving the corresponding problem in a reproducing kernel Hilbert space. The proposed algorithm outperforms state-of-the-art multi-view subspace clustering algorithms on one synthetic and four real-world datasets. | In this section, we give a brief introduction to Sparse Subspace Clustering (SSC) @cite_10 , Low-Rank Representation (LRR) @cite_17 @cite_36 and Low-rank Sparse Subspace Clustering (LRSSC) @cite_5 . | {
"cite_N": [
"@cite_36",
"@cite_5",
"@cite_10",
"@cite_17"
],
"mid": [
"",
"2944309706",
"1993962865",
"79405465"
],
"abstract": [
"",
"An important problem in analyzing big data is subspace clustering , i.e., to represent a collection of points in a high-dimensional space via the union of low-dimensional subspaces. Sparse subspace clustering (SSC) and Low-rank representation (LRR) are the state-of-the-art methods for this task. These two methods are fundamentally similar in that both are based on convex optimization exploiting the intuition of “Self-Expressiveness”. The main difference is that the SSC minimizes the vector @math norm of the representation matrix to induce sparsity while LRR minimizes the nuclear norm (aka trace norm) to promote a low-rank structure. Because the representation matrix is often simultaneously sparse and low-rank, we propose a new algorithm, termed Low-rank sparse subspace clustering (LRSSC), by the combining SSC and LRR, and develop theoretical guarantees of the success of the algorithm. The results reveal interesting insights into the strengths and the weaknesses of SSC and LRR, and demonstrate how the LRSSC can take advantage of both methods in preserving the “Self-Expressiveness Property” and “Graph Connectivity” at the same time. A byproduct of our analysis is that it also expands the theoretical guarantee of SSC to handle cases when the subspaces have arbitrarily small canonical angles but are “nearly independent”.",
"Many real-world problems deal with collections of high-dimensional data, such as images, videos, text, and web documents, DNA microarray data, and more. Often, such high-dimensional data lie close to low-dimensional structures corresponding to several classes or categories to which the data belong. In this paper, we propose and study an algorithm, called sparse subspace clustering, to cluster data points that lie in a union of low-dimensional subspaces. The key idea is that, among the infinitely many possible representations of a data point in terms of other points, a sparse representation corresponds to selecting a few points from the same subspace. This motivates solving a sparse optimization program whose solution is used in a spectral clustering framework to infer the clustering of the data into subspaces. Since solving the sparse optimization program is in general NP-hard, we consider a convex relaxation and show that, under appropriate conditions on the arrangement of the subspaces and the distribution of the data, the proposed minimization program succeeds in recovering the desired sparse representations. The proposed algorithm is efficient and can handle data points near the intersections of subspaces. Another key advantage of the proposed algorithm with respect to the state of the art is that it can deal directly with data nuisances, such as noise, sparse outlying entries, and missing entries, by incorporating the model of the data into the sparse optimization program. We demonstrate the effectiveness of the proposed algorithm through experiments on synthetic data as well as the two real-world problems of motion segmentation and face clustering.",
"We propose low-rank representation (LRR) to segment data drawn from a union of multiple linear (or affine) subspaces. Given a set of data vectors, LRR seeks the lowest-rank representation among all the candidates that represent all vectors as the linear combination of the bases in a dictionary. Unlike the well-known sparse representation (SR), which computes the sparsest representation of each data vector individually, LRR aims at finding the lowest-rank representation of a collection of vectors jointly. LRR better captures the global structure of data, giving a more effective tool for robust subspace segmentation from corrupted data. Both theoretical and experimental results show that LRR is a promising tool for subspace segmentation."
]
} |
1708.08732 | 2748391982 | Abstract Most existing approaches address multi-view subspace clustering problem by constructing the affinity matrix on each view separately and afterwards propose how to extend spectral clustering algorithm to handle multi-view data. This paper presents an approach to multi-view subspace clustering that learns a joint subspace representation by constructing affinity matrix shared among all views. Relying on the importance of both low-rank and sparsity constraints in the construction of the affinity matrix, we introduce the objective that balances between the agreement across different views, while at the same time encourages sparsity and low-rankness of the solution. Related low-rank and sparsity constrained optimization problem is for each view solved using the alternating direction method of multipliers. Furthermore, we extend our approach to cluster data drawn from nonlinear subspaces by solving the corresponding problem in a reproducing kernel Hilbert space. The proposed algorithm outperforms state-of-the-art multi-view subspace clustering algorithms on one synthetic and four real-world datasets. | Denote the SVD of @math as @math . The minimizer of equation ) is uniquely given by @cite_17 : In the cases when data is contaminated by noise, the following problem needs to be solved: The optimal solution of equation ) has been derived in @cite_20 : where @math , @math and @math . Matrices are partitioned according to the sets @math and @math . | {
"cite_N": [
"@cite_20",
"@cite_17"
],
"mid": [
"2037549374",
"79405465"
],
"abstract": [
"We consider the problem of fitting one or more subspaces to a collection of data points drawn from the subspaces and corrupted by noise outliers. We pose this problem as a rank minimization problem, where the goal is to decompose the corrupted data matrix as the sum of a clean, self-expressive, low-rank dictionary plus a matrix of noise outliers. Our key contribution is to show that, for noisy data, this non-convex problem can be solved very efficiently and in closed form from the SVD of the noisy data matrix. Remarkably, this is true for both one or more subspaces. An important difference with respect to existing methods is that our framework results in a polynomial thresholding of the singular values with minimal shrinkage. Indeed, a particular case of our framework in the case of a single subspace leads to classical PCA, which requires no shrinkage. In the case of multiple subspaces, our framework provides an affinity matrix that can be used to cluster the data according to the sub-spaces. In the case of data corrupted by outliers, a closed-form solution appears elusive. We thus use an augmented Lagrangian optimization framework, which requires a combination of our proposed polynomial thresholding operator with the more traditional shrinkage-thresholding operator.",
"We propose low-rank representation (LRR) to segment data drawn from a union of multiple linear (or affine) subspaces. Given a set of data vectors, LRR seeks the lowest-rank representation among all the candidates that represent all vectors as the linear combination of the bases in a dictionary. Unlike the well-known sparse representation (SR), which computes the sparsest representation of each data vector individually, LRR aims at finding the lowest-rank representation of a collection of vectors jointly. LRR better captures the global structure of data, giving a more effective tool for robust subspace segmentation from corrupted data. Both theoretical and experimental results show that LRR is a promising tool for subspace segmentation."
]
} |
1708.08732 | 2748391982 | Abstract Most existing approaches address multi-view subspace clustering problem by constructing the affinity matrix on each view separately and afterwards propose how to extend spectral clustering algorithm to handle multi-view data. This paper presents an approach to multi-view subspace clustering that learns a joint subspace representation by constructing affinity matrix shared among all views. Relying on the importance of both low-rank and sparsity constraints in the construction of the affinity matrix, we introduce the objective that balances between the agreement across different views, while at the same time encourages sparsity and low-rankness of the solution. Related low-rank and sparsity constrained optimization problem is for each view solved using the alternating direction method of multipliers. Furthermore, we extend our approach to cluster data drawn from nonlinear subspaces by solving the corresponding problem in a reproducing kernel Hilbert space. The proposed algorithm outperforms state-of-the-art multi-view subspace clustering algorithms on one synthetic and four real-world datasets. | Sparse Subspace Clustering (SSC) @cite_10 requires that each data point is represented by a small number of data points from its own subspace and it amounts to solve the following minimization problem: The @math norm is used as the tightest convex relaxation of the @math quasi-norm that counts the number of nonzero elements of the solution. Constraint @math is used to avoid trivial solution of representing a data point as a linear combination of itself. | {
"cite_N": [
"@cite_10"
],
"mid": [
"1993962865"
],
"abstract": [
"Many real-world problems deal with collections of high-dimensional data, such as images, videos, text, and web documents, DNA microarray data, and more. Often, such high-dimensional data lie close to low-dimensional structures corresponding to several classes or categories to which the data belong. In this paper, we propose and study an algorithm, called sparse subspace clustering, to cluster data points that lie in a union of low-dimensional subspaces. The key idea is that, among the infinitely many possible representations of a data point in terms of other points, a sparse representation corresponds to selecting a few points from the same subspace. This motivates solving a sparse optimization program whose solution is used in a spectral clustering framework to infer the clustering of the data into subspaces. Since solving the sparse optimization program is in general NP-hard, we consider a convex relaxation and show that, under appropriate conditions on the arrangement of the subspaces and the distribution of the data, the proposed minimization program succeeds in recovering the desired sparse representations. The proposed algorithm is efficient and can handle data points near the intersections of subspaces. Another key advantage of the proposed algorithm with respect to the state of the art is that it can deal directly with data nuisances, such as noise, sparse outlying entries, and missing entries, by incorporating the model of the data into the sparse optimization program. We demonstrate the effectiveness of the proposed algorithm through experiments on synthetic data as well as the two real-world problems of motion segmentation and face clustering."
]
} |
1708.08732 | 2748391982 | Abstract Most existing approaches address multi-view subspace clustering problem by constructing the affinity matrix on each view separately and afterwards propose how to extend spectral clustering algorithm to handle multi-view data. This paper presents an approach to multi-view subspace clustering that learns a joint subspace representation by constructing affinity matrix shared among all views. Relying on the importance of both low-rank and sparsity constraints in the construction of the affinity matrix, we introduce the objective that balances between the agreement across different views, while at the same time encourages sparsity and low-rankness of the solution. Related low-rank and sparsity constrained optimization problem is for each view solved using the alternating direction method of multipliers. Furthermore, we extend our approach to cluster data drawn from nonlinear subspaces by solving the corresponding problem in a reproducing kernel Hilbert space. The proposed algorithm outperforms state-of-the-art multi-view subspace clustering algorithms on one synthetic and four real-world datasets. | If data is contaminated by noise, the following minimization problem needs to be solved: This problem can be efficiently solved using ADMM optimization procedure @cite_41 . | {
"cite_N": [
"@cite_41"
],
"mid": [
"2164278908"
],
"abstract": [
"Many problems of recent interest in statistics and machine learning can be posed in the framework of convex optimization. Due to the explosion in size and complexity of modern datasets, it is increasingly important to be able to solve problems with a very large number of features or training examples. As a result, both the decentralized collection or storage of these datasets as well as accompanying distributed solution methods are either necessary or at least highly desirable. In this review, we argue that the alternating direction method of multipliers is well suited to distributed convex optimization, and in particular to large-scale problems arising in statistics, machine learning, and related areas. The method was developed in the 1970s, with roots in the 1950s, and is equivalent or closely related to many other algorithms, such as dual decomposition, the method of multipliers, Douglas–Rachford splitting, Spingarn's method of partial inverses, Dykstra's alternating projections, Bregman iterative algorithms for l1 problems, proximal methods, and others. After briefly surveying the theory and history of the algorithm, we discuss applications to a wide variety of statistical and machine learning problems of recent interest, including the lasso, sparse logistic regression, basis pursuit, covariance selection, support vector machines, and many others. We also discuss general distributed optimization, extensions to the nonconvex setting, and efficient implementation, including some details on distributed MPI and Hadoop MapReduce implementations."
]
} |
1708.08732 | 2748391982 | Abstract Most existing approaches address multi-view subspace clustering problem by constructing the affinity matrix on each view separately and afterwards propose how to extend spectral clustering algorithm to handle multi-view data. This paper presents an approach to multi-view subspace clustering that learns a joint subspace representation by constructing affinity matrix shared among all views. Relying on the importance of both low-rank and sparsity constraints in the construction of the affinity matrix, we introduce the objective that balances between the agreement across different views, while at the same time encourages sparsity and low-rankness of the solution. Related low-rank and sparsity constrained optimization problem is for each view solved using the alternating direction method of multipliers. Furthermore, we extend our approach to cluster data drawn from nonlinear subspaces by solving the corresponding problem in a reproducing kernel Hilbert space. The proposed algorithm outperforms state-of-the-art multi-view subspace clustering algorithms on one synthetic and four real-world datasets. | Low-Rank Sparse Subspace Clustering (LRSSC) @cite_5 combines low-rank and sparsity constraints: In the case of the corrupted data the following problem needs to be solved to approximate @math : Once matrix @math is obtained by LRR, SSC or LRSSC approach, the affinity matrix @math is calculated as: Given affinity matrix @math , spectral clustering @cite_39 @cite_22 finds cluster membership of data points by applying k-means clustering to the eigenvectors of the graph Laplacian matrix @math computed from the affinity matrix @math . | {
"cite_N": [
"@cite_5",
"@cite_22",
"@cite_39"
],
"mid": [
"2944309706",
"",
"2121947440"
],
"abstract": [
"An important problem in analyzing big data is subspace clustering , i.e., to represent a collection of points in a high-dimensional space via the union of low-dimensional subspaces. Sparse subspace clustering (SSC) and Low-rank representation (LRR) are the state-of-the-art methods for this task. These two methods are fundamentally similar in that both are based on convex optimization exploiting the intuition of “Self-Expressiveness”. The main difference is that the SSC minimizes the vector @math norm of the representation matrix to induce sparsity while LRR minimizes the nuclear norm (aka trace norm) to promote a low-rank structure. Because the representation matrix is often simultaneously sparse and low-rank, we propose a new algorithm, termed Low-rank sparse subspace clustering (LRSSC), by the combining SSC and LRR, and develop theoretical guarantees of the success of the algorithm. The results reveal interesting insights into the strengths and the weaknesses of SSC and LRR, and demonstrate how the LRSSC can take advantage of both methods in preserving the “Self-Expressiveness Property” and “Graph Connectivity” at the same time. A byproduct of our analysis is that it also expands the theoretical guarantee of SSC to handle cases when the subspaces have arbitrarily small canonical angles but are “nearly independent”.",
"",
"We propose a novel approach for solving the perceptual grouping problem in vision. Rather than focusing on local features and their consistencies in the image data, our approach aims at extracting the global impression of an image. We treat image segmentation as a graph partitioning problem and propose a novel global criterion, the normalized cut, for segmenting the graph. The normalized cut criterion measures both the total dissimilarity between the different groups as well as the total similarity within the groups. We show that an efficient computational technique based on a generalized eigenvalue problem can be used to optimize this criterion. We applied this approach to segmenting static images, as well as motion sequences, and found the results to be very encouraging."
]
} |
1708.08744 | 2750637936 | In higher educational institutes, many students have to struggle hard to complete different courses since there is no dedicated support offered to students who need special attention in the registered courses. Machine learning techniques can be utilized for students' grades prediction in different courses. Such techniques would help students to improve their performance based on predicted grades and would enable instructors to identify such individuals who might need assistance in the courses. In this paper, we use Collaborative Filtering (CF), Matrix Factorization (MF), and Restricted Boltzmann Machines (RBM) techniques to systematically analyze a real-world data collected from Information Technology University (ITU), Lahore, Pakistan. We evaluate the academic performance of ITU students who got admission in the bachelor's degree program in ITU's Electrical Engineering department. The RBM technique is found to be better than the other techniques used in predicting the students' performance in the particular course. | @cite_9 proposed matrix factorization models for predicting student performance of Algebra and Bridge to Algebra courses. The factorization techniques are useful in case of sparse data and absence of students' background knowledge and tasks. They split the data into trainset and testset. The data represents the log files of interactions between students and computer aided tutoring systems. @cite_4 extended the research and used tensor-based factorization to predict student success. They formulated the problem of predicting student performance as a recommender system problem and proposed tensor-based factorization techniques to add the temporal effect of student performance. The system saves success failure logs of students on exercises as they interact with the system. | {
"cite_N": [
"@cite_9",
"@cite_4"
],
"mid": [
"123693119",
"152180625"
],
"abstract": [
"Recommender systems are widely used in many areas, especially in ecommerce. Recently, they are also applied in e-learning for recommending learning objects (e.g. papers) to students. This chapter introduces state-of-the-art recommender system techniques which can be used not only for recommending objects like tasks exercises to the students but also for predicting student performance. We formulate the problem of predicting student performance as a recommender system problem and present matrix factorization methods, which are currently known as the most effective recommendation approaches, to implicitly take into account the prevailing latent factors (e.g. “slip” and “guess”) for predicting student performance. As a learner’s knowledge improves over time, too, we propose tensor factorization methods to take the temporal effect into account. Finally, some experimental results and discussions are provided to validate the proposed approach.",
"Recommender systems are widely used in many areas, especially in e-commerce. Recently, they are also applied in technology enhanced learning such as recommending resources (e.g. papers, books,...) to the learners (students). In this study, we propose using state-of-the-art recommender system techniques for predicting student performance. We introduce and formulate the problem of predicting student performance in the context of recommender systems. We present the matrix factorization methods, known as the most effective recommendation approaches, to implicitly take into account the latent factors, e.g. “slip” and “guess”, in predicting student performance. Moreover, the knowledge of the learners has been improved over the time, thus, we propose tensor factorization methods to take the temporal effect into account. Experimental results show that the proposed approaches can improve the prediction results."
]
} |
1708.08744 | 2750637936 | In higher educational institutes, many students have to struggle hard to complete different courses since there is no dedicated support offered to students who need special attention in the registered courses. Machine learning techniques can be utilized for students' grades prediction in different courses. Such techniques would help students to improve their performance based on predicted grades and would enable instructors to identify such individuals who might need assistance in the courses. In this paper, we use Collaborative Filtering (CF), Matrix Factorization (MF), and Restricted Boltzmann Machines (RBM) techniques to systematically analyze a real-world data collected from Information Technology University (ITU), Lahore, Pakistan. We evaluate the academic performance of ITU students who got admission in the bachelor's degree program in ITU's Electrical Engineering department. The RBM technique is found to be better than the other techniques used in predicting the students' performance in the particular course. | Grade prediction accuracy using Matrix Factorization (MF) method degrades when dealing with small sample sizes. @cite_30 investigated different recommender system techniques to accurately predict the students' next term course grades as well as within the class assessment performance of George Mason University (GMU), University of Minnesota (UMN) and Stanford University (SU). Their study revealed that both Personalized Multi-Linear Regression models (PLMR) and advance Matrix Factorization (MF) techniques could predict next term grades with lower error rate than traditional methods. PLMR was also useful for predicting grades on assessments within a traditional class or online course by incorporating features captured through students' interaction with LMS and MOOC server logs. | {
"cite_N": [
"@cite_30"
],
"mid": [
"2341421431"
],
"abstract": [
"To help solve the ongoing problem of student retention, new expected performance-prediction techniques are needed to facilitate degree planning and determine who might be at risk of failing or dropping a class. Personalized multiregression and matrix factorization approaches based on recommender systems, initially developed for e-commerce applications, accurately forecast students' grades in future courses as well as on in-class assessments."
]
} |
1708.08744 | 2750637936 | In higher educational institutes, many students have to struggle hard to complete different courses since there is no dedicated support offered to students who need special attention in the registered courses. Machine learning techniques can be utilized for students' grades prediction in different courses. Such techniques would help students to improve their performance based on predicted grades and would enable instructors to identify such individuals who might need assistance in the courses. In this paper, we use Collaborative Filtering (CF), Matrix Factorization (MF), and Restricted Boltzmann Machines (RBM) techniques to systematically analyze a real-world data collected from Information Technology University (ITU), Lahore, Pakistan. We evaluate the academic performance of ITU students who got admission in the bachelor's degree program in ITU's Electrical Engineering department. The RBM technique is found to be better than the other techniques used in predicting the students' performance in the particular course. | Educational Data Mining utilizes data mining techniques to discover novel knowledge originating in educational settings @cite_19 . EDM can be used for decision making in refining repetitive curricula and admission criteria of educational institutions @cite_16 . @cite_14 applied the EDM approach to analyze the effects of core Computer Science courses and provide novel information for refining repetitive curricula to enhance the success rate of the students. They utilized the historical log file of all the students of the Department of Mathematical Information Technology (DMIT) at the University of Jyväskylä in Finland. They analyzed patterns observed in the historical log file from the student database for enhanced profiling of the core courses and the indication of study skills that support timely and successful graduation. They trained multilayer perceptron neural network model with cross-validation to demonstrate the constructed nonlinear regression model. In their study, they found that the general learning capabilities can better predict the students' success than specific IT skills. | {
"cite_N": [
"@cite_19",
"@cite_14",
"@cite_16"
],
"mid": [
"1539947443",
"1876282706",
"2155587956"
],
"abstract": [
"We review the history and current trends in the field of Educational Data Mining (EDM). We consider the methodological profile of research in the early years of EDM, compared to in 2008 and 2009, and discuss trends and shifts in the research conducted by this community. In particular, we discuss the increased emphasis on prediction, the emergence of work using existing models to make scientific discoveries (“discovery with models”), and the reduction in the frequency of relationship mining within the EDM community. We discuss two ways that researchers have attempted to categorize the diversity of research in educational data mining research, and review the types of research problems that these methods have been used to address. The mostcited papers in EDM between 1995 and 2005 are listed, and their influence on the EDM community (and beyond the EDM community) is discussed.",
"Curricula for Computer Science degrees are characterized by the strong occupational orientation of the discipline. In the BSc degree structure, with clearly separated CS core studies, learning skills between these and other required studies may vary a lot, showing in student’s overall performance. To analyze such a situation, we apply nonstandard educational data mining techniques on a preprocessed log file of the passed courses. The joint variation of the course grades are studied through correlation analysis while the intrinsic groups of students are created and analyzed using a special clustering technique. Since not all students have attended all the courses, there is a nonstructured sparsity pattern to cope with. Finally, multilayer perceptron neural network with cross-validation based generalization assurance is trained and analyzed using analytic mean sensitivity to explain the nonlinear regression model constructed. Local (within-methods) and global (between-methods) triangulation of different analysis methods is argued to improve the technical soundness of the presented approaches, giving more confidence on our final conclusion that general learning capabilities predict the success of students better than specific IT skills obtained as part of the core studies.",
"Educational Data Mining (EDM) is an emerging multidisciplinary research area, in which methods and techniques for exploring data originating from various educational information systems have been developed. EDM is both a learning science, as well as a rich application area for data mining, due to the growing availability of educational data. EDM contributes to the study of how students learn, and the settings in which they learn. It enables data-driven decision making for improving the current educational practice and learning material. We present a brief overview of EDM and introduce four selected EDM papers representing a crosscut of different application areas for data mining in education."
]
} |
1708.08744 | 2750637936 | In higher educational institutes, many students have to struggle hard to complete different courses since there is no dedicated support offered to students who need special attention in the registered courses. Machine learning techniques can be utilized for students' grades prediction in different courses. Such techniques would help students to improve their performance based on predicted grades and would enable instructors to identify such individuals who might need assistance in the courses. In this paper, we use Collaborative Filtering (CF), Matrix Factorization (MF), and Restricted Boltzmann Machines (RBM) techniques to systematically analyze a real-world data collected from Information Technology University (ITU), Lahore, Pakistan. We evaluate the academic performance of ITU students who got admission in the bachelor's degree program in ITU's Electrical Engineering department. The RBM technique is found to be better than the other techniques used in predicting the students' performance in the particular course. | Dropout early warning systems help higher education institutions to identify students at risk, and to identify interventions that may help to increase the student retention rate of the institutes. @cite_34 utilized the Wisconsin DEWS approach to predict the student dropout risk. They introduced flexible series of DEWS software modules that can adapt to new data, new algorithms, and new outcome variables to predict the dropout risk as well as impute key predictors. | {
"cite_N": [
"@cite_34"
],
"mid": [
"161938675"
],
"abstract": [
"The state of Wisconsin has one of the highest four year graduation rates in the nation, but deep disparities among student subgroups remain. To address this the state has created the Wisconsin Dropout Early Warning System (DEWS), a predictive model of student dropout risk for students in grades six through nine. The Wisconsin DEWS is in use statewide and currently provides predictions on the likelihood of graduation for over 225,000 students. DEWS represents a novel statistical learning based approach to the challenge of assessing the risk of non-graduation for students and provides highly accurate predictions for students in the middle grades without expanding beyond mandated administrative data collections. Similar dropout early warning systems are in place in many jurisdictions across the country. Prior research has shown that in many cases the indicators used by such systems do a poor job of balancing the trade off between correct classification of likely dropouts and false-alarm (, 2013). Building on this work, DEWS uses the receiver-operating characteristic (ROC) metric to identify the best possible set of statistical models for making predictions about individual students. This paper describes the DEWS approach and the software behind it, which leverages the open source statistical language R (R Core Team, 2013). As a result DEWS is a flexible series of software modules that can adapt to new data, new algorithms, and new outcome variables to not only predict dropout, but also impute key predictors as well. The design and implementation of each of these modules is described in detail as well as the open-source R package, EWStools, that serves as the core of DEWS (Knowles, 2014)."
]
} |
1708.08744 | 2750637936 | In higher educational institutes, many students have to struggle hard to complete different courses since there is no dedicated support offered to students who need special attention in the registered courses. Machine learning techniques can be utilized for students' grades prediction in different courses. Such techniques would help students to improve their performance based on predicted grades and would enable instructors to identify such individuals who might need assistance in the courses. In this paper, we use Collaborative Filtering (CF), Matrix Factorization (MF), and Restricted Boltzmann Machines (RBM) techniques to systematically analyze a real-world data collected from Information Technology University (ITU), Lahore, Pakistan. We evaluate the academic performance of ITU students who got admission in the bachelor's degree program in ITU's Electrical Engineering department. The RBM technique is found to be better than the other techniques used in predicting the students' performance in the particular course. | Hidden Markov model has been used widely to model student learning. @cite_23 investigated solutions of hidden Markov model and concluded that the utilization of a maximum likelihood test should be the preferred method for finding parameter values for the hidden Markov Model. @cite_1 in a separate study developed and analyzed a new fitting procedure for Bayesian Knowledge Tracing and concluded that empirical probabilities had the comparable predictive accuracy to that of expectation maximization. | {
"cite_N": [
"@cite_1",
"@cite_23"
],
"mid": [
"160448112",
"1737233464"
],
"abstract": [
"Student modeling is an important component of ITS research because it can help guide the behavior of a running tutor and help researchers understand how students learn. Due to its predictive accuracy, interpretability and ability to infer student knowledge, Corbett & Anderson’s Bayesian Knowledge Tracing is one of the most popular student models. However, researchers have discovered problems with some of the most popular methods of fitting it. These problems include: multiple sets of highly dissimilar parameters predicting the data equally well (identifiability), local minima, degenerate parameters, and computational cost during fitting. Some researchers have proposed new fitting procedures to combat these problems, but are more complex and not completely successful at eliminating the problems they set out to prevent. We instead fit parameters by estimating the mostly likely point that each student learned the skill, developing a new method that avoids the above problems while achieving similar predictive accuracy.",
"Bayesian knowledge tracing has been used widely to model student learning. However, the name “Bayesian knowledge tracing” has been applied to two related, but distinct, models: The first is the Bayesian knowledge tracing Markov chain which predicts the student-averaged probability of a correct application of a skill. We present an analytical solution to this model and show that it is a function of three parameters and has the functional form of an exponential. The second form is the Bayesian knowledge tracing hidden Markov model which can use the individual student's performance at each opportunity to apply a skill to update the conditional probability that the student has learned that skill. We use a fixed point analysis to study solutions of this model and find a range of parameters where it has the desired behavior. *Revised version, Feb 2017"
]
} |
1708.08288 | 2749568986 | Abstract We address the problem of transferring the style of a headshot photo to face images. Existing methods using a single exemplar lead to inaccurate results when the exemplar does not contain sufficient stylized facial components for a given photo. In this work, we propose an algorithm to stylize face images using multiple exemplars containing different subjects in the same style. Patch correspondences between an input photo and multiple exemplars are established using a Markov Random Field (MRF), which enables accurate local energy transfer via Laplacian stacks. As image patches from multiple exemplars are used, the boundaries of facial components on the target image are inevitably inconsistent. The artifacts are removed by a post-processing step using an edge-preserving filter. Experimental results show that the proposed algorithm consistently produces visually pleasing results. | : These methods typically learn a mapping function using one exemplar to adjust the tone and lighting of the input photo. In @cite_22 , a transformation function is estimated over the entire image to map one distribution into another for color transfer. A multiple layer style transfer method is proposed in @cite_20 where an input image is decomposed into base and detail layers where the style is transferred independently . Further improvement is made in @cite_19 where a multi-scale approach is presented to reduce artifacts through an image pyramid. In @cite_4 , a color grading approach is developed by using color distribution transfer. A graph regularization for color processing is proposed in @cite_23 . To reduce time complexity, an efficient method is proposed in @cite_0 based on the generalized patchmatch algorithm @cite_24 . It uses a holistic non-linear parametric color model to address dense correspondence problem. We note these algorithms are effective in transferring image styles holistically at the expense of capturing fine details, which are well transferred using the proposed method. | {
"cite_N": [
"@cite_4",
"@cite_22",
"@cite_0",
"@cite_19",
"@cite_24",
"@cite_23",
"@cite_20"
],
"mid": [
"2006957355",
"2141015396",
"2106505277",
"",
"1763426478",
"2140124895",
"2147821504"
],
"abstract": [
"This article proposes an original method for grading the colours between different images or shots. The first stage of the method is to find a one-to-one colour mapping that transfers the palette of an example target picture to the original picture. This is performed using an original and parameter free algorithm that is able to transform any N-dimensional probability density function into another one. The proposed algorithm is iterative, non-linear and has a low computational cost. Applying the colour mapping on the original picture allows reproducing the same 'feel' as the target picture, but can also increase the graininess of the original picture, especially if the colour dynamic of the two pictures is very different. The second stage of the method is to reduce this grain artefact through an efficient post-processing algorithm that intends to preserve the gradient field of the original picture.",
"This article proposes an original method to estimate a continuous transformation that maps one N-dimensional distribution to another. The method is iterative, non-linear, and is shown to converge. Only 1D marginal distribution is used in the estimation process, hence involving low computation costs. As an illustration this mapping is applied to color transfer between two images of different contents. The paper also serves as a central focal point for collecting together the research activity in this area and relating it to the important problem of automated color grading",
"This paper presents a new efficient method for recovering reliable local sets of dense correspondences between two images with some shared content. Our method is designed for pairs of images depicting similar regions acquired by different cameras and lenses, under non-rigid transformations, under different lighting, and over different backgrounds. We utilize a new coarse-to-fine scheme in which nearest-neighbor field computations using Generalized PatchMatch [ 2010] are interleaved with fitting a global non-linear parametric color model and aggregating consistent matching regions using locally adaptive constraints. Compared to previous correspondence approaches, our method combines the best of two worlds: It is dense, like optical flow and stereo reconstruction methods, and it is also robust to geometric and photometric variations, like sparse feature matching. We demonstrate the usefulness of our method using three applications for automatic example-based photograph enhancement: adjusting the tonal characteristics of a source image to match a reference, transferring a known mask to a new image, and kernel estimation for image deblurring.",
"",
"PatchMatch is a fast algorithm for computing dense approximate nearest neighbor correspondences between patches of two image regions [1]. This paper generalizes PatchMatch in three ways: (1) to find k nearest neighbors, as opposed to just one, (2) to search across scales and rotations, in addition to just translations, and (3) to match using arbitrary descriptors and distances, not just sum-of-squared-differences on patch colors. In addition, we offer new search and parallelization strategies that further accelerate the method, and we show performance improvements over standard kd-tree techniques across a variety of inputs. In contrast to many previous matching algorithms, which for efficiency reasons have restricted matching to sparse interest points, or spatially proximate matches, our algorithm can efficiently find global, dense matches, even while matching across all scales and rotations. This is especially useful for computer vision applications, where our algorithm can be used as an efficient general-purpose component. We explore a variety of vision applications: denoising, finding forgeries by detecting cloned regions, symmetry detection, and object detection.",
"Nowadays color image processing is an essential issue in computer vision. Variational formulations provide a framework for color image restoration, smoothing and segmentation problems. The solutions of variational models can be obtained by minimizing appropriate energy functions and this minimization is usually performed by continuous partial differential equations (PDEs). The problem is usually considered as a regularization matter which minimizes a smoothness plus a regularization term. In this paper, we propose a general discrete regularization framework defined on weighted graphs of arbitrary topologies which can be seen as a discrete analogue of classical regularization theory. The smoothness term of the regularization uses a discrete definition of the p-Laplace operator. With this formulation, we propose a family of fast and simple anisotropic linear and nonlinear filters which do not involve PDEs. The proposed approach can be useful to process color images for restoration, denoising and segmentation purposes.",
"We introduce a new approach to tone management for photographs. Whereas traditional tone-mapping operators target a neutral and faithful rendition of the input image, we explore pictorial looks by controlling visual qualities such as the tonal balance and the amount of detail. Our method is based on a two-scale non-linear decomposition of an image. We modify the different layers based on their histograms and introduce a technique that controls the spatial variation of detail. We introduce a Poisson correction that prevents potential gradient reversal and preserves detail. In addition to directly controlling the parameters, the user can transfer the look of a model photograph to the picture being edited."
]
} |
1708.08288 | 2749568986 | Abstract We address the problem of transferring the style of a headshot photo to face images. Existing methods using a single exemplar lead to inaccurate results when the exemplar does not contain sufficient stylized facial components for a given photo. In this work, we propose an algorithm to stylize face images using multiple exemplars containing different subjects in the same style. Patch correspondences between an input photo and multiple exemplars are established using a Markov Random Field (MRF), which enables accurate local energy transfer via Laplacian stacks. As image patches from multiple exemplars are used, the boundaries of facial components on the target image are inevitably inconsistent. The artifacts are removed by a post-processing step using an edge-preserving filter. Experimental results show that the proposed algorithm consistently produces visually pleasing results. | : These methods transfer the color and tone based on the distributions on the exemplars. In @cite_9 , a local method is proposed for regional color transfer between two natural images by probabilistic segmentation, and a scheme based on expectation maximization is proposed to impose spatial and color smoothness. An exemplar-based style transfer method is proposed in @cite_5 where local affine color transformation model is developed to render natural images during different time of the day. In addition to color or tone transfer, numerous face photo decomposition methods based on edge-preserving filters @cite_14 @cite_8 @cite_16 are developed for makeup transfer @cite_2 and relighting @cite_18 . From an identity-specific collection of face images, an algorithm is developed to enhance low-quality photos based on high-quality ones by exploiting holistic and face-specific regions (e.g., deblurring, light transfer, and super resolution) @cite_3 . The training and input photos used in @cite_3 are from the same subject, and their goal is for image enhancement. | {
"cite_N": [
"@cite_18",
"@cite_14",
"@cite_8",
"@cite_9",
"@cite_3",
"@cite_2",
"@cite_5",
"@cite_16"
],
"mid": [
"2086665547",
"2141957843",
"2008719012",
"2160530465",
"2160777052",
"2121293528",
"2019969451",
"2125188192"
],
"abstract": [
"This article proposes a novel image-based method to transfer illumination from a reference face image to a target face image through edge-preserving filters. According to our method, only a single reference image, without any knowledge of the 3D geometry or material information of the target face, is needed. We first decompose the lightness layers of the reference and the target images into large-scale and detail layers through weighted least square (WLS) filter after face alignment. The large-scale layer of the reference image is filtered with the guidance of the target image. Adaptive parameter selection schemes for the edge-preserving filters is proposed in the above two filtering steps. The final relit result is obtained by replacing the large-scale layer of the target image with that of the reference image. We acquire convincing relit result on numerous target and reference face images with different lighting effects and genders. Comparisons with previous work show that our method is less affected by geometry differences and can preserve better the identification structure and skin color of the target face.",
"Many recent computational photography techniques decompose an image into a piecewise smooth base layer, containing large scale variations in intensity, and a residual detail layer capturing the smaller scale details in the image. In many of these applications, it is important to control the spatial scale of the extracted details, and it is often desirable to manipulate details at multiple scales, while avoiding visual artifacts. In this paper we introduce a new way to construct edge-preserving multi-scale image decompositions. We show that current basedetail decomposition techniques, based on the bilateral filter, are limited in their ability to extract detail at arbitrary scales. Instead, we advocate the use of an alternative edge-preserving smoothing operator, based on the weighted least squares optimization framework, which is particularly well suited for progressive coarsening of images and for multi-scale detail extraction. After describing this operator, we show how to use it to construct edge-preserving multi-scale decompositions, and compare it to the bilateral filter, as well as to other schemes. Finally, we demonstrate the effectiveness of our edge-preserving decompositions in the context of LDR and HDR tone mapping, detail enhancement, and other applications.",
"This paper formulates both the median filter and bilateral filter as a cost volume aggregation problem whose computational complexity is independent of the filter kernel size. Unlike most of the previous works, the proposed framework results in a general bilateral filter that can have arbitrary spatial @math 1 and arbitrary range filter kernels. This bilateral filter takes about 3.5 s to exactly filter a one megapixel 8-bit grayscale image on a 3.2 GHz Intel Core i7 CPU. In practice, the intensity range and spatial domain can be downsampled to improve the efficiency. This compression can maintain very high accuracy (e.g., 40 dB) but over @math 100× faster.",
"We address the problem of regional color transfer between two natural images by probabilistic segmentation. We use a new expectation-maximization (EM) scheme to impose both spatial and color smoothness to infer natural connectivity among pixels. Unlike previous work, our method takes local color information into consideration, and segment image with soft region boundaries for seamless color transfer and compositing. Our modified EM method has two advantages in color manipulation: first, subject to different levels of color smoothness in image space, our algorithm produces an optimal number of regions upon convergence, where the color statistics in each region can be adequately characterized by a component of a Gaussian mixture model (GMM). Second, we allow a pixel to fall in several regions according to our estimated probability distribution in the EM step, resulting in a transparency-like ratio for compositing different regions seamlessly. Hence, natural color transition across regions can be achieved, where the necessary intra-region and inter-region smoothness are enforced without losing original details. We demonstrate results on a variety of applications including image deblurring, enhanced color transfer, and colorizing gray scale images. Comparisons with previous methods are also presented.",
"We describe a framework for improving the quality of personal photos by using a person's favorite photographs as examples. We observe that the majority of a person's photographs include the faces of a photographer's family and friends and often the errors in these photographs are the most disconcerting. We focus on correcting these types of images and use common faces across images to automatically perform both global and face-specific corrections. Our system achieves this by using face detection to align faces between “good” and “bad” photos such that properties of the good examples can be used to correct a bad photo. These “personal” photos provide strong guidance for a number of operations and, as a result, enable a number of high-quality image processing operations. We illustrate the power and generality of our approach by presenting a novel deblurring algorithm, and we show corrections that perform sharpening, superresolution, in-painting of over- and underexposured regions, and white-balancing.",
"This paper introduces an approach of creating face makeup upon a face image with another image as the style example. Our approach is analogous to physical makeup, as we modify the color and skin detail while preserving the face structure. More precisely, we first decompose the two images into three layers: face structure layer, skin detail layer, and color layer. Thereafter, we transfer information from each layer of one image to corresponding layer of the other image. One major advantage of the proposed method lies in that only one example image is required. This renders face makeup by example very convenient and practical. Equally, this enables some additional interesting applications, such as applying makeup by a portraiture. The experiment results demonstrate the effectiveness of the proposed approach in faithfully transferring makeup.",
"We introduce \"time hallucination\": synthesizing a plausible image at a different time of day from an input image. This challenging task often requires dramatically altering the color appearance of the picture. In this paper, we introduce the first data-driven approach to automatically creating a plausible-looking photo that appears as though it were taken at a different time of day. The time of day is specified by a semantic time label, such as \"night\". Our approach relies on a database of time-lapse videos of various scenes. These videos provide rich information about the variations in color appearance of a scene throughout the day. Our method transfers the color appearance from videos with a similar scene as the input photo. We propose a locally affine model learned from the video for the transfer, allowing our model to synthesize new color data while retaining image details. We show that this model can hallucinate a wide range of different times of day. The model generates a large sparse linear system, which can be solved by off-the-shelf solvers. We validate our methods by synthesizing transforming photos of various outdoor scenes to four times of interest: daytime, the golden hour, the blue hour, and nighttime.",
"In this paper, we propose a novel explicit image filter called guided filter. Derived from a local linear model, the guided filter computes the filtering output by considering the content of a guidance image, which can be the input image itself or another different image. The guided filter can be used as an edge-preserving smoothing operator like the popular bilateral filter [1], but it has better behaviors near edges. The guided filter is also a more generic concept beyond smoothing: It can transfer the structures of the guidance image to the filtering output, enabling new filtering applications like dehazing and guided feathering. Moreover, the guided filter naturally has a fast and nonapproximate linear time algorithm, regardless of the kernel size and the intensity range. Currently, it is one of the fastest edge-preserving filters. Experiments show that the guided filter is both effective and efficient in a great variety of computer vision and computer graphics applications, including edge-aware smoothing, detail enhancement, HDR compression, image matting feathering, dehazing, joint upsampling, etc."
]
} |
1708.08288 | 2749568986 | Abstract We address the problem of transferring the style of a headshot photo to face images. Existing methods using a single exemplar lead to inaccurate results when the exemplar does not contain sufficient stylized facial components for a given photo. In this work, we propose an algorithm to stylize face images using multiple exemplars containing different subjects in the same style. Patch correspondences between an input photo and multiple exemplars are established using a Markov Random Field (MRF), which enables accurate local energy transfer via Laplacian stacks. As image patches from multiple exemplars are used, the boundaries of facial components on the target image are inevitably inconsistent. The artifacts are removed by a post-processing step using an edge-preserving filter. Experimental results show that the proposed algorithm consistently produces visually pleasing results. | : A local method that transfers the face style of an exemplar to the input face image is proposed in @cite_12 . It first generates dense correspondence between an input photo and one selected exemplar. Then it decomposes each image into a Laplacian stack before transferring the local energy in each image frequency subband within each layer. Finally all the stacks are aggregated to generate the output image. Since the style represented by the local energy is precisely transferred in multiple layers, it has the advantage to handle detailed facial components. Compared to the holistic methods, local approaches can better capture the region details and thus facilitate face stylization . However, if the components appeared in the exemplars and the input photos are significantly different, the resulting images are likely to contain undesired effects. In this work, we use multiple exemplars to solve this problem. | {
"cite_N": [
"@cite_12"
],
"mid": [
"2106395586"
],
"abstract": [
"Headshot portraits are a popular subject in photography but to achieve a compelling visual style requires advanced skills that a casual photographer will not have. Further, algorithms that automate or assist the stylization of generic photographs do not perform well on headshots due to the feature-specific, local retouching that a professional photographer typically applies to generate such portraits. We introduce a technique to transfer the style of an example headshot photo onto a new one. This can allow one to easily reproduce the look of renowned artists. At the core of our approach is a new multiscale technique to robustly transfer the local statistics of an example portrait onto a new one. This technique matches properties such as the local contrast and the overall lighting direction while being tolerant to the unavoidable differences between the faces of two different people. Additionally, because artists sometimes produce entire headshot collections in a common style, we show how to automatically find a good example to use as a reference for a given portrait, enabling style transfer without the user having to search for a suitable example for each input. We demonstrate our approach on data taken in a controlled environment as well as on a large set of photos downloaded from the Internet. We show that we can successfully handle styles by a variety of different artists."
]
} |
1708.08282 | 2753295253 | In school, a teacher plays an important role in various classroom teaching patterns. Likewise to this human learning activity, the learning using privileged information (LUPI) paradigm provides additional information generated by the teacher to 'teach' learning models during the training stage. Therefore, this novel learning paradigm is a typical Teacher-Student Interaction mechanism. This paper is the first to present a random vector functional link network based on the LUPI paradigm, called RVFL+. Rather than simply combining two existing approaches, the newly-derived RVFL+ fills the gap between classical randomized neural networks and the newfashioned LUPI paradigm, which offers an alternative way to train RVFL networks. Moreover, the proposed RVFL+ can perform in conjunction with the kernel trick for highly complicated nonlinear feature learning, which is termed KRVFL+. Furthermore, the statistical property of the proposed RVFL+ is investigated, and we present a sharp and high-quality generalization error bound based on the Rademacher complexity. Competitive experimental results on 14 real-world datasets illustrate the great effectiveness and efficiency of the novel RVFL+ and KRVFL+, which can achieve better generalization performance than state-of-the-art methods. | Over the last three decades, randomization based methods, including random projection @cite_11 , random forests @cite_43 , bagging @cite_28 , stochastic configuration networks (SCN) @cite_27 @cite_34 , and random vector functional link networks (RVFL) @cite_18 , etc., play important roles in machine learning community. We refer to @cite_48 @cite_49 for great surveys of the randomized neural networks. | {
"cite_N": [
"@cite_18",
"@cite_28",
"@cite_48",
"@cite_43",
"@cite_27",
"@cite_49",
"@cite_34",
"@cite_11"
],
"mid": [
"1996640396",
"100189773",
"2753648062",
"",
"2593382986",
"2286961399",
"",
"2154209944"
],
"abstract": [
"Abstract In this paper we explore and discuss the learning and generalization characteristics of the random vector version of the Functional-link net and compare these with those attainable with the GDR algorithm. This is done for a well-behaved deterministic function and for real-world data. It seems that ‘ overtraining ’ occurs for stochastic mappings. Otherwise there is saturation of training.",
"",
"In big data fields, with increasing computing capability, artificial neural networks have shown great strength in solving data classification and regression problems. The traditional training of neural networks depends generally on the error back propagation method to iteratively tune all the parameters. When the number of hidden layers increases, this kind of training has many problems such as slow convergence, time consuming, and local minima. To avoid these problems, neural networks with random weights (NNRW) are proposed in which the weights between the hidden layer and input layer are randomly selected and the weights between the output layer and hidden layer are obtained analytically. Researchers have shown that NNRW has much lower training complexity in comparison with the traditional training of feed-forward neural networks. This paper objectively reviews the advantages and disadvantages of NNRW model, tries to reveal the essence of NNRW, gives our comments and remarks on NNRW, and provides some useful guidelines for users to choose a mechanism to train a feed-forward neural network.",
"",
"This paper contributes to the development of randomized methods for neural networks. The proposed learner model is generated incrementally by stochastic configuration (SC) algorithms, termed SC networks (SCNs). In contrast to the existing randomized learning algorithms for single layer feed-forward networks, we randomly assign the input weights and biases of the hidden nodes in the light of a supervisory mechanism, and the output weights are analytically evaluated in either a constructive or selective manner. As fundamentals of SCN-based data modeling techniques, we establish some theoretical results on the universal approximation property. Three versions of SC algorithms are presented for data regression and classification problems in this paper. Simulation results concerning both data regression and classification indicate some remarkable merits of our proposed SCNs in terms of less human intervention on the network size setting, the scope adaptation of random parameters, fast learning, and sound generalization.",
"As a powerful tool for data regression and classification, neural networks have received considerable attention from researchers in fields such as machine learning, statistics, computer vision and so on. There exists a large body of research work on network training, among which most of them tune the parameters iteratively. Such methods often suffer from local minima and slow convergence. It has been shown that randomization based training methods can significantly boost the performance or efficiency of neural networks. Among these methods, most approaches use randomization either to change the data distributions, and or to fix a part of the parameters or network configurations.źThis article presents a comprehensive survey of the earliest work and recent advances as well as some suggestions for future research.",
"",
"Inspired by theories of sparse representation and compressed sensing, this paper presents a simple, novel, yet very powerful approach for texture classification based on random projection, suitable for large texture database applications. At the feature extraction stage, a small set of random features is extracted from local image patches. The random features are embedded into a bag--of-words model to perform texture classification; thus, learning and classification are carried out in a compressed domain. The proposed unconventional random feature extraction is simple, yet by leveraging the sparse nature of texture images, our approach outperforms traditional feature extraction methods which involve careful design and complex steps. We have conducted extensive experiments on each of the CUReT, the Brodatz, and the MSRC databases, comparing the proposed approach to four state-of-the-art texture classification methods: Patch, Patch-MRF, MR8, and LBP. We show that our approach leads to significant improvements in classification accuracy and reductions in feature dimensionality."
]
} |
1708.08414 | 2952874307 | Various defense schemes --- which determine the presence of an attack on the in-vehicle network --- have recently been proposed. However, they fail to identify which Electronic Control Unit (ECU) actually mounted the attack. Clearly, pinpointing the attacker ECU is essential for fast efficient forensic, isolation, security patch, etc. To meet this need, we propose a novel scheme, called Viden (Voltage-based attacker identification), which can identify the attacker ECU by measuring and utilizing voltages on the in-vehicle network. The first phase of Viden, called ACK learning, determines whether or not the measured voltage signals really originate from the genuine message transmitter. Viden then exploits the voltage measurements to construct and update the transmitter ECUs' voltage profiles as their fingerprints. It finally uses the voltage profiles to identify the attacker ECU. Since Viden adapts its profiles to changes inside outside of the vehicle, it can pinpoint the attacker ECU under various conditions. Moreover, its efficiency and design-compliance with modern in-vehicle network implementations make Viden practical and easily deployable. Our extensive experimental evaluations on both a CAN bus prototype and two real vehicles have shown that Viden can accurately fingerprint ECUs based solely on voltage measurements and thus identify the attacker ECU with a low false identification rate of 0.2 . | A clock-based intrusion detection system (CIDS) was proposed in @cite_25 to detect intrusions by fingerprinting ECUs on CAN. CIDS derived the fingerprints by extracting the ECUs' clock skews from message arrival times. While the main objective of CIDS was to detect intrusions, the authors of @cite_25 mentioned that the thus-derived fingerprints may also be used for attacker identification, but only when attack messages are injected periodically . In other words, if the attacker transmits messages aperiodically , then CIDS cannot identify the attacker ECU, i.e., the adversary can evade CIDS as far as attacker identification is concerned. takes an entirely different approach: looking at attack messages from the perspective of ECUs' output voltages on CAN. This allows to accurately identify the attacker ECU irrespective of how and when the attacker injects its messages, which is crucial for attacker identification. | {
"cite_N": [
"@cite_25"
],
"mid": [
"2464037380"
],
"abstract": [
"As more software modules and external interfaces are getting added on vehicles, new attacks and vulnerabilities are emerging. Researchers have demonstrated how to compromise in-vehicle Electronic Control Units (ECUs) and control the vehicle maneuver. To counter these vulnerabilities, various types of defense mechanisms have been proposed, but they have not been able to meet the need of strong protection for safety-critical ECUs against in-vehicle network attacks. To mitigate this deficiency, we propose an anomaly-based intrusion detection system (IDS), called Clock-based IDS (CIDS). It measures and then exploits the intervals of periodic in-vehicle messages for fingerprinting ECUs. The thus-derived fingerprints are then used for constructing a baseline of ECUs' clock behaviors with the Recursive Least Squares (RLS) algorithm. Based on this baseline, CIDS uses Cumulative Sum (CUSUM) to detect any abnormal shifts in the identification errors - a clear sign of intrusion. This allows quick identification of in-vehicle network intrusions with a low false-positive rate of 0.055 . Unlike state-of-the-art IDSs, if an attack is detected, CIDS's fingerprinting of ECUs also facilitates a rootcause analysis; identifying which ECU mounted the attack. Our experiments on a CAN bus prototype and on real vehicles have shown CIDS to be able to detect a wide range of in-vehicle network attacks."
]
} |
1708.08414 | 2952874307 | Various defense schemes --- which determine the presence of an attack on the in-vehicle network --- have recently been proposed. However, they fail to identify which Electronic Control Unit (ECU) actually mounted the attack. Clearly, pinpointing the attacker ECU is essential for fast efficient forensic, isolation, security patch, etc. To meet this need, we propose a novel scheme, called Viden (Voltage-based attacker identification), which can identify the attacker ECU by measuring and utilizing voltages on the in-vehicle network. The first phase of Viden, called ACK learning, determines whether or not the measured voltage signals really originate from the genuine message transmitter. Viden then exploits the voltage measurements to construct and update the transmitter ECUs' voltage profiles as their fingerprints. It finally uses the voltage profiles to identify the attacker ECU. Since Viden adapts its profiles to changes inside outside of the vehicle, it can pinpoint the attacker ECU under various conditions. Moreover, its efficiency and design-compliance with modern in-vehicle network implementations make Viden practical and easily deployable. Our extensive experimental evaluations on both a CAN bus prototype and two real vehicles have shown that Viden can accurately fingerprint ECUs based solely on voltage measurements and thus identify the attacker ECU with a low false identification rate of 0.2 . | Instead of fingerprinting ECUs based on message timings, as in , some researchers also proposed to fingerprint them via voltage measurements. The authors of @cite_24 used the Mean Squared Error (MSE) of voltage measurements as fingerprints of ECUs. However, they were shown to be valid only for the voltages measured during the transmission of CAN message IDs, and more importantly when voltages were measured on a low-speed (10Kbps) CAN bus; this is far from contemporary vehicles that usually operate on a 500Kbps CAN bus. | {
"cite_N": [
"@cite_24"
],
"mid": [
"2072724180"
],
"abstract": [
"The CAN (Controller Area Network) bus, i.e., the de facto standard for connecting ECUs inside cars, is increasingly becoming exposed to some of the most sophisticated security threats. Due to its broadcast nature and ID oriented communication, each node is sightless in regards to the source of the received messages and assuring source identification is an uneasy challenge. While recent research has focused on devising security in CAN networks by the use of cryptography at the protocol layer, such solutions are not always an alternative due to increased communication and computational overheads, not to mention backward compatibility issues. In this work we set steps for a distinct approach, namely, we try to take authentication up to unique physical characteristics of the frames that are placed by each node on the bus. For this we analyze the frames by taking measurements of the voltage, filtering the signal and examining mean square errors and convolutions in order to uniquely identify each potential sender. Our experimental results show that distinguishing between certain nodes is clearly possible and by clever choices of transceivers and frame IDs each message can be precisely linked to its sender."
]
} |
1708.08414 | 2952874307 | Various defense schemes --- which determine the presence of an attack on the in-vehicle network --- have recently been proposed. However, they fail to identify which Electronic Control Unit (ECU) actually mounted the attack. Clearly, pinpointing the attacker ECU is essential for fast efficient forensic, isolation, security patch, etc. To meet this need, we propose a novel scheme, called Viden (Voltage-based attacker identification), which can identify the attacker ECU by measuring and utilizing voltages on the in-vehicle network. The first phase of Viden, called ACK learning, determines whether or not the measured voltage signals really originate from the genuine message transmitter. Viden then exploits the voltage measurements to construct and update the transmitter ECUs' voltage profiles as their fingerprints. It finally uses the voltage profiles to identify the attacker ECU. Since Viden adapts its profiles to changes inside outside of the vehicle, it can pinpoint the attacker ECU under various conditions. Moreover, its efficiency and design-compliance with modern in-vehicle network implementations make Viden practical and easily deployable. Our extensive experimental evaluations on both a CAN bus prototype and two real vehicles have shown that Viden can accurately fingerprint ECUs based solely on voltage measurements and thus identify the attacker ECU with a low false identification rate of 0.2 . | To overcome these difficulties, researchers proposed to extract other time and frequency domain features of voltage measurements (e.g., RMS amplitude) and use them as inputs for classification; more specifically, supervised (batch) learning algorithms (e.g., SVM) @cite_16 . This way, they were able to fingerprint ECUs with enhanced accuracy and was successful on high-speed CAN buses. However, this solution was neither practical nor attractive for attacker identification for the following reasons. First, it required not only a high sampling rate (2.5 GSamples sec), but also the use of the extended CAN frame format with 29-bit IDs, which is seldom used (due to its bandwidth waste) in contemporary vehicles; most vehicles use the standard format with 11-bit IDs. Moreover, since the modeling was done via batch learning, unpredictable changes in the CAN bus (e.g., temperature, battery level) and adversary's behaviors can lead to false identifications. These will be detailed later when we discuss the details of . | {
"cite_N": [
"@cite_16"
],
"mid": [
"2465629572"
],
"abstract": [
"In the last several decades, the automotive industry has come to incorporate the latest Information and Communications (ICT) technology, increasingly replacing mechanical components of vehicles with electronic components. These electronic control units (ECUs) communicate with each other in an in-vehicle network that makes the vehicle both safer and easier to drive. Controller Area Networks (CANs) are the current standard for such high quality in-vehicle communication. Unfortunately, however, CANs do not currently offer protection against security attacks. In particular, they do not allow for message authentication and hence are open to attacks that replay ECU messages for malicious purposes. Applying the classic cryptographic method of message authentication code (MAC) is not feasible since the CAN data frame is not long enough to include a sufficiently long MAC to provide effective authentication. In this paper, we propose a novel identification method, which works in the physical layer of an in-vehicle CAN network. Our method identifies ECUs using inimitable characteristics of signals enabling detection of a compromised or alien ECU being used in a replay attack. Unlike previous attempts to address security issues in the in-vehicle CAN network, our method works by simply adding a monitoring unit to the existing network, making it deployable in current systems and compliant with required CAN standards. Our experimental results show that the bit string and classification algorithm that we utilized yielded more accurate identification of compromised ECUs than any other method proposed to date. The false positive rate is more than 2 times lower than the method proposed by P.-S. This paper is also the first to identify potential attack models that systems should be able to detect."
]
} |
1708.08417 | 2753711514 | Many applications such as autonomous navigation, urban planning and asset monitoring, rely on the availability of accurate information about objects and their geolocations. In this paper we propose to automatically detect and compute the GPS coordinates of recurring stationary objects of interest using street view imagery. Our processing pipeline relies on two fully convolutional neural networks: the first segments objects in the images while the second estimates their distance from the camera. To geolocate all the detected objects coherently we propose a novel custom Markov Random Field model to perform objects triangulation. The novelty of the resulting pipeline is the combined use of monocular depth estimation and triangulation to enable automatic mapping of complex scenes with multiple visually similar objects of interest. We validate experimentally the effectiveness of our approach on two object classes: traffic lights and telegraph poles. The experiments report high object recall rates and GPS accuracy within 2 meters, which is comparable with the precision of single-frequency GPS receivers. | In the last decade a considerable effort has been directed towards intelligent use of street view imagery in multiple applied areas: mapping @cite_7 @cite_5 , image referencing @cite_24 , navigation @cite_11 , rendering and visualization @cite_14 @cite_22 , etc. Methods designed in @cite_11 @cite_5 use street view combined with aerial imagery to achieve fine-grained road segmentation, and object detection (trees), respectively. These methods rely on object discovery through street view imagery and recover position from aerial data. @cite_24 street view images are used as reference to geolocate query images by resorting to scene matching. The GSV imagery is employed in @cite_14 @cite_22 in conjunction with social media, like Twitter, to perform visualization and 3D rendering. | {
"cite_N": [
"@cite_14",
"@cite_22",
"@cite_7",
"@cite_24",
"@cite_5",
"@cite_11"
],
"mid": [
"2583330224",
"2485353833",
"2084636801",
"2103163130",
"2438072089",
"2464204616"
],
"abstract": [
"A visible area detection method that depends on online available information is proposed.Visual popularity and sentiment are calculated based on images and their accompanying text that people share in social mediaThe visual popularity information is presented in a 3D virtual environment using Unreal engine. Display Omitted This paper proposes to use a geotagged virtual world for the visualization of peoples visual interest and their sentiment as captured from their social network activities. Using mobile devices, people widely share their experiences and the things they find interesting through social networks. We experimentally show that accumulating information over a period of time from multiple social network users allows to efficiently map and visualize popular landmarks as found in cities such as Rome in Italy and Dublin in Ireland. The proposed approach is also sensitive to temporal and spatial events that attract visual attention. We visualize the calculated popularity on 3D virtual cities using game engine technologies.",
"This paper presents an immersive geo-spatial social media system for virtual and augmented reality environments. With the rapid growth of photo-sharing social media sites such as Flickr, Pinterest, and Instagram, geo-tagged photographs are now ubiquitous. However, the current systems for their navigation are unsatisfyingly one- or two-dimensional. In this paper, we present our prototype system, Social Street View, which renders the geo-tagged social media in its natural geo-spatial context provided by immersive maps, such as Google Street View. This paper presents new algorithms for fusing and laying out the social media in an aesthetically pleasing manner with geospatial renderings, validates them with respect to visual saliency metrics, suggests spatio-temporal filters, and presents a system architecture that is able to stream geo-tagged social media and render it across a range of display platforms spanning tablets, desktops, head-mounted displays, and large-area room-sized curved tiled displays. The paper concludes by exploring several potential use cases including immersive social storytelling, learning about culture and crowd-sourced tourism.",
"Low-vision and blind bus riders often rely on known physical landmarks to help locate and verify bus stop locations (e.g., by searching for an expected shelter, bench, or newspaper bin). However, there are currently few, if any, methods to determine this information a priori via computational tools or services. In this article, we introduce and evaluate a new scalable method for collecting bus stop location and landmark descriptions by combining online crowdsourcing and Google Street View (GSV). We conduct and report on three studies: (i) a formative interview study of 18 people with visual impairments to inform the design of our crowdsourcing tool, (ii) a comparative study examining differences between physical bus stop audit data and audits conducted virtually with GSV, and (iii) an online study of 153 crowd workers on Amazon Mechanical Turk to examine the feasibility of crowdsourcing bus stop audits using our custom tool with GSV. Our findings reemphasize the importance of landmarks in nonvisual navigation, demonstrate that GSV is a viable bus stop audit dataset, and show that minimally trained crowd workers can find and identify bus stop landmarks with 82.5p accuracy across 150 bus stop locations (87.3p with simple quality control).",
"Estimating geographic information from an image is an excellent, difficult high-level computer vision problem whose time has come. The emergence of vast amounts of geographically-calibrated image data is a great reason for computer vision to start looking globally - on the scale of the entire planet! In this paper, we propose a simple algorithm for estimating a distribution over geographic locations from a single image using a purely data-driven scene matching approach. For this task, we leverage a dataset of over 6 million GPS-tagged images from the Internet. We represent the estimated image location as a probability distribution over the Earthpsilas surface. We quantitatively evaluate our approach in several geolocation tasks and demonstrate encouraging performance (up to 30 times better than chance). We show that geolocation estimates can provide the basis for numerous other image understanding tasks such as population density estimation, land cover estimation or urban rural classification.",
"Each corner of the inhabited world is imaged from multiple viewpoints with increasing frequency. Online map services like Google Maps or Here Maps provide direct access to huge amounts of densely sampled, georeferenced images from street view and aerial perspective. There is an opportunity to design computer vision systems that will help us search, catalog and monitor public infrastructure, buildings and artifacts. We explore the architecture and feasibility of such a system. The main technical challenge is combining test time information from multiple views of each geographic location (e.g., aerial and street views). We implement two modules: det2geo, which detects the set of locations of objects belonging to a given category, and geo2cat, which computes the fine-grained category of the object at a given location. We introduce a solution that adapts state-of the-art CNN-based object detectors and classifiers. We test our method on \"Pasadena Urban Trees\", a new dataset of 80,000 trees with geographic and species annotations, and show that combining multiple views significantly improves both tree detection and tree species classification, rivaling human performance.",
"In this paper we present an approach to enhance existing maps with fine grained segmentation categories such as parking spots and sidewalk, as well as the number and location of road lanes. Towards this goal, we propose an efficient approach that is able to estimate these fine grained categories by doing joint inference over both, monocular aerial imagery, as well as ground images taken from a stereo camera pair mounted on top of a car. Important to this is reasoning about the alignment between the two types of imagery, as even when the measurements are taken with sophisticated GPS+IMU systems, this alignment is not sufficiently accurate. We demonstrate the effectiveness of our approach on a new dataset which enhances KITTI [8] with aerial images taken with a camera mounted on an airplane and flying around the city of Karlsruhe, Germany."
]
} |
1708.08417 | 2753711514 | Many applications such as autonomous navigation, urban planning and asset monitoring, rely on the availability of accurate information about objects and their geolocations. In this paper we propose to automatically detect and compute the GPS coordinates of recurring stationary objects of interest using street view imagery. Our processing pipeline relies on two fully convolutional neural networks: the first segments objects in the images while the second estimates their distance from the camera. To geolocate all the detected objects coherently we propose a novel custom Markov Random Field model to perform objects triangulation. The novelty of the resulting pipeline is the combined use of monocular depth estimation and triangulation to enable automatic mapping of complex scenes with multiple visually similar objects of interest. We validate experimentally the effectiveness of our approach on two object classes: traffic lights and telegraph poles. The experiments report high object recall rates and GPS accuracy within 2 meters, which is comparable with the precision of single-frequency GPS receivers. | combining test time information from multiple views of each geographic location (e.g., aerial and street views), dataset: Pasadena Urban Trees @cite_5 | {
"cite_N": [
"@cite_5"
],
"mid": [
"2438072089"
],
"abstract": [
"Each corner of the inhabited world is imaged from multiple viewpoints with increasing frequency. Online map services like Google Maps or Here Maps provide direct access to huge amounts of densely sampled, georeferenced images from street view and aerial perspective. There is an opportunity to design computer vision systems that will help us search, catalog and monitor public infrastructure, buildings and artifacts. We explore the architecture and feasibility of such a system. The main technical challenge is combining test time information from multiple views of each geographic location (e.g., aerial and street views). We implement two modules: det2geo, which detects the set of locations of objects belonging to a given category, and geo2cat, which computes the fine-grained category of the object at a given location. We introduce a solution that adapts state-of the-art CNN-based object detectors and classifiers. We test our method on \"Pasadena Urban Trees\", a new dataset of 80,000 trees with geographic and species annotations, and show that combining multiple views significantly improves both tree detection and tree species classification, rivaling human performance."
]
} |
1708.08417 | 2753711514 | Many applications such as autonomous navigation, urban planning and asset monitoring, rely on the availability of accurate information about objects and their geolocations. In this paper we propose to automatically detect and compute the GPS coordinates of recurring stationary objects of interest using street view imagery. Our processing pipeline relies on two fully convolutional neural networks: the first segments objects in the images while the second estimates their distance from the camera. To geolocate all the detected objects coherently we propose a novel custom Markov Random Field model to perform objects triangulation. The novelty of the resulting pipeline is the combined use of monocular depth estimation and triangulation to enable automatic mapping of complex scenes with multiple visually similar objects of interest. We validate experimentally the effectiveness of our approach on two object classes: traffic lights and telegraph poles. The experiments report high object recall rates and GPS accuracy within 2 meters, which is comparable with the precision of single-frequency GPS receivers. | joint inference over both, monocular aerial imagery, as well as ground images taken from a stereo camera pair mounted on top of a car @cite_11 | {
"cite_N": [
"@cite_11"
],
"mid": [
"2464204616"
],
"abstract": [
"In this paper we present an approach to enhance existing maps with fine grained segmentation categories such as parking spots and sidewalk, as well as the number and location of road lanes. Towards this goal, we propose an efficient approach that is able to estimate these fine grained categories by doing joint inference over both, monocular aerial imagery, as well as ground images taken from a stereo camera pair mounted on top of a car. Important to this is reasoning about the alignment between the two types of imagery, as even when the measurements are taken with sophisticated GPS+IMU systems, this alignment is not sufficiently accurate. We demonstrate the effectiveness of our approach on a new dataset which enhances KITTI [8] with aerial images taken with a camera mounted on an airplane and flying around the city of Karlsruhe, Germany."
]
} |
1708.08417 | 2753711514 | Many applications such as autonomous navigation, urban planning and asset monitoring, rely on the availability of accurate information about objects and their geolocations. In this paper we propose to automatically detect and compute the GPS coordinates of recurring stationary objects of interest using street view imagery. Our processing pipeline relies on two fully convolutional neural networks: the first segments objects in the images while the second estimates their distance from the camera. To geolocate all the detected objects coherently we propose a novel custom Markov Random Field model to perform objects triangulation. The novelty of the resulting pipeline is the combined use of monocular depth estimation and triangulation to enable automatic mapping of complex scenes with multiple visually similar objects of interest. We validate experimentally the effectiveness of our approach on two object classes: traffic lights and telegraph poles. The experiments report high object recall rates and GPS accuracy within 2 meters, which is comparable with the precision of single-frequency GPS receivers. | Image segmentation is one of the central tasks in object extraction. A multitude of approaches have been proposed to address this problem: starting from elementary pattern analysis techniques, such as Hough transform and local gradients, through feature extraction-based tools that rely on cascades of weak classifiers @cite_10 , to more advanced machine learning methods such as random forests, support vector machines and convolutional neural networks (CNNs) @cite_16 @cite_20 . The latter have recently pushed the machine vision techniques to new heights by allowing automatic feature selection and unprecedented capacity to efficiently learn from huge volumes of data. FCNNs are a natural form of CNNs when addressing dense segmentation since they allow location information to be retained through the decision process by relying solely on convolutional layers @cite_19 . Their output can be in the form of bounding boxes @cite_13 @cite_17 or segmentation maps. The latter are obtained via deconvolutions only @cite_19 or by resorting to conditional random fields @cite_15 . | {
"cite_N": [
"@cite_10",
"@cite_19",
"@cite_15",
"@cite_16",
"@cite_13",
"@cite_20",
"@cite_17"
],
"mid": [
"2164598857",
"2395611524",
"",
"2949650786",
"639708223",
"1686810756",
"2193145675"
],
"abstract": [
"This paper describes a machine learning approach for visual object detection which is capable of processing images extremely rapidly and achieving high detection rates. This work is distinguished by three key contributions. The first is the introduction of a new image representation called the \"integral image\" which allows the features used by our detector to be computed very quickly. The second is a learning algorithm, based on AdaBoost, which selects a small number of critical visual features from a larger set and yields extremely efficient classifiers. The third contribution is a method for combining increasingly more complex classifiers in a \"cascade\" which allows background regions of the image to be quickly discarded while spending more computation on promising object-like regions. The cascade can be viewed as an object specific focus-of-attention mechanism which unlike previous approaches provides statistical guarantees that discarded regions are unlikely to contain the object of interest. In the domain of face detection the system yields detection rates comparable to the best previous systems. Used in real-time applications, the detector runs at 15 frames per second without resorting to image differencing or skin color detection.",
"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, improve on the previous best result 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, the VGG net, and GoogLeNet) into fully convolutional networks and transfer their learned representations by fine-tuning 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 networks achieve improved segmentation of PASCAL VOC (30 relative improvement to 67.2 mean IU on 2012), NYUDv2, SIFT Flow, and PASCAL-Context, while inference takes one tenth of a second for a typical image.",
"",
"Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers---8x deeper than VGG nets but still having lower complexity. An ensemble of these residual nets achieves 3.57 error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28 relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.",
"State-of-the-art object detection networks depend on region proposal algorithms to hypothesize object locations. Advances like SPPnet [1] and Fast R-CNN [2] have reduced the running time of these detection networks, exposing region proposal computation as a bottleneck. In this work, we introduce a Region Proposal Network (RPN) that shares full-image convolutional features with the detection network, thus enabling nearly cost-free region proposals. An RPN is a fully convolutional network that simultaneously predicts object bounds and objectness scores at each position. The RPN is trained end-to-end to generate high-quality region proposals, which are used by Fast R-CNN for detection. We further merge RPN and Fast R-CNN into a single network by sharing their convolutional features—using the recently popular terminology of neural networks with ’attention’ mechanisms, the RPN component tells the unified network where to look. For the very deep VGG-16 model [3] , our detection system has a frame rate of 5 fps ( including all steps ) on a GPU, while achieving state-of-the-art object detection accuracy on PASCAL VOC 2007, 2012, and MS COCO datasets with only 300 proposals per image. In ILSVRC and COCO 2015 competitions, Faster R-CNN and RPN are the foundations of the 1st-place winning entries in several tracks. Code has been made publicly available.",
"In this work we investigate the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting. Our main contribution is a thorough evaluation of networks of increasing depth using an architecture with very small (3x3) convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers. These findings were the basis of our ImageNet Challenge 2014 submission, where our team secured the first and the second places in the localisation and classification tracks respectively. We also show that our representations generalise well to other datasets, where they achieve state-of-the-art results. We have made our two best-performing ConvNet models publicly available to facilitate further research on the use of deep visual representations in computer vision.",
"We present a method for detecting objects in images using a single deep neural network. Our approach, named SSD, discretizes the output space of bounding boxes into a set of default boxes over different aspect ratios and scales per feature map location. At prediction time, the network generates scores for the presence of each object category in each default box and produces adjustments to the box to better match the object shape. Additionally, the network combines predictions from multiple feature maps with different resolutions to naturally handle objects of various sizes. SSD is simple relative to methods that require object proposals because it completely eliminates proposal generation and subsequent pixel or feature resampling stages and encapsulates all computation in a single network. This makes SSD easy to train and straightforward to integrate into systems that require a detection component. Experimental results on the PASCAL VOC, COCO, and ILSVRC datasets confirm that SSD has competitive accuracy to methods that utilize an additional object proposal step and is much faster, while providing a unified framework for both training and inference. For (300 300 ) input, SSD achieves 74.3 mAP on VOC2007 test at 59 FPS on a Nvidia Titan X and for (512 512 ) input, SSD achieves 76.9 mAP, outperforming a comparable state of the art Faster R-CNN model. Compared to other single stage methods, SSD has much better accuracy even with a smaller input image size. Code is available at https: github.com weiliu89 caffe tree ssd."
]
} |
1708.08417 | 2753711514 | Many applications such as autonomous navigation, urban planning and asset monitoring, rely on the availability of accurate information about objects and their geolocations. In this paper we propose to automatically detect and compute the GPS coordinates of recurring stationary objects of interest using street view imagery. Our processing pipeline relies on two fully convolutional neural networks: the first segments objects in the images while the second estimates their distance from the camera. To geolocate all the detected objects coherently we propose a novel custom Markov Random Field model to perform objects triangulation. The novelty of the resulting pipeline is the combined use of monocular depth estimation and triangulation to enable automatic mapping of complex scenes with multiple visually similar objects of interest. We validate experimentally the effectiveness of our approach on two object classes: traffic lights and telegraph poles. The experiments report high object recall rates and GPS accuracy within 2 meters, which is comparable with the precision of single-frequency GPS receivers. | Estimation of camera-to-object distances and, more generally, 3D scene analysis from RGB images, can be addressed in several ways. The most explored are the stereo-vision approaches @cite_9 that estimate camera-to-object distances from multi-view stereo image disparity analysis or perform scene reconstruction via Structure-from-Motion methods @cite_21 . These rely on various assumptions about camera positions and trajectory, and typically require a rich set of input RGB images and a certain form of knowledge about the analyzed scene. These classes of methods rely heavily on feature extraction and matching. Another way to address scene depth evaluation has been explored in several recent studies @cite_4 @cite_12 @cite_2 . The central idea in these is to recover depth information from monocular images based on the extensive scene interpretation capabilities of FCNNs by training them on RGB and depth or disparity images. These methods achieve consistent yet approximate results relying solely on the information available in the color bands. | {
"cite_N": [
"@cite_4",
"@cite_9",
"@cite_21",
"@cite_2",
"@cite_12"
],
"mid": [
"2520707372",
"2160014001",
"2464339273",
"2124907686",
"2963591054"
],
"abstract": [
"Learning based methods have shown very promising results for the task of depth estimation in single images. However, most existing approaches treat depth prediction as a supervised regression problem and as a result, require vast quantities of corresponding ground truth depth data for training. Just recording quality depth data in a range of environments is a challenging problem. In this paper, we innovate beyond existing approaches, replacing the use of explicit depth data during training with easier-to-obtain binocular stereo footage. We propose a novel training objective that enables our convolutional neural network to learn to perform single image depth estimation, despite the absence of ground truth depth data. Ex-ploiting epipolar geometry constraints, we generate disparity images by training our network with an image reconstruction loss. We show that solving for image reconstruction alone results in poor quality depth images. To overcome this problem, we propose a novel training loss that enforces consistency between the disparities produced relative to both the left and right images, leading to improved performance and robustness compared to existing approaches. Our method produces state of the art results for monocular depth estimation on the KITTI driving dataset, even outperforming supervised methods that have been trained with ground truth depth.",
"This paper presents a quantitative comparison of several multi-view stereo reconstruction algorithms. Until now, the lack of suitable calibrated multi-view image datasets with known ground truth (3D shape models) has prevented such direct comparisons. In this paper, we first survey multi-view stereo algorithms and compare them qualitatively using a taxonomy that differentiates their key properties. We then describe our process for acquiring and calibrating multiview image datasets with high-accuracy ground truth and introduce our evaluation methodology. Finally, we present the results of our quantitative comparison of state-of-the-art multi-view stereo reconstruction algorithms on six benchmark datasets. The datasets, evaluation details, and instructions for submitting new models are available online at http: vision.middlebury.edu mview.",
"We consider the problem of robust rotation optimization in Structure from Motion applications. A number of different approaches have been recently proposed, with solutions that are at times incompatible, and at times complementary. The goal of this paper is to survey and compare these ideas in a unified manner, and to benchmark their robustness against the presence of outliers. In all, we have tested more than forty variants of a these methods (including novel ones), and we find the best performing combination.",
"Predicting the depth (or surface normal) of a scene from single monocular color images is a challenging task. This paper tackles this challenging and essentially underdetermined problem by regression on deep convolutional neural network (DCNN) features, combined with a post-processing refining step using conditional random fields (CRF). Our framework works at two levels, super-pixel level and pixel level. First, we design a DCNN model to learn the mapping from multi-scale image patches to depth or surface normal values at the super-pixel level. Second, the estimated super-pixel depth or surface normal is refined to the pixel level by exploiting various potentials on the depth or surface normal map, which includes a data term, a smoothness term among super-pixels and an auto-regression term characterizing the local structure of the estimation map. The inference problem can be efficiently solved because it admits a closed-form solution. Experiments on the Make3D and NYU Depth V2 datasets show competitive results compared with recent state-of-the-art methods.",
"This paper addresses the problem of estimating the depth map of a scene given a single RGB image. We propose a fully convolutional architecture, encompassing residual learning, to model the ambiguous mapping between monocular images and depth maps. In order to improve the output resolution, we present a novel way to efficiently learn feature map up-sampling within the network. For optimization, we introduce the reverse Huber loss that is particularly suited for the task at hand and driven by the value distributions commonly present in depth maps. Our model is composed of a single architecture that is trained end-to-end and does not rely on post-processing techniques, such as CRFs or other additional refinement steps. As a result, it runs in real-time on images or videos. In the evaluation, we show that the proposed model contains fewer parameters and requires fewer training data than the current state of the art, while outperforming all approaches on depth estimation. Code and models are publicly available."
]
} |
1708.07769 | 2749904558 | We show how standard Newtonian N-body simulations can be interpreted in terms of the weak-field limit of general relativity by employing the recently developed Newtonian motion gauge. Our framework allows the inclusion of radiation perturbations and the non-linear evolution of matter. We show how to construct the weak-field metric by combining Newtonian simulations with results from Einstein-Boltzmann codes. We discuss observational effects on weak lensing and ray tracing, identifying important relativistic corrections. | In this section we discuss the connection between our work, the analytic method of @cite_33 , the relativistic ray tracing code liger and the relativistic N-body code gevolution . | {
"cite_N": [
"@cite_33"
],
"mid": [
"2068569185"
],
"abstract": [
"On large scales, comparable to the horizon, the observable clustering properties of galaxies are affected by various general relativistic effects. To calculate these effects one needs to consistently solve for the metric, densities, and velocities in a specific coordinate system or gauge. The method of choice for simulating large-scale structure is numerical N-body simulations which are performed in the Newtonian limit. Even though one might worry that the use of the Newtonian approximation would make it impossible to use these simulations to compute properties on very large scales, we show that the simulations are still solving the dynamics correctly even for long modes and we give formulas to obtain the position of particles in the conformal Newtonian gauge given the positions computed in the simulation. We also give formulas to convert from the output coordinates of N-body simulations to the observable coordinates of the particles."
]
} |
1708.07769 | 2749904558 | We show how standard Newtonian N-body simulations can be interpreted in terms of the weak-field limit of general relativity by employing the recently developed Newtonian motion gauge. Our framework allows the inclusion of radiation perturbations and the non-linear evolution of matter. We show how to construct the weak-field metric by combining Newtonian simulations with results from Einstein-Boltzmann codes. We discuss observational effects on weak lensing and ray tracing, identifying important relativistic corrections. | In the previous work by Chisari and Zaldarriaga (CZ) @cite_33 , a dictionary between simulations and general relativistic perturbations in the Poisson gauge has been developed, using a different ansatz restricted to linear perturbation theory in a pressureless Universe. As expected, their findings correspond to our more general result in the appropriate limits. | {
"cite_N": [
"@cite_33"
],
"mid": [
"2068569185"
],
"abstract": [
"On large scales, comparable to the horizon, the observable clustering properties of galaxies are affected by various general relativistic effects. To calculate these effects one needs to consistently solve for the metric, densities, and velocities in a specific coordinate system or gauge. The method of choice for simulating large-scale structure is numerical N-body simulations which are performed in the Newtonian limit. Even though one might worry that the use of the Newtonian approximation would make it impossible to use these simulations to compute properties on very large scales, we show that the simulations are still solving the dynamics correctly even for long modes and we give formulas to obtain the position of particles in the conformal Newtonian gauge given the positions computed in the simulation. We also give formulas to convert from the output coordinates of N-body simulations to the observable coordinates of the particles."
]
} |
1708.07769 | 2749904558 | We show how standard Newtonian N-body simulations can be interpreted in terms of the weak-field limit of general relativity by employing the recently developed Newtonian motion gauge. Our framework allows the inclusion of radiation perturbations and the non-linear evolution of matter. We show how to construct the weak-field metric by combining Newtonian simulations with results from Einstein-Boltzmann codes. We discuss observational effects on weak lensing and ray tracing, identifying important relativistic corrections. | In the equations of CZ @cite_33 , the comoving curvature perturbation @math is further replaced by the potential @math using the relation 3 = 5 ,. This relation is only valid in the absence of radiation and or dark energy. At the times where N-body simulations are usually initialised, this is of course fairly accurate, but the same relation may no longer be applied at the final time. | {
"cite_N": [
"@cite_33"
],
"mid": [
"2068569185"
],
"abstract": [
"On large scales, comparable to the horizon, the observable clustering properties of galaxies are affected by various general relativistic effects. To calculate these effects one needs to consistently solve for the metric, densities, and velocities in a specific coordinate system or gauge. The method of choice for simulating large-scale structure is numerical N-body simulations which are performed in the Newtonian limit. Even though one might worry that the use of the Newtonian approximation would make it impossible to use these simulations to compute properties on very large scales, we show that the simulations are still solving the dynamics correctly even for long modes and we give formulas to obtain the position of particles in the conformal Newtonian gauge given the positions computed in the simulation. We also give formulas to convert from the output coordinates of N-body simulations to the observable coordinates of the particles."
]
} |
1708.07932 | 1985914750 | There are so many vehicles in the world and the number of vehicles is increasing rapidly. To alleviate the parking problems caused by that, the smart parking system has been developed. The parking planning is one of the most important parts of it. An effective parking planning strategy makes the better use of parking resources possible. In this paper, we present a feasible method to do parking planning. We transform the parking planning problem into a kind of linear assignment problem. We take vehicles as jobs and parking spaces as agents. We take distances between vehicles and parking spaces as costs for agents doing jobs. Then we design an algorithm for this particular assignment problem and solve the parking planning problem. The method proposed can give timely and efficient guide information to vehicles for a real time smart parking system. Finally, we show the effectiveness of the method with experiments over some data, which can simulate the situation of doing parking planning in the real world. | T. Litman @cite_4 presents a comprehensive implementation guide on parking management, including parking planning practices. The study of M. Idris @cite_5 reviews the evolution of parking space occupancy detection. The detection technology makes the status of the parking space available for the parking management system. S. Zeitman @cite_3 invents a communication system for the parking management system. They provide a method to establish communication between municipality, driver and parking spaces. S. Shaheen @cite_2 documents the research and feasibility analysis for the design and implementation of parking management field test. Their report gives a description of the parking field test and its technology in detail. R. Lu @cite_0 provides a solution for smart parking scheme through taking advantage of communication between vehicules. Our method is different since we study the problem of processing parking planning on a real time smart parking system, then sending the guide information to vehicles. | {
"cite_N": [
"@cite_4",
"@cite_3",
"@cite_0",
"@cite_2",
"@cite_5"
],
"mid": [
"",
"1509031731",
"2171278763",
"2188556114",
"2060854363"
],
"abstract": [
"",
"A parking management communication system including a central control unit (12) having a database (14), a central interface unit (16) and at least one user interface (20, 22, and 26), the central interface unit (16) being in communication with the at least one user interface unit (20, 22, and 26) via at least one of wired and wireless communication link.",
"Searching for a vacant parking space in a congested area or a large parking lot and preventing auto theft are major concerns to our daily lives. In this paper, we propose a new smart parking scheme for large parking lots through vehicular communication. The proposed scheme can provide the drivers with real-time parking navigation service, intelligent anti- theft protection, and friendly parking information dissemination. Performance analysis via extensive simulations demonstrates its efficiency and practicality. Keywords— Vehicular communications; smart parking; navi- gation; anti-theft; information dissemination; security & privacy",
"In almost every major city in the U.S. and internationally, parking problems are ubiquitous. It is well known that the limited availability of parking contributes to roadway congestion, air pollution, and driver frustration and that the cost of expanding traditional parking capacity is frequently prohibitive. However, less research has addressed the effect of insufficient parking at transit stations on transit use. In the San Francisco Bay Area, parking has recently been at or near capacity at many of the 31 Bay Area Rapid Transit (BART) District stations with parking facilities. Smart parking management technologies may provide a cost-effective tool to address near-term parking constraints at BART transit stations. This report presents early findings from an application of advanced parking technologies to maximize existing parking capacity at the Rockridge BART station, which was launched in December 2004 in the East San Francisco Bay Area. The smart parking system includes traffic sensors that count the number of vehicles entering and exiting the parking lots at the station. A reservation system allows travelers to reserve spaces by Internet, personal digital assistant (PDA), phone, and cell phone. The real-time information obtained from the sensors and the reservation system is displayed on variable message signs (VMS) (on Highway 24 leading to the station) to alert drivers of parking space availability. Before and after surveys and focus groups will be used to evaluate the travel effects, economic potential, and system technology of the field test. This report consists of three major sections: * A literature review in which the effectiveness of different types and applications of smart parking management systems are evaluated; * A feasibility analysis, including focus groups, surveys, and observational analyses, which guides the development and initial evaluation of the smart parking field test; and * A smart parking project description, which includes the applied demonstration design and technology.",
"Service technology such as Web services has been one of the mainstream technologies in today’s software development. Distributed services may suffer from communication problems or contain faults themselves, and hence service consumers may experience service interruption. A solution is to create services which can tolerate faults so that failures can be made transparent to the consumers. Since there are many patterns of software fault tolerance available, we end up with a question of which pattern should be applied to a particular service. This paper attempts to recommend to service developers the patterns for fault tolerant services. A recommendation model is proposed based on characteristics of the service itself and of the service provision environment. Once a fault tolerance pattern is chosen, a fault tolerant version of the service can be created as a WS-BPEL service. A software tool is developed to assist in pattern recommendation and generation of the fault tolerant service version."
]
} |
1708.07819 | 2748021867 | This work addresses the problem of semantic foggy scene understanding (SFSU). Although extensive research has been performed on image dehazing and on semantic scene understanding with clear-weather images, little attention has been paid to SFSU. Due to the difficulty of collecting and annotating foggy images, we choose to generate synthetic fog on real images that depict clear-weather outdoor scenes, and then leverage these partially synthetic data for SFSU by employing state-of-the-art convolutional neural networks (CNN). In particular, a complete pipeline to add synthetic fog to real, clear-weather images using incomplete depth information is developed. We apply our fog synthesis on the Cityscapes dataset and generate Foggy Cityscapes with 20,550 images. SFSU is tackled in two ways: (1) with typical supervised learning, and (2) with a novel type of semi-supervised learning, which combines (1) with an unsupervised supervision transfer from clear-weather images to their synthetic foggy counterparts. In addition, we carefully study the usefulness of image dehazing for SFSU. For evaluation, we present Foggy Driving, a dataset with 101 real-world images depicting foggy driving scenes, which come with ground truth annotations for semantic segmentation and object detection. Extensive experiments show that (1) supervised learning with our synthetic data significantly improves the performance of state-of-the-art CNN for SFSU on Foggy Driving; (2) our semi-supervised learning strategy further improves performance; and (3) image dehazing marginally advances SFSU with our learning strategy. The datasets, models and code are made publicly available. | Our work bears resemblance to works from the broad field of transfer learning. Model adaptation across weather conditions to semantically segment simple road scenes is studied in @cite_42 . More recently, a domain adversarial based approach was proposed to adapt semantic segmentation models both at pixel level and feature level from simulated to real environments @cite_27 . Our work generates synthetic fog from clear-weather data to close the domain gap. Combining our method and the aforementioned transfer learning methods is a promising direction for future work. The supervision transfer from clear weather to foggy weather in this paper is inspired by the stream of work on model distillation imitation @cite_7 @cite_48 @cite_18 . Our approach is similar in that knowledge is transferred from one domain (model) to another by using paired data samples as a bridge. | {
"cite_N": [
"@cite_18",
"@cite_7",
"@cite_48",
"@cite_42",
"@cite_27"
],
"mid": [
"1920864768",
"1821462560",
"753847829",
"2018726624",
"2767657961"
],
"abstract": [
"Metric learning has proved very successful. However, human annotations are necessary. In this paper, we propose an unsupervised method, dubbed Metric Imitation (MI), where metrics over cheap features (target features, TFs) are learned by imitating the standard metrics over more sophisticated, off-the-shelf features (source features, SFs) by transferring view-independent property manifold structures. In particular, MI consists of: 1) quantifying the properties of source metrics as manifold geometry, 2) transferring the manifold from source domain to target domain, and 3) learning a mapping of TFs so that the manifold is approximated as well as possible in the mapped feature domain. MI is useful in at least two scenarios where: 1) TFs are more efficient computationally and in terms of memory than SFs; and 2) SFs contain privileged information, but are not available during testing. For the former, MI is evaluated on image clustering, category-based image retrieval, and instance-based object retrieval, with three SFs and three TFs. For the latter, MI is tested on the task of example-based image super-resolution, where high-resolution patches are taken as SFs and low-resolution patches as TFs. Experiments show that MI is able to provide good metrics while avoiding expensive data labeling efforts and that it achieves state-of-the-art performance for image super-resolution. In addition, manifold transfer is an interesting direction of transfer learning.",
"A very simple way to improve the performance of almost any machine learning algorithm is to train many different models on the same data and then to average their predictions. Unfortunately, making predictions using a whole ensemble of models is cumbersome and may be too computationally expensive to allow deployment to a large number of users, especially if the individual models are large neural nets. Caruana and his collaborators have shown that it is possible to compress the knowledge in an ensemble into a single model which is much easier to deploy and we develop this approach further using a different compression technique. We achieve some surprising results on MNIST and we show that we can significantly improve the acoustic model of a heavily used commercial system by distilling the knowledge in an ensemble of models into a single model. We also introduce a new type of ensemble composed of one or more full models and many specialist models which learn to distinguish fine-grained classes that the full models confuse. Unlike a mixture of experts, these specialist models can be trained rapidly and in parallel.",
"In this work we propose a technique that transfers supervision between images from different modalities. We use learned representations from a large labeled modality as supervisory signal for training representations for a new unlabeled paired modality. Our method enables learning of rich representations for unlabeled modalities and can be used as a pre-training procedure for new modalities with limited labeled data. We transfer supervision from labeled RGB images to unlabeled depth and optical flow images and demonstrate large improvements for both these cross modal supervision transfers.",
"Semantic road labeling is a key component of systems that aim at assisted or even autonomous driving. Considering that such systems continuously operate in the real-world, unforeseen conditions not represented in any conceivable training procedure are likely to occur on a regular basis. In order to equip systems with the ability to cope with such situations, we would like to enable adaptation to such new situations and conditions at runtime. Existing adaptive methods for image labeling either require labeled data from the new condition or even operate globally on a complete test set. None of this is a desirable mode of operation for a system as described above where new images arrive sequentially and conditions may vary. We study the effect of changing test conditions on scene labeling methods based on a new diverse street scene dataset. We propose a novel approach that can operate in such conditions and is based on a sequential Bayesian model update in order to robustly integrate the arriving images into the adapting procedure.",
"Domain adaptation is critical for success in new, unseen environments. Adversarial adaptation models applied in feature spaces discover domain invariant representations, but are difficult to visualize and sometimes fail to capture pixel-level and low-level domain shifts. Recent work has shown that generative adversarial networks combined with cycle-consistency constraints are surprisingly effective at mapping images between domains, even without the use of aligned image pairs. We propose a novel discriminatively-trained Cycle-Consistent Adversarial Domain Adaptation model. CyCADA adapts representations at both the pixel-level and feature-level, enforces cycle-consistency while leveraging a task loss, and does not require aligned pairs. Our model can be applied in a variety of visual recognition and prediction settings. We show new state-of-the-art results across multiple adaptation tasks, including digit classification and semantic segmentation of road scenes demonstrating transfer from synthetic to real world domains."
]
} |
1708.07860 | 2750912449 | We investigate methods for combining multiple self-supervised tasks--i.e., supervised tasks where data can be collected without manual labeling--in order to train a single visual representation. First, we provide an apples-to-apples comparison of four different self-supervised tasks using the very deep ResNet-101 architecture. We then combine tasks to jointly train a network. We also explore lasso regularization to encourage the network to factorize the information in its representation, and methods for "harmonizing" network inputs in order to learn a more unified representation. We evaluate all methods on ImageNet classification, PASCAL VOC detection, and NYU depth prediction. Our results show that deeper networks work better, and that combining tasks--even via a naive multi-head architecture--always improves performance. Our best joint network nearly matches the PASCAL performance of a model pre-trained on ImageNet classification, and matches the ImageNet network on NYU depth prediction. | Tasks that use auxiliary information such as multi-modal information beyond pixels include: predicting sound given videos @cite_42 , predicting camera motion given two images of the same scene @cite_0 @cite_16 @cite_3 , or predicting what robotic motion caused a change in a scene @cite_29 @cite_10 @cite_28 @cite_44 @cite_19 . However, non-visual information can be difficult to obtain: estimating motion requires IMU measurements, running robots is still expensive, and sound is complex and difficult to evaluate quantitatively. | {
"cite_N": [
"@cite_28",
"@cite_29",
"@cite_42",
"@cite_3",
"@cite_0",
"@cite_44",
"@cite_19",
"@cite_16",
"@cite_10"
],
"mid": [
"2528973876",
"2338684808",
"2951353470",
"",
"2198618282",
"2473208550",
"2964335674",
"2951590555",
"2949098821"
],
"abstract": [
"There has been a recent paradigm shift in robotics to data-driven learning for planning and control. Due to large number of experiences required for training, most of these approaches use a self-supervised paradigm: using sensors to measure success failure. However, in most cases, these sensors provide weak supervision at best. In this work, we propose an adversarial learning framework that pits an adversary against the robot learning the task. In an effort to defeat the adversary, the original robot learns to perform the task with more robustness leading to overall improved performance. We show that this adversarial framework forces the the robot to learn a better grasping model in order to overcome the adversary. By grasping 82 of presented novel objects compared to 68 without an adversary, we demonstrate the utility of creating adversaries. We also demonstrate via experiments that having robots in adversarial setting might be a better learning strategy as compared to having collaborative multiple robots.",
"What is the right supervisory signal to train visual representations? Current approaches in computer vision use category labels from datasets such as ImageNet to train ConvNets. However, in case of biological agents, visual representation learning does not require millions of semantic labels. We argue that biological agents use physical interactions with the world to learn visual representations unlike current vision systems which just use passive observations (images and videos downloaded from web). For example, babies push objects, poke them, put them in their mouth and throw them to learn representations. Towards this goal, we build one of the first systems on a Baxter platform that pushes, pokes, grasps and observes objects in a tabletop environment. It uses four different types of physical interactions to collect more than 130K datapoints, with each datapoint providing supervision to a shared ConvNet architecture allowing us to learn visual representations. We show the quality of learned representations by observing neuron activations and performing nearest neighbor retrieval on this learned representation. Quantitatively, we evaluate our learned ConvNet on image classification tasks and show improvements compared to learning without external data. Finally, on the task of instance retrieval, our network outperforms the ImageNet network on recall@1 by 3",
"The sound of crashing waves, the roar of fast-moving cars -- sound conveys important information about the objects in our surroundings. In this work, we show that ambient sounds can be used as a supervisory signal for learning visual models. To demonstrate this, we train a convolutional neural network to predict a statistical summary of the sound associated with a video frame. We show that, through this process, the network learns a representation that conveys information about objects and scenes. We evaluate this representation on several recognition tasks, finding that its performance is comparable to that of other state-of-the-art unsupervised learning methods. Finally, we show through visualizations that the network learns units that are selective to objects that are often associated with characteristic sounds.",
"",
"Understanding how images of objects and scenes behave in response to specific ego-motions is a crucial aspect of proper visual development, yet existing visual learning methods are conspicuously disconnected from the physical source of their images. We propose to exploit proprioceptive motor signals to provide unsupervised regularization in convolutional neural networks to learn visual representations from egocentric video. Specifically, we enforce that our learned features exhibit equivariance, i.e, they respond predictably to transformations associated with distinct ego-motions. With three datasets, we show that our unsupervised feature learning approach significantly outperforms previous approaches on visual recognition and next-best-view prediction tasks. In the most challenging test, we show that features learned from video captured on an autonomous driving platform improve large-scale scene recognition in static images from a disjoint domain.",
"We investigate an experiential learning paradigm for acquiring an internal model of intuitive physics. Our model is evaluated on a real-world robotic manipulation task that requires displacing objects to target locations by poking. The robot gathered over 400 hours of experience by executing more than 100K pokes on different objects. We propose a novel approach based on deep neural networks for modeling the dynamics of robot's interactions directly from images, by jointly estimating forward and inverse models of dynamics. The inverse model objective provides supervision to construct informative visual features, which the forward model can then predict and in turn regularize the feature space for the inverse model. The interplay between these two objectives creates useful, accurate models that can then be used for multi-step decision making. This formulation has the additional benefit that it is possible to learn forward models in an abstract feature space and thus alleviate the need of predicting pixels. Our experiments show that this joint modeling approach outperforms alternative methods.",
"Recently, end-to-end learning frameworks are gaining prevalence in the field of robot control. These frameworks input states images and directly predict the torques or the action parameters. However, these approaches are often critiqued due to their huge data requirements for learning a task. The argument of the difficulty in scalability to multiple tasks is well founded, since training these tasks often require hundreds or thousands of examples. But do end-to-end approaches need to learn a unique model for every task? Intuitively, it seems that sharing across tasks should help since all tasks require some common understanding of the environment. In this paper, we attempt to take the next step in data-driven end-to-end learning frameworks: move from the realm of task-specific models to joint learning of multiple robot tasks. In an astonishing result we show that models with multi-task learning tend to perform better than task-specific models trained with same amounts of data. For example, a deep-network learned with 2.5K grasp and 2.5K push examples performs better on grasping than a network trained on 5K grasp examples.",
"The dominant paradigm for feature learning in computer vision relies on training neural networks for the task of object recognition using millions of hand labelled images. Is it possible to learn useful features for a diverse set of visual tasks using any other form of supervision? In biology, living organisms developed the ability of visual perception for the purpose of moving and acting in the world. Drawing inspiration from this observation, in this work we investigate if the awareness of egomotion can be used as a supervisory signal for feature learning. As opposed to the knowledge of class labels, information about egomotion is freely available to mobile agents. We show that given the same number of training images, features learnt using egomotion as supervision compare favourably to features learnt using class-label as supervision on visual tasks of scene recognition, object recognition, visual odometry and keypoint matching.",
"Current learning-based robot grasping approaches exploit human-labeled datasets for training the models. However, there are two problems with such a methodology: (a) since each object can be grasped in multiple ways, manually labeling grasp locations is not a trivial task; (b) human labeling is biased by semantics. While there have been attempts to train robots using trial-and-error experiments, the amount of data used in such experiments remains substantially low and hence makes the learner prone to over-fitting. In this paper, we take the leap of increasing the available training data to 40 times more than prior work, leading to a dataset size of 50K data points collected over 700 hours of robot grasping attempts. This allows us to train a Convolutional Neural Network (CNN) for the task of predicting grasp locations without severe overfitting. In our formulation, we recast the regression problem to an 18-way binary classification over image patches. We also present a multi-stage learning approach where a CNN trained in one stage is used to collect hard negatives in subsequent stages. Our experiments clearly show the benefit of using large-scale datasets (and multi-stage training) for the task of grasping. We also compare to several baselines and show state-of-the-art performance on generalization to unseen objects for grasping."
]
} |
1708.07860 | 2750912449 | We investigate methods for combining multiple self-supervised tasks--i.e., supervised tasks where data can be collected without manual labeling--in order to train a single visual representation. First, we provide an apples-to-apples comparison of four different self-supervised tasks using the very deep ResNet-101 architecture. We then combine tasks to jointly train a network. We also explore lasso regularization to encourage the network to factorize the information in its representation, and methods for "harmonizing" network inputs in order to learn a more unified representation. We evaluate all methods on ImageNet classification, PASCAL VOC detection, and NYU depth prediction. Our results show that deeper networks work better, and that combining tasks--even via a naive multi-head architecture--always improves performance. Our best joint network nearly matches the PASCAL performance of a model pre-trained on ImageNet classification, and matches the ImageNet network on NYU depth prediction. | Thus, many works use raw pixels. In videos, time can be a source of supervision. One can simply predict future @cite_8 @cite_34 , although such predictions may be difficult to evaluate. One way to simplify the problem is to ask a network to temporally order a set of frames sampled from a video @cite_11 . Another is to note that objects generally appear across many frames: thus, we can train features to remain invariant as a video progresses @cite_20 @cite_35 @cite_31 @cite_25 @cite_30 . Finally, motion cues can separate foreground objects from background. Neural networks can be asked to re-produce these motion-based boundaries without seeing motion @cite_17 @cite_23 . | {
"cite_N": [
"@cite_30",
"@cite_35",
"@cite_8",
"@cite_17",
"@cite_23",
"@cite_31",
"@cite_34",
"@cite_25",
"@cite_20",
"@cite_11"
],
"mid": [
"",
"2146444479",
"2952390294",
"2950341389",
"2204233249",
"2145038566",
"",
"219040644",
"2096388912",
""
],
"abstract": [
"",
"Invariant features of temporally varying signals are useful for analysis and classification. Slow feature analysis (SFA) is a new method for learning invariant or slowly varying features from a vectorial input signal. It is based on a nonlinear expansion of the input signal and application of principal component analysis to this expanded signal and its time derivative. It is guaranteed to find the optimal solution within a family of functions directly and can learn to extract a large number of decorrelated features, which are ordered by their degree of invariance. SFA can be applied hierarchically to process high-dimensional input signals and extract complex features. SFA is applied first to complex cell tuning properties based on simple cell output, including disparity and motion. Then more complicated input-output functions are learned by repeated application of SFA. Finally, a hierarchical network of SFA modules is presented as a simple model of the visual system. The same unstructured network can learn translation, size, rotation, contrast, or, to a lesser degree, illumination invariance for one-dimensional objects, depending on only the training stimulus. Surprisingly, only a few training objects suffice to achieve good generalization to new objects. The generated representation is suitable for object recognition. Performance degrades if the network is trained to learn multiple invariances simultaneously.",
"In a given scene, humans can often easily predict a set of immediate future events that might happen. However, generalized pixel-level anticipation in computer vision systems is difficult because machine learning struggles with the ambiguity inherent in predicting the future. In this paper, we focus on predicting the dense trajectory of pixels in a scene, specifically what will move in the scene, where it will travel, and how it will deform over the course of one second. We propose a conditional variational autoencoder as a solution to this problem. In this framework, direct inference from the image shapes the distribution of possible trajectories, while latent variables encode any necessary information that is not available in the image. We show that our method is able to successfully predict events in a wide variety of scenes and can produce multiple different predictions when the future is ambiguous. Our algorithm is trained on thousands of diverse, realistic videos and requires absolutely no human labeling. In addition to non-semantic action prediction, we find that our method learns a representation that is applicable to semantic vision tasks.",
"This paper presents a novel yet intuitive approach to unsupervised feature learning. Inspired by the human visual system, we explore whether low-level motion-based grouping cues can be used to learn an effective visual representation. Specifically, we use unsupervised motion-based segmentation on videos to obtain segments, which we use as 'pseudo ground truth' to train a convolutional network to segment objects from a single frame. Given the extensive evidence that motion plays a key role in the development of the human visual system, we hope that this straightforward approach to unsupervised learning will be more effective than cleverly designed 'pretext' tasks studied in the literature. Indeed, our extensive experiments show that this is the case. When used for transfer learning on object detection, our representation significantly outperforms previous unsupervised approaches across multiple settings, especially when training data for the target task is scarce.",
"Data-driven approaches for edge detection have proven effective and achieve top results on modern benchmarks. However, all current data-driven edge detectors require manual supervision for training in the form of hand-labeled region segments or object boundaries. Specifically, human annotators mark semantically meaningful edges which are subsequently used for training. Is this form of strong, high-level supervision actually necessary to learn to accurately detect edges? In this work we present a simple yet effective approach for training edge detectors without human supervision. To this end we utilize motion, and more specifically, the only input to our method is noisy semi-dense matches between frames. We begin with only a rudimentary knowledge of edges (in the form of image gradients), and alternate between improving motion estimation and edge detection in turn. Using a large corpus of video data, we show that edge detectors trained using our unsupervised scheme approach the performance of the same methods trained with full supervision (within 3-5 ). Finally, we show that when using a deep network for the edge detector, our approach provides a novel pre-training scheme for object detection.",
"This work proposes a learning method for deep architectures that takes advantage of sequential data, in particular from the temporal coherence that naturally exists in unlabeled video recordings. That is, two successive frames are likely to contain the same object or objects. This coherence is used as a supervisory signal over the unlabeled data, and is used to improve the performance on a supervised task of interest. We demonstrate the effectiveness of this method on some pose invariant object and face recognition tasks.",
"",
"Is strong supervision necessary for learning a good visual representation? Do we really need millions of semantically-labeled images to train a Convolutional Neural Network (CNN)? In this paper, we present a simple yet surprisingly powerful approach for unsupervised learning of CNN. Specifically, we use hundreds of thousands of unlabeled videos from the web to learn visual representations. Our key idea is that visual tracking provides the supervision. That is, two patches connected by a track should have similar visual representation in deep feature space since they probably belong to same object or object part. We design a Siamese-triplet network with a ranking loss function to train this CNN representation. Without using a single image from ImageNet, just using 100K unlabeled videos and the VOC 2012 dataset, we train an ensemble of unsupervised networks that achieves 52 mAP (no bounding box regression). This performance comes tantalizingly close to its ImageNet-supervised counterpart, an ensemble which achieves a mAP of 54.4 . We also show that our unsupervised network can perform competitively in other tasks such as surface-normal estimation.",
"The visual system can reliably identify objects even when the retinal image is transformed considerably by commonly occurring changes in the environment. A local learning rule is proposed, which allows a network to learn to generalize across such transformations. During the learning phase, the network is exposed to temporal sequences of patterns undergoing the transformation. An application of the algorithm is presented in which the network learns invariance to shift in retinal position. Such a principle may be involved in the development of the characteristic shift invariance property of complex cells in the primary visual cortex, and also in the development of more complicated invariance properties of neurons in higher visual areas.",
""
]
} |
1708.07860 | 2750912449 | We investigate methods for combining multiple self-supervised tasks--i.e., supervised tasks where data can be collected without manual labeling--in order to train a single visual representation. First, we provide an apples-to-apples comparison of four different self-supervised tasks using the very deep ResNet-101 architecture. We then combine tasks to jointly train a network. We also explore lasso regularization to encourage the network to factorize the information in its representation, and methods for "harmonizing" network inputs in order to learn a more unified representation. We evaluate all methods on ImageNet classification, PASCAL VOC detection, and NYU depth prediction. Our results show that deeper networks work better, and that combining tasks--even via a naive multi-head architecture--always improves performance. Our best joint network nearly matches the PASCAL performance of a model pre-trained on ImageNet classification, and matches the ImageNet network on NYU depth prediction. | Self-supervised learning can also work with a single image. One can hide a part of the image and ask the network to make predictions about the hidden part. The network can be tasked with generating pixels, either by filling in holes @cite_38 @cite_37 , or recovering color after images have been converted to grayscale @cite_18 @cite_1 . Again, evaluating the quality of generated pixels is difficult. To simplify the task, one can extract multiple patches at random from an image, and then ask the network to position the patches relative to each other @cite_22 @cite_13 . Finally, one can form a surrogate class'' by taking a single image and altering it many times via translations, rotations, and color shifts @cite_40 , to create a synthetic categorization problem. | {
"cite_N": [
"@cite_38",
"@cite_37",
"@cite_18",
"@cite_22",
"@cite_1",
"@cite_40",
"@cite_13"
],
"mid": [
"2342877626",
"2556707115",
"",
"2950187998",
"2950064337",
"2148349024",
"2321533354"
],
"abstract": [
"We present an unsupervised visual feature learning algorithm driven by context-based pixel prediction. By analogy with auto-encoders, we propose Context Encoders -- a convolutional neural network trained to generate the contents of an arbitrary image region conditioned on its surroundings. In order to succeed at this task, context encoders need to both understand the content of the entire image, as well as produce a plausible hypothesis for the missing part(s). When training context encoders, we have experimented with both a standard pixel-wise reconstruction loss, as well as a reconstruction plus an adversarial loss. The latter produces much sharper results because it can better handle multiple modes in the output. We found that a context encoder learns a representation that captures not just appearance but also the semantics of visual structures. We quantitatively demonstrate the effectiveness of our learned features for CNN pre-training on classification, detection, and segmentation tasks. Furthermore, context encoders can be used for semantic inpainting tasks, either stand-alone or as initialization for non-parametric methods.",
"We introduce a simple semi-supervised learning approach for images based on in-painting using an adversarial loss. Images with random patches removed are presented to a generator whose task is to fill in the hole, based on the surrounding pixels. The in-painted images are then presented to a discriminator network that judges if they are real (unaltered training images) or not. This task acts as a regularizer for standard supervised training of the discriminator. Using our approach we are able to directly train large VGG-style networks in a semi-supervised fashion. We evaluate on STL-10 and PASCAL datasets, where our approach obtains performance comparable or superior to existing methods.",
"",
"This work explores the use of spatial context as a source of free and plentiful supervisory signal for training a rich visual representation. Given only a large, unlabeled image collection, we extract random pairs of patches from each image and train a convolutional neural net to predict the position of the second patch relative to the first. We argue that doing well on this task requires the model to learn to recognize objects and their parts. We demonstrate that the feature representation learned using this within-image context indeed captures visual similarity across images. For example, this representation allows us to perform unsupervised visual discovery of objects like cats, people, and even birds from the Pascal VOC 2011 detection dataset. Furthermore, we show that the learned ConvNet can be used in the R-CNN framework and provides a significant boost over a randomly-initialized ConvNet, resulting in state-of-the-art performance among algorithms which use only Pascal-provided training set annotations.",
"We develop a fully automatic image colorization system. Our approach leverages recent advances in deep networks, exploiting both low-level and semantic representations. As many scene elements naturally appear according to multimodal color distributions, we train our model to predict per-pixel color histograms. This intermediate output can be used to automatically generate a color image, or further manipulated prior to image formation. On both fully and partially automatic colorization tasks, we outperform existing methods. We also explore colorization as a vehicle for self-supervised visual representation learning.",
"Current methods for training convolutional neural networks depend on large amounts of labeled samples for supervised training. In this paper we present an approach for training a convolutional neural network using only unlabeled data. We train the network to discriminate between a set of surrogate classes. Each surrogate class is formed by applying a variety of transformations to a randomly sampled 'seed' image patch. We find that this simple feature learning algorithm is surprisingly successful when applied to visual object recognition. The feature representation learned by our algorithm achieves classification results matching or outperforming the current state-of-the-art for unsupervised learning on several popular datasets (STL-10, CIFAR-10, Caltech-101).",
"We propose a novel unsupervised learning approach to build features suitable for object detection and classification. The features are pre-trained on a large dataset without human annotation and later transferred via fine-tuning on a different, smaller and labeled dataset. The pre-training consists of solving jigsaw puzzles of natural images. To facilitate the transfer of features to other tasks, we introduce the context-free network (CFN), a siamese-ennead convolutional neural network. The features correspond to the columns of the CFN and they process image tiles independently (i.e., free of context). The later layers of the CFN then use the features to identify their geometric arrangement. Our experimental evaluations show that the learned features capture semantically relevant content. We pre-train the CFN on the training set of the ILSVRC2012 dataset and transfer the features on the combined training and validation set of Pascal VOC 2007 for object detection (via fast RCNN) and classification. These features outperform all current unsupervised features with (51.8 , ) for detection and (68.6 , ) for classification, and reduce the gap with supervised learning ( (56.5 , ) and (78.2 , ) respectively)."
]
} |
1708.07860 | 2750912449 | We investigate methods for combining multiple self-supervised tasks--i.e., supervised tasks where data can be collected without manual labeling--in order to train a single visual representation. First, we provide an apples-to-apples comparison of four different self-supervised tasks using the very deep ResNet-101 architecture. We then combine tasks to jointly train a network. We also explore lasso regularization to encourage the network to factorize the information in its representation, and methods for "harmonizing" network inputs in order to learn a more unified representation. We evaluate all methods on ImageNet classification, PASCAL VOC detection, and NYU depth prediction. Our results show that deeper networks work better, and that combining tasks--even via a naive multi-head architecture--always improves performance. Our best joint network nearly matches the PASCAL performance of a model pre-trained on ImageNet classification, and matches the ImageNet network on NYU depth prediction. | Our work is also related to multi-task learning. Several recent works have trained deep visual representations using multiple tasks @cite_24 @cite_46 @cite_33 @cite_41 , including one work @cite_4 which combines no less than 7 tasks. Usually the goal is to create a single representation that works well for every task, and perhaps share knowledge between tasks. Surprisingly, however, previous work has shown little transfer between diverse tasks. Kokkinos @cite_4 , for example, found a slight dip in performance with 7 tasks versus 2. Note that our work is not primarily concerned with the performance on the self-supervised tasks we combine: we evaluate on a separate set of semantic evaluation tasks.'' Some previous self-supervised learning literature has suggested performance gains from combining self-supervised tasks @cite_19 @cite_3 , although these works used relatively similar tasks within relatively restricted domains where extra information was provided besides pixels. In this work, we find that pre-training on multiple diverse self-supervised tasks using only pixels yields strong performance. | {
"cite_N": [
"@cite_4",
"@cite_33",
"@cite_41",
"@cite_3",
"@cite_24",
"@cite_19",
"@cite_46"
],
"mid": [
"2963498646",
"2950209802",
"",
"",
"1487583988",
"2964335674",
"2951713345"
],
"abstract": [
"In this work we train in an end-to-end manner a convolutional neural network (CNN) that jointly handles low-, mid-, and high-level vision tasks in a unified architecture. Such a network can act like a swiss knife for vision tasks, we call it an UberNet to indicate its overarching nature. The main contribution of this work consists in handling challenges that emerge when scaling up to many tasks. We introduce techniques that facilitate (i) training a deep architecture while relying on diverse training sets and (ii) training many (potentially unlimited) tasks with a limited memory budget. This allows us to train in an end-to-end manner a unified CNN architecture that jointly handles (a) boundary detection (b) normal estimation (c) saliency estimation (d) semantic segmentation (e) human part segmentation (f) semantic boundary detection, (g) region proposal generation and object detection. We obtain competitive performance while jointly addressing all tasks in 0.7 seconds on a GPU. Our system will be made publicly available.",
"There are multiple cues in an image which reveal what action a person is performing. For example, a jogger has a pose that is characteristic for jogging, but the scene (e.g. road, trail) and the presence of other joggers can be an additional source of information. In this work, we exploit the simple observation that actions are accompanied by contextual cues to build a strong action recognition system. We adapt RCNN to use more than one region for classification while still maintaining the ability to localize the action. We call our system R*CNN. The action-specific models and the feature maps are trained jointly, allowing for action specific representations to emerge. R*CNN achieves 90.2 mean AP on the PASAL VOC Action dataset, outperforming all other approaches in the field by a significant margin. Last, we show that R*CNN is not limited to action recognition. In particular, R*CNN can also be used to tackle fine-grained tasks such as attribute classification. We validate this claim by reporting state-of-the-art performance on the Berkeley Attributes of People dataset.",
"",
"",
"We present an integrated framework for using Convolutional Networks for classification, localization and detection. We show how a multiscale and sliding window approach can be efficiently implemented within a ConvNet. We also introduce a novel deep learning approach to localization by learning to predict object boundaries. Bounding boxes are then accumulated rather than suppressed in order to increase detection confidence. We show that different tasks can be learned simultaneously using a single shared network. This integrated framework is the winner of the localization task of the ImageNet Large Scale Visual Recognition Challenge 2013 (ILSVRC2013) and obtained very competitive results for the detection and classifications tasks. In post-competition work, we establish a new state of the art for the detection task. Finally, we release a feature extractor from our best model called OverFeat.",
"Recently, end-to-end learning frameworks are gaining prevalence in the field of robot control. These frameworks input states images and directly predict the torques or the action parameters. However, these approaches are often critiqued due to their huge data requirements for learning a task. The argument of the difficulty in scalability to multiple tasks is well founded, since training these tasks often require hundreds or thousands of examples. But do end-to-end approaches need to learn a unique model for every task? Intuitively, it seems that sharing across tasks should help since all tasks require some common understanding of the environment. In this paper, we attempt to take the next step in data-driven end-to-end learning frameworks: move from the realm of task-specific models to joint learning of multiple robot tasks. In an astonishing result we show that models with multi-task learning tend to perform better than task-specific models trained with same amounts of data. For example, a deep-network learned with 2.5K grasp and 2.5K push examples performs better on grasping than a network trained on 5K grasp examples.",
"In this paper we address three different computer vision tasks using a single basic architecture: depth prediction, surface normal estimation, and semantic labeling. We use a multiscale convolutional network that is able to adapt easily to each task using only small modifications, regressing from the input image to the output map directly. Our method progressively refines predictions using a sequence of scales, and captures many image details without any superpixels or low-level segmentation. We achieve state-of-the-art performance on benchmarks for all three tasks."
]
} |
1708.07785 | 2749710155 | We address the problem of identifying individual cetaceans from images showing the trailing edge of their fins. Given the trailing edge from an unknown individual, we produce a ranking of known individuals from a database. The nicks and notches along the trailing edge define an individual's unique signature. We define a representation based on integral curvature that is robust to changes in viewpoint and pose, and captures the pattern of nicks and notches in a local neighborhood at multiple scales. We explore two ranking methods that use this representation. The first uses a dynamic programming time-warping algorithm to align two representations, and interprets the alignment cost as a measure of similarity. This algorithm also exploits learned spatial weights to downweight matches from regions of unstable curvature. The second interprets the representation as a feature descriptor. Feature keypoints are defined at the local extrema of the representation. Descriptors for the set of known individuals are stored in a tree structure, which allows us to perform queries given the descriptors from an unknown trailing edge. We evaluate the top-k accuracy on two real-world datasets to demonstrate the effectiveness of the curvature representation, achieving top-1 accuracy scores of approximately 95 and 80 for bottlenose dolphins and humpback whales, respectively. | The curvature computed at points along a contour is often used to represent shape information @cite_20 @cite_15 @cite_0 . For differential curvature, this is defined as the change of the angle of the normal vector along the length of the contour @cite_26 . This representation is sensitive to noise, however, and instead we use integral curvature. A fixed shape is placed at points along the contour, while measuring the area of the intersection of the shape with the contour @cite_22 . Using integral curvature to capture shape information is a key part of the Leafsnap system @cite_24 , which classifies leaves by using curvature histograms at multiple scales as shape features. We briefly compare to Leafsnap in . | {
"cite_N": [
"@cite_26",
"@cite_22",
"@cite_0",
"@cite_24",
"@cite_15",
"@cite_20"
],
"mid": [
"109375385",
"2078380962",
"",
"2213241010",
"",
"1991956880"
],
"abstract": [
"Descriptors based on orientation histograms are widely used in computer vision. The spatial pooling involved in these representations provides important invariance properties, yet it is also responsible for the loss of important details. In this paper, we suggest a way to preserve the details described by the local curvature. We propose a descriptor that comprises the direction and magnitude of curvature and naturally expands classical orientation histograms like SIFT and HOG. We demonstrate the general benefit of the expansion exemplarily for image classification, object detection, and descriptor matching.",
"Differential invariants of curves and surfaces such as curvatures and their derivatives play a central role in Geometry Processing. They are, however, sensitive to noise and minor perturbations and do not exhibit the desired multi-scale behavior. Recently, the relationships between differential invariants and certain integrals over small neighborhoods have been used to define efficiently computable integral invariants which have both a geometric meaning and useful stability properties. This paper considers integral invariants defined via distance functions, and the stability analysis of integral invariants in general. Such invariants proved useful for many tasks where the computation of shape characteristics is important. A prominent and recent example is the automatic reassembling of broken objects based on correspondences between fracture surfaces.",
"",
"We describe the first mobile app for identifying plant species using automatic visual recognition. The system --- called Leafsnap --- identifies tree species from photographs of their leaves. Key to this system are computer vision components for discarding non-leaf images, segmenting the leaf from an untextured background, extracting features representing the curvature of the leaf's contour over multiple scales, and identifying the species from a dataset of the 184 trees in the Northeastern United States. Our system obtains state-of-the-art performance on the real-world images from the new Leafsnap Dataset --- the largest of its kind. Throughout the paper, we document many of the practical steps needed to produce a computer vision system such as ours, which currently has nearly a million users.",
"",
"We present an approach that directly uses curvature cues in a discriminative way to perform object recognition. We show that integrating curvature information substantially improves detection results over descriptors that solely rely upon histograms of orientated gradients (HoG). The proposed approach is generic in that it can be easily integrated into state-of-the-art object detection systems. Results on two challenging datasets are presented: ETHZ Shape Dataset and INRIA horses Dataset, improving state-of the-art results using HoG by 7.6 and 12.3 in average precision (AP), respectively. In particular, we achieve higher recall at lower false positive rates."
]
} |
1708.07949 | 2752640714 | Implementation of Neuromorphic Systems using post Complementary Met al-Oxide-Semiconductor (CMOS) technology based Memristive Crossbar Array (MCA) has emerged as a promising solution to enable low-power acceleration of neural networks. However, the recent trend to design Deep Neural Networks (DNNs) for achieving human-like cognitive abilities poses significant challenges towards the scalable design of neuromorphic systems (due to the increase in computation storage demands). Network pruning [7] is a powerful technique to remove redundant connections for designing optimally connected (maximally sparse) DNNs. However, such pruning techniques induce irregular connections that are incoherent to the crossbar structure. Eventually they produce DNNs with highly inefficient hardware realizations (in terms of area and energy). In this work, we propose TraNNsformer - an integrated training framework that transforms DNNs to enable their efficient realization on MCA-based systems. TraNNsformer first prunes the connectivity matrix while forming clusters with the remaining connections. Subsequently, it retrains the network to fine tune the connections and reinforce the clusters. This is done iteratively to transform the original connectivity into an optimally pruned and maximally clustered mapping. We evaluated the proposed framework by transforming different Multi-Layer Perceptron (MLP) based Spiking Neural Networks (SNNs) on a wide range of datasets (MNIST, SVHN and CIFAR10) and executing them on MCA-based systems to analyze the area and energy benefits. Without accuracy loss, TraNNsformer reduces the area (energy) consumption by 28 - 55 (49 - 67 ) with respect to the original network. Compared to network pruning, TraNNsformer achieves 28 - 49 (15 - 29 ) area (energy) savings. Furthermore, TraNNsformer is a technology-aware framework that allows mapping a given DNN to any MCA size permissible by the memristive technology for reliable operations. | The quest to reverse-engineer the powerful cognitive capabilities of the human brain inspired the research in artificial neural networks. Although, the precise structure and functioning of the human brain has not been clearly understood, there is sufficient evidence that suggests the hierarchical organization of neurons (cells) in the brain @cite_16 . Deep Neural Networks (DNNs), inspired from this hierarchy of neurons and synapses in the human brain, have achieved outstanding classification accuracies across myriad cognitive applications including computer vision @cite_0 , speech recognition and natural language processing @cite_5 . Eventually, this has led to its ubiquitous presence in recognition applications across a wide variety of computational platforms that we use in our day-to-day life. For example, Siri, Google Now, Cortana are intelligent personal assistants developed by Apple, Google and Microsoft respectively and run DNNs to recognize external inputs (image, voice etc.). | {
"cite_N": [
"@cite_0",
"@cite_5",
"@cite_16"
],
"mid": [
"2618530766",
"179875071",
"1645800954"
],
"abstract": [
"We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 different classes. On the test data, we achieved top-1 and top-5 error rates of 37.5 and 17.0 , respectively, which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implementation of the convolution operation. To reduce overfitting in the fully connected layers we employed a recently developed regularization method called \"dropout\" that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3 , compared to 26.2 achieved by the second-best entry.",
"A new recurrent neural network based language model (RNN LM) with applications to speech recognition is presented. Results indicate that it is possible to obtain around 50 reduction of perplexity by using mixture of several RNN LMs, compared to a state of the art backoff language model. Speech recognition experiments show around 18 reduction of word error rate on the Wall Street Journal task when comparing models trained on the same amount of data, and around 5 on the much harder NIST RT05 task, even when the backoff model is trained on much more data than the RNN LM. We provide ample empirical evidence to suggest that connectionist language models are superior to standard n-gram techniques, except their high computational (training) complexity. Index Terms: language modeling, recurrent neural networks, speech recognition",
"Deep neural networks such as Convolutional Networks (ConvNets) and Deep Belief Networks (DBNs) represent the state-of-the-art for many machine learning and computer vision classification problems. To overcome the large computational cost of deep networks, spiking deep networks have recently been proposed, given the specialized hardware now available for spiking neural networks (SNNs). However, this has come at the cost of performance losses due to the conversion from analog neural networks (ANNs) without a notion of time, to sparsely firing, event-driven SNNs. Here we analyze the effects of converting deep ANNs into SNNs with respect to the choice of parameters for spiking neurons such as firing rates and thresholds. We present a set of optimization techniques to minimize performance loss in the conversion process for ConvNets and fully connected deep networks. These techniques yield networks that outperform all previous SNNs on the MNIST database to date, and many networks here are close to maximum performance after only 20 ms of simulated time. The techniques include using rectified linear units (ReLUs) with zero bias during training, and using a new weight normalization method to help regulate firing rates. Our method for converting an ANN into an SNN enables low-latency classification with high accuracies already after the first output spike, and compared with previous SNN approaches it yields improved performance without increased training time. The presented analysis and optimization techniques boost the value of spiking deep networks as an attractive framework for neuromorphic computing platforms aimed at fast and efficient pattern recognition."
]
} |
1708.07949 | 2752640714 | Implementation of Neuromorphic Systems using post Complementary Met al-Oxide-Semiconductor (CMOS) technology based Memristive Crossbar Array (MCA) has emerged as a promising solution to enable low-power acceleration of neural networks. However, the recent trend to design Deep Neural Networks (DNNs) for achieving human-like cognitive abilities poses significant challenges towards the scalable design of neuromorphic systems (due to the increase in computation storage demands). Network pruning [7] is a powerful technique to remove redundant connections for designing optimally connected (maximally sparse) DNNs. However, such pruning techniques induce irregular connections that are incoherent to the crossbar structure. Eventually they produce DNNs with highly inefficient hardware realizations (in terms of area and energy). In this work, we propose TraNNsformer - an integrated training framework that transforms DNNs to enable their efficient realization on MCA-based systems. TraNNsformer first prunes the connectivity matrix while forming clusters with the remaining connections. Subsequently, it retrains the network to fine tune the connections and reinforce the clusters. This is done iteratively to transform the original connectivity into an optimally pruned and maximally clustered mapping. We evaluated the proposed framework by transforming different Multi-Layer Perceptron (MLP) based Spiking Neural Networks (SNNs) on a wide range of datasets (MNIST, SVHN and CIFAR10) and executing them on MCA-based systems to analyze the area and energy benefits. Without accuracy loss, TraNNsformer reduces the area (energy) consumption by 28 - 55 (49 - 67 ) with respect to the original network. Compared to network pruning, TraNNsformer achieves 28 - 49 (15 - 29 ) area (energy) savings. Furthermore, TraNNsformer is a technology-aware framework that allows mapping a given DNN to any MCA size permissible by the memristive technology for reliable operations. | The tremendous improvement in DNN performance has been possible due to the surge in the scale of the DNNs. classified handwritten digits in 1998 with less than 1M synapses @cite_11 . proposed AlexNet and won the ImageNet challenge in 2012 by using 650k neurons and 60M synapses @cite_0 . In 2014, developed VGGNet using 138M synapses @cite_9 . proposed a DNN to convert image to natural language employing 130M Convolutional Neural Network (CNN) synapses and 100M Recurrent Neural Network (RNN) synapses @cite_13 . | {
"cite_N": [
"@cite_0",
"@cite_9",
"@cite_13",
"@cite_11"
],
"mid": [
"2618530766",
"1686810756",
"2951805548",
"2310919327"
],
"abstract": [
"We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 different classes. On the test data, we achieved top-1 and top-5 error rates of 37.5 and 17.0 , respectively, which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implementation of the convolution operation. To reduce overfitting in the fully connected layers we employed a recently developed regularization method called \"dropout\" that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3 , compared to 26.2 achieved by the second-best entry.",
"In this work we investigate the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting. Our main contribution is a thorough evaluation of networks of increasing depth using an architecture with very small (3x3) convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers. These findings were the basis of our ImageNet Challenge 2014 submission, where our team secured the first and the second places in the localisation and classification tracks respectively. We also show that our representations generalise well to other datasets, where they achieve state-of-the-art results. We have made our two best-performing ConvNet models publicly available to facilitate further research on the use of deep visual representations in computer vision.",
"We present a model that generates natural language descriptions of images and their regions. Our approach leverages datasets of images and their sentence descriptions to learn about the inter-modal correspondences between language and visual data. Our alignment model is based on a novel combination of Convolutional Neural Networks over image regions, bidirectional Recurrent Neural Networks over sentences, and a structured objective that aligns the two modalities through a multimodal embedding. We then describe a Multimodal Recurrent Neural Network architecture that uses the inferred alignments to learn to generate novel descriptions of image regions. We demonstrate that our alignment model produces state of the art results in retrieval experiments on Flickr8K, Flickr30K and MSCOCO datasets. We then show that the generated descriptions significantly outperform retrieval baselines on both full images and on a new dataset of region-level annotations.",
""
]
} |
1708.07949 | 2752640714 | Implementation of Neuromorphic Systems using post Complementary Met al-Oxide-Semiconductor (CMOS) technology based Memristive Crossbar Array (MCA) has emerged as a promising solution to enable low-power acceleration of neural networks. However, the recent trend to design Deep Neural Networks (DNNs) for achieving human-like cognitive abilities poses significant challenges towards the scalable design of neuromorphic systems (due to the increase in computation storage demands). Network pruning [7] is a powerful technique to remove redundant connections for designing optimally connected (maximally sparse) DNNs. However, such pruning techniques induce irregular connections that are incoherent to the crossbar structure. Eventually they produce DNNs with highly inefficient hardware realizations (in terms of area and energy). In this work, we propose TraNNsformer - an integrated training framework that transforms DNNs to enable their efficient realization on MCA-based systems. TraNNsformer first prunes the connectivity matrix while forming clusters with the remaining connections. Subsequently, it retrains the network to fine tune the connections and reinforce the clusters. This is done iteratively to transform the original connectivity into an optimally pruned and maximally clustered mapping. We evaluated the proposed framework by transforming different Multi-Layer Perceptron (MLP) based Spiking Neural Networks (SNNs) on a wide range of datasets (MNIST, SVHN and CIFAR10) and executing them on MCA-based systems to analyze the area and energy benefits. Without accuracy loss, TraNNsformer reduces the area (energy) consumption by 28 - 55 (49 - 67 ) with respect to the original network. Compared to network pruning, TraNNsformer achieves 28 - 49 (15 - 29 ) area (energy) savings. Furthermore, TraNNsformer is a technology-aware framework that allows mapping a given DNN to any MCA size permissible by the memristive technology for reliable operations. | Prior work on learning structured connectivity in DNNs during the training process has focused on transformations for CMOS based architectures (Graphics Processing Unit) @cite_2 @cite_17 . While, post-CMOS based MCA architectures were explored in @cite_20 , they focused on clustering the synapses after the training process finishes. Our work distinguishes from the prior works as it proposes an integrated training framework to maximize the benefits from post-CMOS MCA based systems by designing DNN transformations, which conform to MCA's structural rigidity. Additionally, we also show that offline clustering (clustering after training) is unable to preserve the benefits of sparsity at the hardware level. | {
"cite_N": [
"@cite_17",
"@cite_20",
"@cite_2"
],
"mid": [
"",
"2001480795",
"2952826672"
],
"abstract": [
"",
"In implementations of neuromorphic computing systems (NCS), memristor and its crossbar topology have been widely used to realize fully connected neural networks. However, many neural networks utilized in real applications often have a sparse connectivity, which is hard to be efficiently mapped to a crossbar structure. Moreover, the scale of the neural networks is normally much larger than that can be offered by the latest integration technology of memristor crossbars. In this work, we propose AutoNCS -- an EDA framework that can automate the NCS designs that combine memristor crossbars and discrete synapse modules. The connections of the neural networks are clustered to improve the utilization of the memristor elements in crossbar structures by taking into account the physical design cost of the NCS. Our results show that AutoNCS can substantially enhance the utilization efficiency of memristor crossbars while reducing the wirelength, area and delay of the physical designs of the NCS.",
"Real time application of deep learning algorithms is often hindered by high computational complexity and frequent memory accesses. Network pruning is a promising technique to solve this problem. However, pruning usually results in irregular network connections that not only demand extra representation efforts but also do not fit well on parallel computation. We introduce structured sparsity at various scales for convolutional neural networks, which are channel wise, kernel wise and intra kernel strided sparsity. This structured sparsity is very advantageous for direct computational resource savings on embedded computers, parallel computing environments and hardware based systems. To decide the importance of network connections and paths, the proposed method uses a particle filtering approach. The importance weight of each particle is assigned by computing the misclassification rate with corresponding connectivity pattern. The pruned network is re-trained to compensate for the losses due to pruning. While implementing convolutions as matrix products, we particularly show that intra kernel strided sparsity with a simple constraint can significantly reduce the size of kernel and feature map matrices. The pruned network is finally fixed point optimized with reduced word length precision. This results in significant reduction in the total storage size providing advantages for on-chip memory based implementations of deep neural networks."
]
} |
1708.07989 | 2750721850 | We focus on a scenario where two wireless source nodes wish to exchange confidential information via an RF energy harvesting untrusted two-way relay. Despite its cooperation in forwarding the information, the relay is considered untrusted out of the concern that it might attempt to decode the confidential information that is being relayed. To discourage the eavesdropping intention of the relay, we use a friendly jammer. Under the total power constraint, to maximize the sum-secrecy rate, we allocate the power among the sources and the jammer optimally and calculate the optimal power splitting ratio to balance between the energy harvesting and the information processing at the relay. We further examine the effect of imperfect channel state information at both sources on the sum-secrecy rate. Numerical results highlight the role of the jammer in achieving the secure communication under channel estimation errors. We have shown that, as the channel estimation error on any of the channels increases, the power allocated to the jammer decreases to abate the interference caused to the confidential information reception due to the imperfect cancellation of jammer's signal. | @cite_0 , authors show that the cooperation by an untrusted relay can be beneficial and can achieve higher secrecy rate than just treating the untrusted relay as a pure eavesdropper. In @cite_9 , authors investigate the secure communication in untrusted two-way relay systems with the help of external friendly jammers and show that, though it is possible to achieve a non-zero secrecy rate without the friendly jammers, the secrecy rate at both sources can effectively be improved with the help from an external friendly jammer. @cite_12 , authors have focused on improving the energy efficiency while achieving the minimum secrecy rate for the untrusted two-way relay. The works in @cite_0 @cite_11 @cite_20 @cite_22 @cite_23 @cite_9 @cite_12 assume that the relay is a conventional node and has a stable power supply. As to energy harvesting untrusted relaying, the works in @cite_5 @cite_6 @cite_21 analyze the effect of untrusted energy harvesting one-way relay on the secure communication between two legitimate nodes. To the best of our knowledge, for energy harvesting two-way untrusted relay, the problem of achieving the secure communication has not been yet studied in the literature. | {
"cite_N": [
"@cite_22",
"@cite_9",
"@cite_21",
"@cite_6",
"@cite_0",
"@cite_23",
"@cite_5",
"@cite_12",
"@cite_20",
"@cite_11"
],
"mid": [
"2016807091",
"2963791781",
"2175451653",
"2098294262",
"2097480027",
"2189096384",
"2962874228",
"2318872164",
"1970578484",
"2045570050"
],
"abstract": [
"This paper studies the problem of secure transmission in dual-hop cooperative networks with untrusted relays, where each relay acts as both a potential helper and an eavesdropper. A security-aware relaying scheme is proposed, which employs the alternate jamming and secrecy-enhanced relay selection to prevent the confidential message from being eavesdropped by the untrusted relays. To evaluate the performance of the proposed strategies, we derive the lower bound of the achievable ergodic secrecy rate (ESR), and conduct the asymptotic analysis to examine how the ESR scales as the number of relays increases.",
"In this paper, we consider a two-way relay system where the two sources can only communicate through an untrusted intermediate relay and investigate the physical layer security issue in this two-way untrusted relay scenario. Specifically, we regard the intermediate relay as an eavesdropper from which the information transmitted by the sources needs to be kept confidential, despite the fact that its cooperation in relaying this information is essential. We first indicate that a nonzero secrecy rate is indeed achievable in this two-way untrusted relay system even without the help of external friendly jammers. As for the system with friendly jammers, after further analysis, we can obtain the secrecy rate of the sources can be effectively improved by utilizing proper jamming power from the friendly jammers. Then, we formulate a Stackelberg game between the sources and the friendly jammers as a power control scheme to achieve the optimized secrecy rate of the sources, in which the sources are treated as the sole buyer and the friendly jammers are the sellers. In addition, the optimal solutions of the jamming power and the asking prices are given, and a distributed updating algorithm to obtain the Stackelberg equilibrium is provided for the proposed game. Finally, the simulation results verify the properties and efficiency of the proposed Stackelberg-game-based scheme.",
"The secrecy capacity (SC) from the source to a destination in a dual-hop amplify-and-forward (AF) relay network using cooperative jamming (CJ) is investigated. Conditions for achieving a positive SC are derived. An energy constrained untrusted cooperative relaying protocol is considered in which the relay forwards information from the source to the destination while harvesting energy from the source signal and the destination jamming signal. An optimization problem is proposed to maximize the SC subject to the energy constraints. A method is proposed to find the optimal power split ratio (PSR) which maximizes the SC. An approximate expression for the optimal PSR is also derived. Results show that a mutually beneficial relationship exists between source-destination pair and the relay. This allows the relay to achieve a net energy gain while the secrecy of the confidential message from the source to the destination is protected.",
"In this work, we maximize the secrecy rate of the wireless-powered untrusted relay network by jointly designing power splitting (PS) ratio and relay beamforming with the proposed global optimal algorithm (GOA) and local optimal algorithm (LOA). Different from the literature, artificial noise (AN) sent by the destination not only degrades the channel condition of the eavesdropper to improve the secrecy rate, but also becomes a new source of energy powering the untrusted relay based on PS. Hence, it is of high economic benefits and efficiency to take advantage of AN compared with the literature. Simulation results show that LOA can achieve satisfactory secrecy rate performance compared with that of GOA, but with less computation time.",
"We consider the communication scenario where a source-destination pair wishes to keep the information secret from a relay node despite wanting to enlist its help. For this scenario, an interesting question is whether the relay node should be deployed at all. That is, whether cooperation with an untrusted relay node can ever be beneficial. We first provide an achievable secrecy rate for the general untrusted relay channel, and proceed to investigate this question for two types of relay networks with orthogonal components. For the first model, there is an orthogonal link from the source to the relay. For the second model, there is an orthogonal link from the relay to the destination. For the first model, we find the equivocation capacity region and show that answer is negative. In contrast, for the second model, we find that the answer is positive. Specifically, we show, by means of the achievable secrecy rate based on compress-and-forward, that by asking the untrusted relay node to relay information, we can achieve a higher secrecy rate than just treating the relay as an eavesdropper. For a special class of the second model, where the relay is not interfering itself, we derive an upper bound for the secrecy rate using an argument whose net effect is to separate the eavesdropper from the relay. The merit of the new upper bound is demonstrated on two channels that belong to this special class. The Gaussian case of the second model mentioned above benefits from this approach in that the new upper bound improves the previously known bounds. For the Cover-Kim deterministic relay channel, the new upper bound finds the secrecy capacity when the source-destination link is not worse than the source-relay link, by matching with achievable rate we present.",
"In this paper, we investigate secure transmission in untrusted amplify-and-forward half-duplex relaying networks with the help of cooperative jamming at the destination (CJD). Under the assumption of full channel state information (CSI), conventional CJD using self-interference cancelation at the destination is efficient when the untrusted relay has no capability to suppress the jamming signal. However, if the source and destination are equipped with a single antenna and the only untrusted relay is equipped with N multiple antennas, it can remove the jamming signal from the received signal by linear filters and the full multiplexing gain of relaying cannot be achievable with the conventional CJD due to the saturation of the secrecy rate at the high transmit power regime. We propose in this paper new CJD scheme where neither destination nor relay can acquire CSI of relay-destination link. Our proposed scheme utilizes zero-forcing cancelation based on known jamming signals instead of self-interference subtraction, while the untrusted relay cannot suppress the jamming signals due to the lack of CSI. We show that the secrecy rate of the proposed scheme can enjoy a half of multiplexing gain in half-duplex relaying while that of conventional CJD is saturated at high transmit power for N ≥2. The impact of channel estimation error at the destination is also investigated to show the robustness of the proposed scheme against strong estimation errors.",
"The broadcast nature of the wireless medium allows unintended users to eavesdrop on confidential information transmission. In this regard, we investigate the problem of secure communication between a source and a destination via a wireless energy harvesting untrusted node that acts as a helper to relay the information; however, the source and destination nodes wish to keep the information confidential from the relay node. To realize the positive secrecy rate, we use destination-assisted jamming. Being an energy-starved node, the untrusted relay harvests energy from the received radio-frequency (RF) signals, which include the source's information signal and the destination's jamming signal. Thus, we utilize the jamming signal efficiently by leveraging it as a useful energy source. At the relay, to enable energy harvesting and information processing, we adopt power splitting (PS) and time switching (TS) policies. To evaluate the secrecy performance of this proposed scenario, we derive analytical expressions for two important metrics, viz., the secrecy outage probability and the ergodic secrecy rate. The numerical analysis reveals design insights into the effects of different system parameters such as PS ratio, energy harvesting time, target secrecy rate, transmit signal-to-noise ratio (SNR), relay location, and energy conversion efficiency factor, on secrecy performance. Specifically, the PS policy achieves better optimal secrecy outage probability and optimal ergodic secrecy rate than that of the TS policy at higher target secrecy rate and transmit SNR, respectively.",
"In this paper, energy-efficient secure communications via untrusted two-way relaying are investigated considering physical layer security to prevent the relays from intercepting the confidential information of users. The performance metric of secure energy efficiency (EE), defined as the ratio of the secrecy sum rate to the total power consumption, is maximized by jointly optimizing power allocation for all nodes, with the constraints of the maximum allowed power and the minimum target secrecy rate. To deal with this intractable nonconvex optimization problem, some optimization methods termed as fractional programming, alternate optimization, penalty function method, and difference of convex functions programming, are jointly applied to address a solution scheme with comparatively lower complexity. With these above-mentioned optimization methods, the primal problem is transformed into simple subproblems hierarchically so as to adopt the corresponding optimization algorithm. By simulations, the achievable secure EE, the secrecy sum rate and the total transmission power of the proposed scheme are compared with those of secrecy sum rate maximization. It is demonstrated that the proposed scheme can improve secure EE remarkably yet at the cost of secrecy sum rate loss. This fact also reveals the inherent tradeoff between energy and security.",
"We consider the problem of secure transmission in two-hop amplify-and-forward untrusted relay networks. We analyze the ergodic secrecy capacity (ESC) and present compact expressions for the ESC in the high signal-to-noise ratio regime. We also examine the impact of large scale antenna arrays at either the source or the destination. For large antenna arrays at the source, we confirm that the ESC is solely determined by the channel between the relay and the destination. For very large antenna arrays at the destination, we confirm that the ESC is solely determined by the channel between the source and the relay.",
"We investigate a relay network where the source can potentially utilize an untrusted non-regenerative relay to augment its direct transmission of a confidential message to the destination. Since the relay is untrusted, it is desirable to protect the confidential data from it while simultaneously making use of it to increase the reliability of the transmission. We first examine the secrecy outage probability (SOP) of the network assuming a single antenna relay, and calculate the exact SOP for three different schemes: direct transmission without using the relay, conventional non-regenerative relaying, and cooperative jamming by the destination. Subsequently, we conduct an asymptotic analysis of the SOPs to determine the optimal policies in different operating regimes. We then generalize to the multi-antenna relay case and investigate the impact of the number of relay antennas on the secrecy performance. Finally, we study a scenario where the relay has only a single RF chain which necessitates an antenna selection scheme, and we show that unlike the case where all antennas are used, under certain conditions the cooperative jamming scheme with antenna selection provides a diversity advantage for the receiver. Numerical results are presented to verify the theoretical predictions of the preferred transmission policies."
]
} |
1708.07335 | 2748973159 | Frame-level visual features are generally aggregated in time with the techniques such as LSTM, Fisher Vectors, NetVLAD etc. to produce a robust video-level representation. We here introduce a learnable aggregation technique whose primary objective is to retain short-time temporal structure between frame-level features and their spatial interdependencies in the representation. Also, it can be easily adapted to the cases where there have very scarce training samples. We evaluate the method on a real-fake expression prediction dataset to demonstrate its superiority. Our method obtains 65 score on the test dataset in the official MAP evaluation and there is only one misclassified decision with the best reported result in the Chalearn Challenge (i.e. 66:7 ) . Lastly, we believe that this method can be extended to different problems such as action event recognition in future. | : Initial attempts to encode visual content of a video are broadly formulated with hand-crafted features. These features are extracted from either local patches @cite_3 @cite_28 and or local trajectories @cite_29 by exploiting regional gradient features. Later, they are aggregated over frames with various pooling techniques to obtain a single compact representation @cite_19 @cite_21 . Recently, feature extraction and aggregation steps are adapted to learnable modules. Frame-level @cite_32 @cite_26 , spatio-temporal-level @cite_30 @cite_18 and motion-level @cite_37 convolutions are extensively studied to extract robust features from videos. Moreover, trainable pooling techniques are introduced @cite_10 @cite_5 to boost discriminative representation learning rather than unsupervised feature-space clustering @cite_19 @cite_21 . | {
"cite_N": [
"@cite_30",
"@cite_18",
"@cite_26",
"@cite_37",
"@cite_28",
"@cite_29",
"@cite_21",
"@cite_32",
"@cite_3",
"@cite_19",
"@cite_5",
"@cite_10"
],
"mid": [
"2235034809",
"",
"",
"2952186347",
"",
"",
"",
"1950136256",
"2108333036",
"1606858007",
"",
"2950687050"
],
"abstract": [
"Typical human actions last several seconds and exhibit characteristic spatio-temporal structure. Recent methods attempt to capture this structure and learn action representations with convolutional neural networks. Such representations, however, are typically learned at the level of a few video frames failing to model actions at their full temporal extent. In this work we learn video representations using neural networks with long-term temporal convolutions (LTC). We demonstrate that LTC-CNN models with increased temporal extents improve the accuracy of action recognition. We also study the impact of different low-level representations, such as raw values of video pixels and optical flow vector fields and demonstrate the importance of high-quality optical flow estimation for learning accurate action models. We report state-of-the-art results on two challenging benchmarks for human action recognition UCF101 (92.7 ) and HMDB51 (67.2 ).",
"",
"",
"We investigate architectures of discriminatively trained deep Convolutional Networks (ConvNets) for action recognition in video. The challenge is to capture the complementary information on appearance from still frames and motion between frames. We also aim to generalise the best performing hand-crafted features within a data-driven learning framework. Our contribution is three-fold. First, we propose a two-stream ConvNet architecture which incorporates spatial and temporal networks. Second, we demonstrate that a ConvNet trained on multi-frame dense optical flow is able to achieve very good performance in spite of limited training data. Finally, we show that multi-task learning, applied to two different action classification datasets, can be used to increase the amount of training data and improve the performance on both. Our architecture is trained and evaluated on the standard video actions benchmarks of UCF-101 and HMDB-51, where it is competitive with the state of the art. It also exceeds by a large margin previous attempts to use deep nets for video classification.",
"",
"",
"",
"In this paper, we propose a discriminative video representation for event detection over a large scale video dataset when only limited hardware resources are available. The focus of this paper is to effectively leverage deep Convolutional Neural Networks (CNNs) to advance event detection, where only frame level static descriptors can be extracted by the existing CNN toolkits. This paper makes two contributions to the inference of CNN video representation. First, while average pooling and max pooling have long been the standard approaches to aggregating frame level static features, we show that performance can be significantly improved by taking advantage of an appropriate encoding method. Second, we propose using a set of latent concept descriptors as the frame descriptor, which enriches visual information while keeping it computationally affordable. The integration of the two contributions results in a new state-of-the-art performance in event detection over the largest video datasets. Compared to improved Dense Trajectories, which has been recognized as the best video representation for event detection, our new representation improves the Mean Average Precision (mAP) from 27.6 to 36.8 for the TRECVID MEDTest 14 dataset and from 34.0 to 44.6 for the TRECVID MEDTest 13 dataset.",
"In this paper we introduce a 3-dimensional (3D) SIFT descriptor for video or 3D imagery such as MRI data. We also show how this new descriptor is able to better represent the 3D nature of video data in the application of action recognition. This paper will show how 3D SIFT is able to outperform previously used description methods in an elegant and efficient manner. We use a bag of words approach to represent videos, and present a method to discover relationships between spatio-temporal words in order to better describe the video data.",
"The Fisher kernel (FK) is a generic framework which combines the benefits of generative and discriminative approaches. In the context of image classification the FK was shown to extend the popular bag-of-visual-words (BOV) by going beyond count statistics. However, in practice, this enriched representation has not yet shown its superiority over the BOV. In the first part we show that with several well-motivated modifications over the original framework we can boost the accuracy of the FK. On PASCAL VOC 2007 we increase the Average Precision (AP) from 47.9 to 58.3 . Similarly, we demonstrate state-of-the-art accuracy on CalTech 256. A major advantage is that these results are obtained using only SIFT descriptors and costless linear classifiers. Equipped with this representation, we can now explore image classification on a larger scale. In the second part, as an application, we compare two abundant resources of labeled images to learn classifiers: ImageNet and Flickr groups. In an evaluation involving hundreds of thousands of training images we show that classifiers learned on Flickr groups perform surprisingly well (although they were not intended for this purpose) and that they can complement classifiers learned on more carefully annotated datasets.",
"",
"In this work, we introduce a new video representation for action classification that aggregates local convolutional features across the entire spatio-temporal extent of the video. We do so by integrating state-of-the-art two-stream networks with learnable spatio-temporal feature aggregation. The resulting architecture is end-to-end trainable for whole-video classification. We investigate different strategies for pooling across space and time and combining signals from the different streams. We find that: (i) it is important to pool jointly across space and time, but (ii) appearance and motion streams are best aggregated into their own separate representations. Finally, we show that our representation outperforms the two-stream base architecture by a large margin (13 relative) as well as out-performs other baselines with comparable base architectures on HMDB51, UCF101, and Charades video classification benchmarks."
]
} |
1708.07335 | 2748973159 | Frame-level visual features are generally aggregated in time with the techniques such as LSTM, Fisher Vectors, NetVLAD etc. to produce a robust video-level representation. We here introduce a learnable aggregation technique whose primary objective is to retain short-time temporal structure between frame-level features and their spatial interdependencies in the representation. Also, it can be easily adapted to the cases where there have very scarce training samples. We evaluate the method on a real-fake expression prediction dataset to demonstrate its superiority. Our method obtains 65 score on the test dataset in the official MAP evaluation and there is only one misclassified decision with the best reported result in the Chalearn Challenge (i.e. 66:7 ) . Lastly, we believe that this method can be extended to different problems such as action event recognition in future. | Even if there are vast numbers of studies on facial emotion detection, the prediction of real-fake expression is limited and most of which are for still images @cite_36 . Also, they are usually based on hand-crafted features. Therefore, reviewing the literature from the perspective of cognitive science might be more informative in order to understand and propose a proper technique for real-fake expression prediction. In particular, @cite_38 @cite_31 show that small emotional changes in eyes and mount movements can be used to interpret an expression as genuine (real) or deceptive (fake) from the sequence of faces. This indicates that both spatial interdependencies of face parts and their sequential structure should be considered in the final model. | {
"cite_N": [
"@cite_36",
"@cite_38",
"@cite_31"
],
"mid": [
"2005733474",
"2282546360",
""
],
"abstract": [
"We present initial results from the application of an automated facial expression recognition system to spontaneous facial expressions of pain. In this study, 26 participants were videotaped under three experimental conditions: baseline, posed pain, and real pain. The real pain condition consisted of cold pressor pain induced by submerging the arm in ice water. Our goal was to (1) assess whether the automated measurements were consistent with expression measurements obtained by human experts, and (2) develop a classifier to automatically differentiate real from faked pain in a subject-independent manner from the automated measurements. We employed a machine learning approach in a two-stage system. In the first stage, a set of 20 detectors for facial actions from the Facial Action Coding System operated on the continuous video stream. These data were then passed to a second machine learning stage, in which a classifier was trained to detect the difference between expressions of real pain and fake pain. Naive human subjects tested on the same videos were at chance for differentiating faked from real pain, obtaining only 49 accuracy. The automated system was successfully able to differentiate faked from real pain. In an analysis of 26 subjects with faked pain before real pain, the system obtained 88 correct for subject independent discrimination of real versus fake pain on a 2-alternative forced choice. Moreover, the most discriminative facial actions in the automated system were consistent with findings using human expert FACS codes.",
"Hiding true emotions: micro-expressions in eyes retrospectively concealed by mouth movements",
""
]
} |
1708.07576 | 2746618409 | Recurrent neural networks (RNNs) represent the state of the art in translation, image captioning, and speech recognition. They are also capable of learning algorithmic tasks such as long addition, copying, and sorting from a set of training examples. We demonstrate that RNNs can learn decryption algorithms -- the mappings from plaintext to ciphertext -- for three polyalphabetic ciphers (Vigenere, Autokey, and Enigma). Most notably, we demonstrate that an RNN with a 3000-unit Long Short-Term Memory (LSTM) cell can learn the decryption function of the Enigma machine. We argue that our model learns efficient internal representations of these ciphers 1) by exploring activations of individual memory neurons and 2) by comparing memory usage across the three ciphers. To be clear, our work is not aimed at 'cracking' the Enigma cipher. However, we do show that our model can perform elementary cryptanalysis by running known-plaintext attacks on the Vigenere and Autokey ciphers. Our results indicate that RNNs can learn algorithmic representations of black box polyalphabetic ciphers and that these representations are useful for cryptanalysis. | Recurrent models, in particular those that use Long Short-Term Memory (LSTM), are a powerful tool for manipulating sequential data. Notable applications include state-of-the art results in handwriting recognition and generation @cite_17 , speech recognition @cite_8 , machine translation @cite_1 , deep reinforcement learning @cite_12 , and image captioning @cite_15 . Like these works, we frame our application of RNNs as a sequence-to-sequence translation task. Unlike these works, our translation task requires 1) a keyphrase and 2) reconstructing a deterministic algorithm. Fortunately, a wealth of previous work has focused on using RNNs to learn algorithms. | {
"cite_N": [
"@cite_8",
"@cite_1",
"@cite_15",
"@cite_12",
"@cite_17"
],
"mid": [
"2950689855",
"2949888546",
"2951805548",
"",
"1810943226"
],
"abstract": [
"Recurrent neural networks (RNNs) are a powerful model for sequential data. End-to-end training methods such as Connectionist Temporal Classification make it possible to train RNNs for sequence labelling problems where the input-output alignment is unknown. The combination of these methods with the Long Short-term Memory RNN architecture has proved particularly fruitful, delivering state-of-the-art results in cursive handwriting recognition. However RNN performance in speech recognition has so far been disappointing, with better results returned by deep feedforward networks. This paper investigates , which combine the multiple levels of representation that have proved so effective in deep networks with the flexible use of long range context that empowers RNNs. When trained end-to-end with suitable regularisation, we find that deep Long Short-term Memory RNNs achieve a test set error of 17.7 on the TIMIT phoneme recognition benchmark, which to our knowledge is the best recorded score.",
"Deep Neural Networks (DNNs) are powerful models that have achieved excellent performance on difficult learning tasks. Although DNNs work well whenever large labeled training sets are available, they cannot be used to map sequences to sequences. In this paper, we present a general end-to-end approach to sequence learning that makes minimal assumptions on the sequence structure. Our method uses a multilayered Long Short-Term Memory (LSTM) to map the input sequence to a vector of a fixed dimensionality, and then another deep LSTM to decode the target sequence from the vector. Our main result is that on an English to French translation task from the WMT'14 dataset, the translations produced by the LSTM achieve a BLEU score of 34.8 on the entire test set, where the LSTM's BLEU score was penalized on out-of-vocabulary words. Additionally, the LSTM did not have difficulty on long sentences. For comparison, a phrase-based SMT system achieves a BLEU score of 33.3 on the same dataset. When we used the LSTM to rerank the 1000 hypotheses produced by the aforementioned SMT system, its BLEU score increases to 36.5, which is close to the previous best result on this task. The LSTM also learned sensible phrase and sentence representations that are sensitive to word order and are relatively invariant to the active and the passive voice. Finally, we found that reversing the order of the words in all source sentences (but not target sentences) improved the LSTM's performance markedly, because doing so introduced many short term dependencies between the source and the target sentence which made the optimization problem easier.",
"We present a model that generates natural language descriptions of images and their regions. Our approach leverages datasets of images and their sentence descriptions to learn about the inter-modal correspondences between language and visual data. Our alignment model is based on a novel combination of Convolutional Neural Networks over image regions, bidirectional Recurrent Neural Networks over sentences, and a structured objective that aligns the two modalities through a multimodal embedding. We then describe a Multimodal Recurrent Neural Network architecture that uses the inferred alignments to learn to generate novel descriptions of image regions. We demonstrate that our alignment model produces state of the art results in retrieval experiments on Flickr8K, Flickr30K and MSCOCO datasets. We then show that the generated descriptions significantly outperform retrieval baselines on both full images and on a new dataset of region-level annotations.",
"",
"This paper shows how Long Short-term Memory recurrent neural networks can be used to generate complex sequences with long-range structure, simply by predicting one data point at a time. The approach is demonstrated for text (where the data are discrete) and online handwriting (where the data are real-valued). It is then extended to handwriting synthesis by allowing the network to condition its predictions on a text sequence. The resulting system is able to generate highly realistic cursive handwriting in a wide variety of styles."
]
} |
1708.07576 | 2746618409 | Recurrent neural networks (RNNs) represent the state of the art in translation, image captioning, and speech recognition. They are also capable of learning algorithmic tasks such as long addition, copying, and sorting from a set of training examples. We demonstrate that RNNs can learn decryption algorithms -- the mappings from plaintext to ciphertext -- for three polyalphabetic ciphers (Vigenere, Autokey, and Enigma). Most notably, we demonstrate that an RNN with a 3000-unit Long Short-Term Memory (LSTM) cell can learn the decryption function of the Enigma machine. We argue that our model learns efficient internal representations of these ciphers 1) by exploring activations of individual memory neurons and 2) by comparing memory usage across the three ciphers. To be clear, our work is not aimed at 'cracking' the Enigma cipher. However, we do show that our model can perform elementary cryptanalysis by running known-plaintext attacks on the Vigenere and Autokey ciphers. Our results indicate that RNNs can learn algorithmic representations of black box polyalphabetic ciphers and that these representations are useful for cryptanalysis. | Past work has shown that RNNs can find general solutions to algorithmic tasks. Zaremba and Sutskever @cite_11 trained LSTMs to add 9-digit numbers with 99 In the 1930s, Allied nations had not yet captured a physical Enigma machine. Cryptographers such as the Polish Marian Rejewski were forced to compare plaintext, keyphrase, and ciphertext messages with each other to infer the mechanics of the machine. After several years of carefully analyzing these messages, Rejewski and the Polish Cipher Bureau were able to construct 'Enigma doubles' without ever having seen an actual Enigma machine @cite_19 . This is the same problem we trained our model to solve. Like Rejewski, our model uncovers the logic of the Enigma by looking for statistical patterns in a large number of plaintext, keyphrase, and ciphertext examples. We should note, however, that Rejewski needed far less data to make the same generalizations. Later in World War II, British cryptographers led by Alan Turing helped to actually crack the Enigma. As Turing's approach capitalized on operator error and expert knowledge about the Enigma, we consider it beyond the scope of this work @cite_9 . | {
"cite_N": [
"@cite_19",
"@cite_9",
"@cite_11"
],
"mid": [
"1971846196",
"1598444122",
""
],
"abstract": [
"The paper gives a personal view of work in the Polish Cipher Bureau from 1932 to 1939 as mathematicians worked to decipher the codes of the military version of the Enigma. The author, who was a participant, relates details of the device and the successes and frustrations involved in the work. He also describes mathematical principles that enabled him and his colleagues to break successive versions of the Enigma code and to construct technical devices (cyclometers and \"bombs\") that facilitated decipherment of Enigma-coded messages.",
"List of Illustrations. Acknowledgements. Introduction. Prologue. The Betrayal -- Belgium and Germany, 1931. The Leak -- Poland, Belgium and Germany, 1929--38. An Inspired Guess -- Poland, 1932. A Terrible Mistake -- Poland, 1933--9. Flight -- Germany, Poland and England, 1939--40. The First Capture -- Scotland, 1940. Mission Impossible -- Norway and Bletchley Park, 1940. Keeping the Enigma Secret -- France and Bletchley Park, May--September 1940. Deadlock -- Bletchley Park, August--October, 1940. The Italian Affair -- Bletchley Park and the Mediterranean, March 1941. The End of the Beginning -- Norway, March 1941. Breakthrough -- North of Iceland, May 1941. Operation Primrose -- The Atlantic, May 1941. The Knock--Out Blow -- North of Iceland, June 1941. Suspicion -- Bletchley Park, the Atlantic and Berlin, May--October 1941. A Two--Edged Sword -- The Atlantic and the Cape Verde Islands, September 1941. Living Dangerously -- The South Atlantic and Norway, November 1941--March 1942. The Hunt for the Bigram Tables -- Bletchley Park and Norway, December 1941. Black Out -- The Barents Sea, Bletchley Park and the Admiralty, February--July 1942. Breaking the Deadlock -- The Mediterranean and Bletchley Park, October--December 1942. The Turning Point -- South of France, the Mediterranean and the Atlantic, November 1942--September 1943. Trapped -- South of France, November 1942--March 1943. The Arrest -- Berlin, March--September 1943. Sinking the Scharnhorst -- The Barents Sea, December 1943. Operation Covered -- Paris, the Indian Ocean and the Atlantic, August 1943--March 1944. The Last Hiccough -- Germany, France and the South Atlantic, March--June 1944. Epilogue -- Where did they Go? Chronology. Glossary. Appendix 1: Polish Codebreaking Techniques. Appendix 2: The Bombe. Appendix 3: Naval Enigma. Appendix 4: Cillis. Appendix 5: Rodding. Appendix 6: Naval Enigma Offizier. Notes. Bibliography. Index.",
""
]
} |
1708.07576 | 2746618409 | Recurrent neural networks (RNNs) represent the state of the art in translation, image captioning, and speech recognition. They are also capable of learning algorithmic tasks such as long addition, copying, and sorting from a set of training examples. We demonstrate that RNNs can learn decryption algorithms -- the mappings from plaintext to ciphertext -- for three polyalphabetic ciphers (Vigenere, Autokey, and Enigma). Most notably, we demonstrate that an RNN with a 3000-unit Long Short-Term Memory (LSTM) cell can learn the decryption function of the Enigma machine. We argue that our model learns efficient internal representations of these ciphers 1) by exploring activations of individual memory neurons and 2) by comparing memory usage across the three ciphers. To be clear, our work is not aimed at 'cracking' the Enigma cipher. However, we do show that our model can perform elementary cryptanalysis by running known-plaintext attacks on the Vigenere and Autokey ciphers. Our results indicate that RNNs can learn algorithmic representations of black box polyalphabetic ciphers and that these representations are useful for cryptanalysis. | Characterizing unknown ciphers is a central problem in cryptography. The comprehensive @cite_16 lists a wealth of methods, from frequency analysis to chosen-plaintext attacks. Papers such as @cite_4 offer additional methods. While these methods can be effective under the right conditions, they do not generalize well outside certain classes of ciphers. This has led researchers to propose several machine learning-based approaches. One work @cite_7 used genetic algorithms to recover the secret key of a simple substitution cipher. A review paper by @cite_0 proposes cipher classification with machine learning. | {
"cite_N": [
"@cite_0",
"@cite_16",
"@cite_4",
"@cite_7"
],
"mid": [
"2343818797",
"1585665690",
"2752820220",
"2070983455"
],
"abstract": [
"Cryptanalysis is very important step for auditing and checking strength of any cryptosystem. Some of these cryptosystem ensures confidentiality and security of large information exchange from source to destination using symmetric key cryptography. The cryptanalyst investigates the strength and identifies the weakness of the key as well as enciphering algorithm. With the increase in key size, the time and effort required predicting the correct key increases. So, the Trend of increasing key size from 1 Byte to 8 Bytes to strengthen the cryptosystem and hence algorithm continues with compromise on the cost of time and computation. Automatic Variable Key (AVK) based symmetric key cryptosystem is an alternative to this style by fixing up key size and adding security level direction. Whenever any new cryptographic method is invented to replace existing vulnerable cryptographic method, it’s deep analysis from all perspectives (Hacker Cryptanalyst as well as User) is desirable and proper study and evaluation of its performance is must. New cryptic techniques may exploit benefits of advances in computational methods like ANN, GA, SI etc. These techniques for cryptanalysis are changing drastically to reduce cryptographic complexity. In this paper a detailed survey and direction of development work has been conducted. The work compares these new methods with state of art approaches and presents future scope and directions from the cryptic mining perspectives.",
"CRYPTOGRAPHIC PROTOCOLS. Protocol Building Blocks. Basic Protocols. Intermediate Protocols. Advanced Protocols. Esoteric Protocols. CRYPTOGRAPHIC TECHNIQUES. Key Length. Key Management. Algorithm Types and Modes. Using Algorithms. CRYPTOGRAPHIC ALGORITHMS. Data Encryption Standard (DES). Other Block Ciphers. Other Stream Ciphers and Real Random-Sequence Generators. Public-Key Algorithms. Special Algorithms for Protocols. THE REAL WORLD. Example Implementations. Politics. SOURCE CODE.source Code. References.",
"",
"This paper considers a new approach to cryptanalysis based on the application of a directed random search algorithm called a genetic algorithm. It is shown that such a algorithm could be used to discover the key for a simple substitution cipher."
]
} |
1708.07576 | 2746618409 | Recurrent neural networks (RNNs) represent the state of the art in translation, image captioning, and speech recognition. They are also capable of learning algorithmic tasks such as long addition, copying, and sorting from a set of training examples. We demonstrate that RNNs can learn decryption algorithms -- the mappings from plaintext to ciphertext -- for three polyalphabetic ciphers (Vigenere, Autokey, and Enigma). Most notably, we demonstrate that an RNN with a 3000-unit Long Short-Term Memory (LSTM) cell can learn the decryption function of the Enigma machine. We argue that our model learns efficient internal representations of these ciphers 1) by exploring activations of individual memory neurons and 2) by comparing memory usage across the three ciphers. To be clear, our work is not aimed at 'cracking' the Enigma cipher. However, we do show that our model can perform elementary cryptanalysis by running known-plaintext attacks on the Vigenere and Autokey ciphers. Our results indicate that RNNs can learn algorithmic representations of black box polyalphabetic ciphers and that these representations are useful for cryptanalysis. | @cite_13 succeeded in using a small feedforward neural network to decode the Vigenere cipher. Unfortunately, they phrased the task in a cipher-specific manner that dramatically simplified the learning objective. For example, at each step, they gave the model the single keyphrase character that was necessary to perform decryption. One of the things that makes learning the Vigenere cipher difficult is choosing that keyphrase character for a particular time step. We avoid this pitfall by using an approach that generalizes well across several ciphers. | {
"cite_N": [
"@cite_13"
],
"mid": [
"2144683792"
],
"abstract": [
"The problem in cryptanalysis can be described as an unknown and the neural networks are ideal tools for blackbox system identification. In this paper, a mathematical black-box model is developed and system identification techniques are combined with adaptive system techniques, to construct the Neuro-Identifier. The Neuro-Identifier is discussed as a black-box model to attack the target cipher systems. In this paper a new addition in cryptography. Has been presented, and the methods of classical and stream cryptosystems are discussed. The constructing of Neuro-Identifier mode is to achieve two objectives: the first one is to emulator construction Neuro-model for the target cipher system, while the second is to (cryptanalysis) determine the key from given plaintext-ciphertext pair."
]
} |
1708.07576 | 2746618409 | Recurrent neural networks (RNNs) represent the state of the art in translation, image captioning, and speech recognition. They are also capable of learning algorithmic tasks such as long addition, copying, and sorting from a set of training examples. We demonstrate that RNNs can learn decryption algorithms -- the mappings from plaintext to ciphertext -- for three polyalphabetic ciphers (Vigenere, Autokey, and Enigma). Most notably, we demonstrate that an RNN with a 3000-unit Long Short-Term Memory (LSTM) cell can learn the decryption function of the Enigma machine. We argue that our model learns efficient internal representations of these ciphers 1) by exploring activations of individual memory neurons and 2) by comparing memory usage across the three ciphers. To be clear, our work is not aimed at 'cracking' the Enigma cipher. However, we do show that our model can perform elementary cryptanalysis by running known-plaintext attacks on the Vigenere and Autokey ciphers. Our results indicate that RNNs can learn algorithmic representations of black box polyalphabetic ciphers and that these representations are useful for cryptanalysis. | There are other interesting connections between machine learning and cryptography. Abadi and Andersen trained two convolutional neural networks (CNNs) to communicate with each other while hiding information from from a third. The authors argue that the two CNNs learned to use a simple form of encryption to selectively protect information from the eavesdropper. Another work by @cite_3 embedded images directly into the trainable parameters of a neural network. These messages could be recovered using a second neural network. Like our work, these two works use neural networks to encrypt information. Unlike our work, the models were neither recurrent nor were they trained on existing ciphers. | {
"cite_N": [
"@cite_3"
],
"mid": [
"2950398949"
],
"abstract": [
"Echo state networks are simple recurrent neural networks that are easy to implement and train. Despite their simplicity, they show a form of memory and can predict or regenerate sequences of data. We make use of this property to realize a novel neural cryptography scheme. The key idea is to assume that Alice and Bob share a copy of an echo state network. If Alice trains her copy to memorize a message, she can communicate the trained part of the network to Bob who plugs it into his copy to regenerate the message. Considering a byte-level representation of in- and output, the technique applies to arbitrary types of data (texts, images, audio files, etc.) and practical experiments reveal it to satisfy the fundamental cryptographic properties of diffusion and confusion."
]
} |
1708.07718 | 2745530706 | In this paper we present a differential approach to photo-polarimetric shape estimation. We propose several alternative differential constraints based on polarisation and photometric shading information and show how to express them in a unified partial differential system. Our method uses the image ratios technique to combine shading and polarisation information in order to directly reconstruct surface height, without first computing surface normal vectors. Moreover, we are able to remove the non-linearities so that the problem reduces to solving a linear differential problem. We also introduce a new method for estimating a polarisation image from multichannel data and, finally, we show it is possible to estimate the illumination directions in a two source setup, extending the method into an uncalibrated scenario. From a numerical point of view, we use a least-squares formulation of the discrete version of the problem. To the best of our knowledge, this is the first work to consider a unified differential approach to solve photo-polarimetric shape estimation directly for height. Numerical results on synthetic and real-world data confirm the effectiveness of our proposed method. | The polarisation state of light reflected by a surface provides a cue to the material properties of the surface and, via a relationship with surface orientation, the shape. Polarisation has been used for a number of applications, including early work on material segmentation @cite_1 and diffuse specular reflectance separation @cite_6 . However, there has been a resurgent interest @cite_17 @cite_26 @cite_3 @cite_13 in using polarisation information for shape estimation. | {
"cite_N": [
"@cite_26",
"@cite_1",
"@cite_3",
"@cite_6",
"@cite_13",
"@cite_17"
],
"mid": [
"2200911156",
"",
"2386561516",
"",
"1960300181",
"2520696072"
],
"abstract": [
"Coarse depth maps can be enhanced by using the shape information from polarization cues. We propose a framework to combine surface normals from polarization (hereafter polarization normals) with an aligned depth map. Polarization normals have not been used for depth enhancement before. This is because polarization normals suffer from physics-based artifacts, such as azimuthal ambiguity, refractive distortion and fronto-parallel signal degradation. We propose a framework to overcome these key challenges, allowing the benefits of polarization to be used to enhance depth maps. Our results demonstrate improvement with respect to state-of-the-art 3D reconstruction techniques.",
"",
"Shape from Polarization (SfP) estimates surface normals using photos captured at different polarizer rotations. Fundamentally, the SfP model assumes that light is reflected either diffusely or specularly. However, this model is not valid for many real-world surfaces exhibiting a mixture of diffuse and specular properties. To address this challenge, previous methods have used a sequential solution: first, use an existing algorithm to separate the scene into diffuse and specular components, then apply the appropriate SfP model. In this paper, we propose a new method that jointly uses viewpoint and polarization data to holistically separate diffuse and specular components, recover refractive index, and ultimately recover 3D shape. By involving the physics of polarization in the separation process, we demonstrate competitive results with a benchmark method, while recovering additional information (e.g. refractive index).",
"",
"We introduce a method to recover the shape of a smooth dielectric object from polarization images taken with a light source from different directions. We present two constraints on shading and polarization and use both in a single optimization scheme. This integration is motivated by the fact that photometric stereo and polarization-based methods have complementary abilities. The polarization-based method can give strong cues for the surface orientation and refractive index, which are independent of the light direction. However, it has ambiguities in selecting between two ambiguous choices of the surface orientation, in the relationship between refractive index and zenith angle (observing angle), and limited performance for surface points with small zenith angles, where the polarization effect is weak. In contrast, photometric stereo method with multiple light sources can disambiguate the surface orientation and give a strong relationship between the surface normals and light directions. However, it has limited performance for large zenith angles, refractive index estimation, and faces the ambiguity in case the light direction is unknown. Taking their advantages, our proposed method can recover the surface normals for both small and large zenith angles, the light directions, and the refractive indexes of the object. The proposed method is successfully evaluated by simulation and real-world experiments.",
"We present a method for estimating surface height directly from a single polarisation image simply by solving a large, sparse system of linear equations. To do so, we show how to express polarisation constraints as equations that are linear in the unknown depth. The ambiguity in the surface normal azimuth angle is resolved globally when the optimal surface height is reconstructed. Our method is applicable to objects with uniform albedo exhibiting diffuse and specular reflectance. We extend it to an uncalibrated scenario by demonstrating that the illumination (point source or first second order spherical harmonics) can be estimated from the polarisation image, up to a binary convex concave ambiguity. We believe that our method is the first monocular, passive shape-from-x technique that enables well-posed depth estimation with only a single, uncalibrated illumination condition. We present results on glossy objects, including in uncontrolled, outdoor illumination."
]
} |
1708.07718 | 2745530706 | In this paper we present a differential approach to photo-polarimetric shape estimation. We propose several alternative differential constraints based on polarisation and photometric shading information and show how to express them in a unified partial differential system. Our method uses the image ratios technique to combine shading and polarisation information in order to directly reconstruct surface height, without first computing surface normal vectors. Moreover, we are able to remove the non-linearities so that the problem reduces to solving a linear differential problem. We also introduce a new method for estimating a polarisation image from multichannel data and, finally, we show it is possible to estimate the illumination directions in a two source setup, extending the method into an uncalibrated scenario. From a numerical point of view, we use a least-squares formulation of the discrete version of the problem. To the best of our knowledge, this is the first work to consider a unified differential approach to solve photo-polarimetric shape estimation directly for height. Numerical results on synthetic and real-world data confirm the effectiveness of our proposed method. | Shape-from-polarisation The degree to which light is linearly polarised and the orientation associated with maximum reflection are related to the two degrees of freedom of surface orientation. In theory, this polarisation information alone restricts the surface normal at each pixel to two possible directions. Both Atkinson and Hancock @cite_7 and Miyazaki al @cite_2 solve the problem of disambiguating these polarisation normals via propagation from the boundary under an assumption of global convexity. Huynh al @cite_23 also disambiguate polarisation normals with a global convexity assumption but estimate refractive index in addition. These works all used a diffuse polarisation model while Morel al @cite_20 use a specular polarisation model for met als. Recently, Taamazyan al @cite_3 introduced a mixed specular diffuse polarisation model. All of these methods estimate surface normals that must be integrated into a height map. Moreover, since they rely entirely on the weak shape cue provided by polarisation and do not enforce integrability, the results are extremely sensitive to noise. | {
"cite_N": [
"@cite_7",
"@cite_3",
"@cite_23",
"@cite_2",
"@cite_20"
],
"mid": [
"2165101214",
"2386561516",
"2129277418",
"2127670691",
"1964369825"
],
"abstract": [
"When unpolarized light is reflected from a smooth dielectric surface, it becomes partially polarized. This is due to the orientation of dipoles induced in the reflecting medium and applies to both specular and diffuse reflection. This paper is concerned with exploiting polarization by surface reflection, using images of smooth dielectric objects, to recover surface normals and, hence, height. This paper presents the underlying physics of polarization by reflection, starting with the Fresnel equations. These equations are used to interpret images taken with a linear polarizer and digital camera, revealing the shape of the objects. Experimental results are presented that illustrate that the technique is accurate near object limbs, as the theory predicts, with less precise, but still useful, results elsewhere. A detailed analysis of the accuracy of the technique for a variety of materials is presented. A method for estimating refractive indices using a laser and linear polarizer is also given.",
"Shape from Polarization (SfP) estimates surface normals using photos captured at different polarizer rotations. Fundamentally, the SfP model assumes that light is reflected either diffusely or specularly. However, this model is not valid for many real-world surfaces exhibiting a mixture of diffuse and specular properties. To address this challenge, previous methods have used a sequential solution: first, use an existing algorithm to separate the scene into diffuse and specular components, then apply the appropriate SfP model. In this paper, we propose a new method that jointly uses viewpoint and polarization data to holistically separate diffuse and specular components, recover refractive index, and ultimately recover 3D shape. By involving the physics of polarization in the separation process, we demonstrate competitive results with a benchmark method, while recovering additional information (e.g. refractive index).",
"In this paper, we propose an approach to the problem of simultaneous shape and refractive index recovery from multispectral polarisation imagery captured from a single viewpoint. The focus of this paper is on dielectric surfaces which diffusely polarise light transmitted from the dielectric body into the air. The diffuse polarisation of the reflection process is modelled using a Transmitted Radiance Sinusoid curve and the Fresnel transmission theory. We provide a method of estimating the azimuth angle of surface normals from the spectral variation of the phase of polarisation. Moreover, to render the problem of simultaneous estimation of surface orientation and index of refraction well-posed, we enforce a generative model on the material dispersion equations for the index of refraction. This generative model, together with the Fresnel transmission ratio, permit the recovery of the index of refraction and the zenith angle simultaneously. We show results on shape recovery and rendering for real world and synthetic imagery.",
"This paper presents a method to estimate geometrical, photometrical, and environmental information of a single-viewed object in one integrated framework under fixed viewing position and fixed illumination direction. These three types of information are important to render a photorealistic image of a real object. Photometrical information represents the texture and the surface roughness of an object, while geometrical and environmental information represent the 3D shape of an object and the illumination distribution, respectively. The proposed method estimates the 3D shape by computing the surface normal from polarization data, calculates the texture of the object from the diffuse only reflection component, determines the illumination directions from the position of the brightest intensity in the specular reflection component, and finally computes the surface roughness of the object by using the estimated illumination distribution.",
"We propose a new application of « Shape from Polarization » method to reconstruct surface shapes of specular met allic objects. Studying the polarization state of the reflected light is very useful to get information on the surface normals. After reflection, an unpolarized light wave becomes partially linearly polarized. Such a wave, can be described by its three parameters: intensity, degree of polarization, and angle of polarization. By using the refractive index of the surface, a relationship between the degree of polarization and the reflection angle can be established. Unfortunately, the relation commonly used for dielectrics, cannot be applied since the refractive index of met allic surfaces is complex. To get a similar relation, we apply a usual approximation valid in the visible region. The Fresnel reflectance model can also provide a relationship between the angle of polarization and the incidence plane orientation. Thus, the reflection angle and the incidence plane orientation give the surface normals. The shape is finally computed by integrating the normals with a relaxation algorithm. Applications on met allic objects made by stamping and polishing are also described, and show the efficiency of our system to discriminate shape defects. Future works will consist in integrating the system into an automatic process of defects detection."
]
} |
1708.07718 | 2745530706 | In this paper we present a differential approach to photo-polarimetric shape estimation. We propose several alternative differential constraints based on polarisation and photometric shading information and show how to express them in a unified partial differential system. Our method uses the image ratios technique to combine shading and polarisation information in order to directly reconstruct surface height, without first computing surface normal vectors. Moreover, we are able to remove the non-linearities so that the problem reduces to solving a linear differential problem. We also introduce a new method for estimating a polarisation image from multichannel data and, finally, we show it is possible to estimate the illumination directions in a two source setup, extending the method into an uncalibrated scenario. From a numerical point of view, we use a least-squares formulation of the discrete version of the problem. To the best of our knowledge, this is the first work to consider a unified differential approach to solve photo-polarimetric shape estimation directly for height. Numerical results on synthetic and real-world data confirm the effectiveness of our proposed method. | Polarisation with additional cues Rahmann and Canterakis @cite_11 combined a specular polarisation model with stereo cues. Similarly, Atkinson and Hancock @cite_14 used polarisation normals to segment an object into patches, simplifying stereo matching. Stereo polarisation cues have also been used for transparent surface modelling @cite_15 . Huynh al @cite_10 extended their earlier work to use multispectral measurements to estimate both shape and refractive index. Drbohlav and Sara @cite_22 showed how the Bas-relief ambiguity @cite_9 in uncalibrated photometric stereo could be resolved using polarisation. However, this approach requires a polarised light source. Recently, Kadambi al @cite_26 proposed an interesting approach in which a single polarisation image is combined with a depth map obtained by an RGBD camera. The depth map is used to disambiguate the normals and provide a base surface for integration. | {
"cite_N": [
"@cite_14",
"@cite_26",
"@cite_22",
"@cite_9",
"@cite_15",
"@cite_10",
"@cite_11"
],
"mid": [
"2096244937",
"2200911156",
"2118920825",
"",
"2163039610",
"1986123528",
"2140253355"
],
"abstract": [
"This paper presents a novel method for 3D surface reconstruction that uses polarization and shading information from two views. The method relies on the polarization data acquired using a standard digital camera and a linear polarizer. Fresnel theory is used to process the raw images and to obtain initial estimates of surface normals, assuming that the reflection type is diffuse. Based on this idea, the paper presents two novel contributions to the problem of surface reconstruction. The first is a technique to enhance the surface normal estimates by incorporating shading information into the method. This is done using robust statistics to estimate how the measured pixel brightnesses depend on the surface orientation. This gives an estimate of the object material reflectance function, which is used to refine the estimates of the surface normals. The second contribution is to use the refined estimates to establish correspondence between two views of an object. To do this, surface patches are extracted from each view, which are then aligned by minimising an energy functional based on the surface normal estimates and local topographic properties. The optimum alignment parameters for different patch pairs are then used to establish stereo correspondence. This process results in an unambiguous field of surface normals, which can be integrated to recover the surface depth. Our technique is most suited to smooth nonmet allic surfaces. It complements existing stereo algorithms since it does not require salient surface features to obtain correspondences. An extensive set of experiments, yielding reconstructed objects and reflectance functions, are presented and compared to ground truth.",
"Coarse depth maps can be enhanced by using the shape information from polarization cues. We propose a framework to combine surface normals from polarization (hereafter polarization normals) with an aligned depth map. Polarization normals have not been used for depth enhancement before. This is because polarization normals suffer from physics-based artifacts, such as azimuthal ambiguity, refractive distortion and fronto-parallel signal degradation. We propose a framework to overcome these key challenges, allowing the benefits of polarization to be used to enhance depth maps. Our results demonstrate improvement with respect to state-of-the-art 3D reconstruction techniques.",
"Photometric stereo with uncalibrated lights determines surface orientations ambiguously up to any regular transformation. If the surface reflectance model is separable with respect to the illumination and viewing directions, its inherent symmetries enable to design two previously unrecognized constraints on normals that reduce this ambiguity. The two constraints represent projections of normals onto planes perpendicular to the viewing and illumination directions, respectively. We identify the classes of transformations that leave each constraint invariant. We construct the constraints using polarization measurement under the assumption of separable reflectance model for smooth dielectrics. We verify that applying the first constraint together with the integrability constraint results in bas-relief ambiguity, while application of the second constraint on integrable normals reduces the ambiguity to convex concave ambiguity. Importantly, the latter result is also obtained when the first and second constraints alone are combined.",
"",
"We propose a method for measuring surface shapes of transparent objects by using a polarizing filter. Generally, the light reflected from an object is partially polarized. The degree of polarization depends upon the incident angle, which, in turn, depends upon the surface normal. Therefore, we can obtain surface normals of objects by observing the degree of polarization at each surface point. Unfortunately, the correspondence between the degree of polarization and the surface normal is not one to one. Hence, to obtain the correct surface normal, we have to solve the ambiguity problem. In this paper, we introduce a method to solve the ambiguity by comparing the polarization data in two objects, i.e., normal position and tilted with small angle position. We also discuss the geometrical features of the object surface and propose a method for matching two sets of polarization data at identical points on the object surface.",
"In this paper, we address the problem of the simultaneous recovery of the shape and refractive index of an object from a spectro-polarimetric image captured from a single view. Here, we focus on the diffuse polarisation process occuring at dielectric surfaces due to subsurface scattering and transmission from the object surface into the air. The diffuse polarisation of the reflection process is modelled by the Fresnel transmission theory. We present a method for estimating the azimuth angle of surface normals from the spectral variation of the phase of polarisation. Moreover, we estimate the zenith angle of surface normals and index of refraction simultaneously in a well-posed optimisation framework. We achieve well-posedness by introducing two additional constraints to the problem, including the surface integrability and the material dispersion equation. This yields an iterative solution which is computationally efficient due to the use of closed-form solutions for both the zenith angle and the refractive index in each iteration. To demonstrate the effectiveness of our approach, we show results of shape recovery and surface rendering for both real-world and synthetic imagery.",
"Traditional intensity imaging does not offer a general approach for the perception of textureless and specular reflecting surfaces. Intensity based methods for shape reconstruction of specular surfaces rely on virtual (i.e. mirrored) features moving over the surface under viewer motion. We present a novel method based on polarization imaging for shape recovery of specular surfaces. This method overcomes the limitations of the intensity based approach because no virtual features are required. It recovers whole surface patches and not only single curves on the surface. The presented solution is general as it is independent of the illumination. The polarization image encodes the projection of the surface normals onto the image and therefore provides constraints on the surface geometry. Taking polarization images from multiple views produces enough constraints to infer the complete surface shape. The reconstruction problem is solved by an optimization scheme where the surface geometry is modelled by a set of hierarchical basis functions. The optimization algorithm proves to be well converging, accurate and noise resistant. The work is substantiated by experiments on synthetic and real data."
]
} |
1708.07403 | 2748159032 | We present CloudScan; an invoice analysis system that requires zero configuration or upfront annotation. In contrast to previous work, CloudScan does not rely on templates of invoice layout, instead it learns a single global model of invoices that naturally generalizes to unseen invoice layouts. The model is trained using data automatically extracted from end-user provided feedback. This automatic training data extraction removes the requirement for users to annotate the data precisely. We describe a recurrent neural network model that can capture long range context and compare it to a baseline logistic regression model corresponding to the current CloudScan production system. We train and evaluate the system on 8 important fields using a dataset of 326,471 invoices. The recurrent neural network and baseline model achieve 0.891 and 0.887 average F1 scores respectively on seen invoice layouts. For the harder task of unseen invoice layouts, the recurrent neural network model outperforms the baseline with 0.840 average F1 compared to 0.788. | The most directly related works are Intellix @cite_15 by DocuWare and the work by ITESOFT @cite_4 . Both systems require that relevant fields are annotated for a template manually beforehand, which creates a database of templates, fields and automatically extracted keywords and positions for each field. When new documents are received, both systems classify the template automatically using address lookups or machine learning classifiers. Once the template is classified the keywords and positions for each field are used to propose field candidates which are then scored using heuristics such as proximity and uniqueness of the keywords. Having scored the candidates the best one for each field is chosen. | {
"cite_N": [
"@cite_15",
"@cite_4"
],
"mid": [
"2078777599",
"2087291337"
],
"abstract": [
"Automatic information extraction from scanned business documents is especially valuable in the application domain of document archiving. But current systems for automated document processing still require a lot of configuration work that can only be done by experienced users or administrators. We present an approach for information extraction which purely builds on end-user provided training examples and intentionally omits efficient known extraction techniques like rule based extraction that require intense training and or information extraction expertise. Our evaluation on a large corpus of business documents shows competitive results of above 85 F1-measure on 10 commonly used fields like document type, sender, receiver and date. The system is deployed and used inside the commercial document management system DocuWare.",
"In this paper we present an incremental framework aimed at extracting field information from administrative document images in the context of a Digital Mail-room scenario. Given a single training sample in which the user has marked which fields have to be extracted from a particular document class, a document model representing structural relationships among words is built. This model is incrementally refined as the system processes more and more documents from the same class. A reformulation of the tf-idf statistic scheme allows to adjust the importance weights of the structural relationships among words. We report in the experimental section our results obtained with a large dataset of real invoices."
]
} |
1708.07403 | 2748159032 | We present CloudScan; an invoice analysis system that requires zero configuration or upfront annotation. In contrast to previous work, CloudScan does not rely on templates of invoice layout, instead it learns a single global model of invoices that naturally generalizes to unseen invoice layouts. The model is trained using data automatically extracted from end-user provided feedback. This automatic training data extraction removes the requirement for users to annotate the data precisely. We describe a recurrent neural network model that can capture long range context and compare it to a baseline logistic regression model corresponding to the current CloudScan production system. We train and evaluate the system on 8 important fields using a dataset of 326,471 invoices. The recurrent neural network and baseline model achieve 0.891 and 0.887 average F1 scores respectively on seen invoice layouts. For the harder task of unseen invoice layouts, the recurrent neural network model outperforms the baseline with 0.840 average F1 compared to 0.788. | smartFIX @cite_5 uses manually configured rules for each template. @cite_16 learns a database of keywords for each template and fall back to a global database of keywords. @cite_12 uses a database of absolute positions of fields for each template. @cite_14 uses a database of manually created (field, pattern, parser) triplets for each template, designs a probabilistic model for finding the most similar pattern in a template, and extracts the value with the associated parser. | {
"cite_N": [
"@cite_5",
"@cite_14",
"@cite_16",
"@cite_12"
],
"mid": [
"1791544012",
"2058053171",
"",
"1963728304"
],
"abstract": [
"Although the internet offers a wide-spread platform for information interchange, day-to-day work in large companies still means the processing of tens of thousands of printed documents every day. This paper presents the system smartFIX which is a document analysis and understanding system developed by the DFKI spin-off INSIDERS. It permits the processing of documents ranging from fixed format forms to unstructured letters of any format. Apart from the architecture, the main components and system characteristics, we also show some results when applying smartFIX to medical bills and prescriptions.",
"We propose an approach for information extraction for multi-page printed document understanding. The approach is designed for scenarios in which the set of possible document classes, i.e., documents sharing similar content and layout, is large and may evolve over time. Describing a new class is a very simple task: the operator merely provides a few samples and then, by means of a GUI, clicks on the OCR-generated blocks of a document containing the information to be extracted. Our approach is based on probability: we derived a general form for the probability that a sequence of blocks contains the searched information. We estimate the parameters for a new class by applying the maximum likelihood method to the samples of the class. All these parameters depend only on block properties that can be extracted automatically from the operator actions on the GUI. Processing a document of a given class consists in finding the sequence of blocks, which maximizes the corresponding probability for that class. We evaluated experimentally our proposal using 807 multi-page printed documents of different domains (invoices, patents, data-sheets), obtaining very good results—e.g., a success rate often greater than 90 even for classes with just two samples.",
"",
"Archiving official written documents such as invoices, reminders and account statements in business and private area gets more and more important. Creating appropriate index entries for document archives like sender's name, creation date or document number is a tedious manual work. We present a novel approach to handle automatic indexing of documents based on generic positional extraction of index terms. For this purpose we apply the knowledge of document templates stored in a common full text search index to find index positions that were successfully extracted in the past."
]
} |
1708.07403 | 2748159032 | We present CloudScan; an invoice analysis system that requires zero configuration or upfront annotation. In contrast to previous work, CloudScan does not rely on templates of invoice layout, instead it learns a single global model of invoices that naturally generalizes to unseen invoice layouts. The model is trained using data automatically extracted from end-user provided feedback. This automatic training data extraction removes the requirement for users to annotate the data precisely. We describe a recurrent neural network model that can capture long range context and compare it to a baseline logistic regression model corresponding to the current CloudScan production system. We train and evaluate the system on 8 important fields using a dataset of 326,471 invoices. The recurrent neural network and baseline model achieve 0.891 and 0.887 average F1 scores respectively on seen invoice layouts. For the harder task of unseen invoice layouts, the recurrent neural network model outperforms the baseline with 0.840 average F1 compared to 0.788. | Our automatic training data extraction is closely related to the idea of distant supervision @cite_19 where relations are extracted from unstructured text automatically using heuristics. | {
"cite_N": [
"@cite_19"
],
"mid": [
"2107598941"
],
"abstract": [
"Modern models of relation extraction for tasks like ACE are based on supervised learning of relations from small hand-labeled corpora. We investigate an alternative paradigm that does not require labeled corpora, avoiding the domain dependence of ACE-style algorithms, and allowing the use of corpora of any size. Our experiments use Freebase, a large semantic database of several thousand relations, to provide distant supervision. For each pair of entities that appears in some Freebase relation, we find all sentences containing those entities in a large unlabeled corpus and extract textual features to train a relation classifier. Our algorithm combines the advantages of supervised IE (combining 400,000 noisy pattern features in a probabilistic classifier) and unsupervised IE (extracting large numbers of relations from large corpora of any domain). Our model is able to extract 10,000 instances of 102 relations at a precision of 67.6 . We also analyze feature performance, showing that syntactic parse features are particularly helpful for relations that are ambiguous or lexically distant in their expression."
]
} |
1708.07403 | 2748159032 | We present CloudScan; an invoice analysis system that requires zero configuration or upfront annotation. In contrast to previous work, CloudScan does not rely on templates of invoice layout, instead it learns a single global model of invoices that naturally generalizes to unseen invoice layouts. The model is trained using data automatically extracted from end-user provided feedback. This automatic training data extraction removes the requirement for users to annotate the data precisely. We describe a recurrent neural network model that can capture long range context and compare it to a baseline logistic regression model corresponding to the current CloudScan production system. We train and evaluate the system on 8 important fields using a dataset of 326,471 invoices. The recurrent neural network and baseline model achieve 0.891 and 0.887 average F1 scores respectively on seen invoice layouts. For the harder task of unseen invoice layouts, the recurrent neural network model outperforms the baseline with 0.840 average F1 compared to 0.788. | Slot Filling is another related NLP task in which pre-defined slots must be filled from natural text. Our system can be seen as a slot filling task with 8 slots, and the text of a single invoice as input. Neural architectures are also used here, e.g. @cite_20 uses bi-directional RNNs and word embedding to achieve competitive results on the ATIS (Airline Travel Information Systems) benchmark dataset. | {
"cite_N": [
"@cite_20"
],
"mid": [
"2399456070"
],
"abstract": [
"One of the key problems in spoken language understanding (SLU) is the task of slot filling. In light of the recent success of applying deep neural network technologies in domain detection and intent identification, we carried out an in-depth investigation on the use of recurrent neural networks for the more difficult task of slot filling involving sequence discrimination. In this work, we implemented and compared several important recurrent-neural-network architectures, including the Elman-type and Jordan-type recurrent networks and their variants. To make the results easy to reproduce and compare, we implemented these networks on the common Theano neural network toolkit, and evaluated them on the ATIS benchmark. We also compared our results to a conditional random fields (CRF) baseline. Our results show that on this task, both types of recurrent networks outperform the CRF baseline substantially, and a bi-directional Jordantype network that takes into account both past and future dependencies among slots works best, outperforming a CRFbased baseline by 14 in relative error reduction."
]
} |
1708.07632 | 2748434587 | Convolutional neural networks with spatio-temporal 3D kernels (3D CNNs) have an ability to directly extract spatio-temporal features from videos for action recognition. Although the 3D kernels tend to overfit because of a large number of their parameters, the 3D CNNs are greatly improved by using recent huge video databases. However, the architecture of 3D CNNs is relatively shallow against to the success of very deep neural networks in 2D-based CNNs, such as residual networks (ResNets). In this paper, we propose a 3D CNNs based on ResNets toward a better action representation. We describe the training procedure of our 3D ResNets in details. We experimentally evaluate the 3D ResNets on the ActivityNet and Kinetics datasets. The 3D ResNets trained on the Kinetics did not suffer from overfitting despite the large number of parameters of the model, and achieved better performance than relatively shallow networks, such as C3D. Our code and pretrained models (e.g. Kinetics and ActivityNet) are publicly available at this https URL | The HMDB51 @cite_20 and UCF101 @cite_16 are the most successful databases in action recognition. The recent consensus, however, tells that these two databases are not large-scale databases. It is difficult to train good models without overfitting using these databases. More recently, huge databases such as Sports-1M @cite_18 and YouTube-8M @cite_21 are proposed. These databases are big enough whereas their annotations are noisy and only video-level labels (i.e. the frames that do not relate to target activities are included). Such noise and unrelated frames might prevent models from good training. In order to create a successful pretrained model like 2D CNNs trained on ImageNet @cite_0 , the Google DeepMind released the Kinetics human action video dataset @cite_10 . The Kinetics dataset includes 300,000 or over trimmed videos and 400 categories. The size of Kinetics is smaller than Sports-1M and YouTube-8M whereas the quality of annotation is extremely high. | {
"cite_N": [
"@cite_18",
"@cite_21",
"@cite_0",
"@cite_16",
"@cite_10",
"@cite_20"
],
"mid": [
"2308045930",
"",
"2108598243",
"24089286",
"2619947201",
"2126579184"
],
"abstract": [
"",
"",
"The explosion of image data on the Internet has the potential to foster more sophisticated and robust models and algorithms to index, retrieve, organize and interact with images and multimedia data. But exactly how such data can be harnessed and organized remains a critical problem. We introduce here a new database called “ImageNet”, a large-scale ontology of images built upon the backbone of the WordNet structure. ImageNet aims to populate the majority of the 80,000 synsets of WordNet with an average of 500-1000 clean and full resolution images. This will result in tens of millions of annotated images organized by the semantic hierarchy of WordNet. This paper offers a detailed analysis of ImageNet in its current state: 12 subtrees with 5247 synsets and 3.2 million images in total. We show that ImageNet is much larger in scale and diversity and much more accurate than the current image datasets. Constructing such a large-scale database is a challenging task. We describe the data collection scheme with Amazon Mechanical Turk. Lastly, we illustrate the usefulness of ImageNet through three simple applications in object recognition, image classification and automatic object clustering. We hope that the scale, accuracy, diversity and hierarchical structure of ImageNet can offer unparalleled opportunities to researchers in the computer vision community and beyond.",
"We introduce UCF101 which is currently the largest dataset of human actions. It consists of 101 action classes, over 13k clips and 27 hours of video data. The database consists of realistic user uploaded videos containing camera motion and cluttered background. Additionally, we provide baseline action recognition results on this new dataset using standard bag of words approach with overall performance of 44.5 . To the best of our knowledge, UCF101 is currently the most challenging dataset of actions due to its large number of classes, large number of clips and also unconstrained nature of such clips.",
"We describe the DeepMind Kinetics human action video dataset. The dataset contains 400 human action classes, with at least 400 video clips for each action. Each clip lasts around 10s and is taken from a different YouTube video. The actions are human focussed and cover a broad range of classes including human-object interactions such as playing instruments, as well as human-human interactions such as shaking hands. We describe the statistics of the dataset, how it was collected, and give some baseline performance figures for neural network architectures trained and tested for human action classification on this dataset. We also carry out a preliminary analysis of whether imbalance in the dataset leads to bias in the classifiers.",
"With nearly one billion online videos viewed everyday, an emerging new frontier in computer vision research is recognition and search in video. While much effort has been devoted to the collection and annotation of large scalable static image datasets containing thousands of image categories, human action datasets lag far behind. Current action recognition databases contain on the order of ten different action categories collected under fairly controlled conditions. State-of-the-art performance on these datasets is now near ceiling and thus there is a need for the design and creation of new benchmarks. To address this issue we collected the largest action video database to-date with 51 action categories, which in total contain around 7,000 manually annotated clips extracted from a variety of sources ranging from digitized movies to YouTube. We use this database to evaluate the performance of two representative computer vision systems for action recognition and explore the robustness of these methods under various conditions such as camera motion, viewpoint, video quality and occlusion."
]
} |
1708.07632 | 2748434587 | Convolutional neural networks with spatio-temporal 3D kernels (3D CNNs) have an ability to directly extract spatio-temporal features from videos for action recognition. Although the 3D kernels tend to overfit because of a large number of their parameters, the 3D CNNs are greatly improved by using recent huge video databases. However, the architecture of 3D CNNs is relatively shallow against to the success of very deep neural networks in 2D-based CNNs, such as residual networks (ResNets). In this paper, we propose a 3D CNNs based on ResNets toward a better action representation. We describe the training procedure of our 3D ResNets in details. We experimentally evaluate the 3D ResNets on the ActivityNet and Kinetics datasets. The 3D ResNets trained on the Kinetics did not suffer from overfitting despite the large number of parameters of the model, and achieved better performance than relatively shallow networks, such as C3D. Our code and pretrained models (e.g. Kinetics and ActivityNet) are publicly available at this https URL | One of the popular approach of CNN-based action recognition is two-stream CNNs with 2D convolutional kernels. proposed the method that uses RGB and stacked optical flow frames as appearance and motion information, respectively @cite_3 . They showed combining the two-streams improves action recognition accuracy. Many methods based on the two-stream CNNs are proposed to improve action recognition performance @cite_5 @cite_7 @cite_14 @cite_4 . proposed combining two-stream CNNs with ResNets @cite_7 . They showed the architecture of ResNets is effective for action recognition with 2D CNNs. Different from the above mentioned approaches, we focused on 3D CNNs, which recently outperform the 2D CNNs using large-scale video datasets. | {
"cite_N": [
"@cite_14",
"@cite_4",
"@cite_7",
"@cite_3",
"@cite_5"
],
"mid": [
"1944615693",
"2507009361",
"2342662179",
"2952186347",
"2952005526"
],
"abstract": [
"Visual features are of vital importance for human action understanding in videos. This paper presents a new video representation, called trajectory-pooled deep-convolutional descriptor (TDD), which shares the merits of both hand-crafted features [31] and deep-learned features [24]. Specifically, we utilize deep architectures to learn discriminative convolutional feature maps, and conduct trajectory-constrained pooling to aggregate these convolutional features into effective descriptors. To enhance the robustness of TDDs, we design two normalization methods to transform convolutional feature maps, namely spatiotemporal normalization and channel normalization. The advantages of our features come from (i) TDDs are automatically learned and contain high discriminative capacity compared with those hand-crafted features; (ii) TDDs take account of the intrinsic characteristics of temporal dimension and introduce the strategies of trajectory-constrained sampling and pooling for aggregating deep-learned features. We conduct experiments on two challenging datasets: HMD-B51 and UCF101. Experimental results show that TDDs outperform previous hand-crafted features [31] and deep-learned features [24]. Our method also achieves superior performance to the state of the art on these datasets.",
"Deep convolutional networks have achieved great success for visual recognition in still images. However, for action recognition in videos, the advantage over traditional methods is not so evident. This paper aims to discover the principles to design effective ConvNet architectures for action recognition in videos and learn these models given limited training samples. Our first contribution is temporal segment network (TSN), a novel framework for video-based action recognition. which is based on the idea of long-range temporal structure modeling. It combines a sparse temporal sampling strategy and video-level supervision to enable efficient and effective learning using the whole action video. The other contribution is our study on a series of good practices in learning ConvNets on video data with the help of temporal segment network. Our approach obtains the state-the-of-art performance on the datasets of HMDB51 ( ( 69.4 , )) and UCF101 ( ( 94.2 , )). We also visualize the learned ConvNet models, which qualitatively demonstrates the effectiveness of temporal segment network and the proposed good practices (Models and code at https: github.com yjxiong temporal-segment-networks).",
"Recent applications of Convolutional Neural Networks (ConvNets) for human action recognition in videos have proposed different solutions for incorporating the appearance and motion information. We study a number of ways of fusing ConvNet towers both spatially and temporally in order to best take advantage of this spatio-temporal information. We make the following findings: (i) that rather than fusing at the softmax layer, a spatial and temporal network can be fused at a convolution layer without loss of performance, but with a substantial saving in parameters, (ii) that it is better to fuse such networks spatially at the last convolutional layer than earlier, and that additionally fusing at the class prediction layer can boost accuracy, finally (iii) that pooling of abstract convolutional features over spatiotemporal neighbourhoods further boosts performance. Based on these studies we propose a new ConvNet architecture for spatiotemporal fusion of video snippets, and evaluate its performance on standard benchmarks where this architecture achieves state-of-the-art results.",
"We investigate architectures of discriminatively trained deep Convolutional Networks (ConvNets) for action recognition in video. The challenge is to capture the complementary information on appearance from still frames and motion between frames. We also aim to generalise the best performing hand-crafted features within a data-driven learning framework. Our contribution is three-fold. First, we propose a two-stream ConvNet architecture which incorporates spatial and temporal networks. Second, we demonstrate that a ConvNet trained on multi-frame dense optical flow is able to achieve very good performance in spite of limited training data. Finally, we show that multi-task learning, applied to two different action classification datasets, can be used to increase the amount of training data and improve the performance on both. Our architecture is trained and evaluated on the standard video actions benchmarks of UCF-101 and HMDB-51, where it is competitive with the state of the art. It also exceeds by a large margin previous attempts to use deep nets for video classification.",
"Two-stream Convolutional Networks (ConvNets) have shown strong performance for human action recognition in videos. Recently, Residual Networks (ResNets) have arisen as a new technique to train extremely deep architectures. In this paper, we introduce spatiotemporal ResNets as a combination of these two approaches. Our novel architecture generalizes ResNets for the spatiotemporal domain by introducing residual connections in two ways. First, we inject residual connections between the appearance and motion pathways of a two-stream architecture to allow spatiotemporal interaction between the two streams. Second, we transform pretrained image ConvNets into spatiotemporal networks by equipping these with learnable convolutional filters that are initialized as temporal residual connections and operate on adjacent feature maps in time. This approach slowly increases the spatiotemporal receptive field as the depth of the model increases and naturally integrates image ConvNet design principles. The whole model is trained end-to-end to allow hierarchical learning of complex spatiotemporal features. We evaluate our novel spatiotemporal ResNet using two widely used action recognition benchmarks where it exceeds the previous state-of-the-art."
]
} |
1708.07517 | 2748489050 | We show how a simple convolutional neural network (CNN) can be trained to accurately and robustly regress 6 degrees of freedom (6DoF) 3D head pose, directly from image intensities. We further explain how this FacePoseNet (FPN) can be used to align faces in 2D and 3D as an alternative to explicit facial landmark detection for these tasks. We claim that in many cases the standard means of measuring landmark detector accuracy can be misleading when comparing different face alignments. Instead, we compare our FPN with existing methods by evaluating how they affect face recognition accuracy on the IJB-A and IJB-B benchmarks: using the same recognition pipeline, but varying the face alignment method. Our results show that (a) better landmark detection accuracy measured on the 300W benchmark does not necessarily imply better face recognition accuracy. (b) Our FPN provides superior 2D and 3D face alignment on both benchmarks. Finally, (c), FPN aligns faces at a small fraction of the computational cost of comparably accurate landmark detectors. For many purposes, FPN is thus a far faster and far more accurate face alignment method than using facial landmark detectors. | Applications of facial landmark detectors. Facial landmark detection is big business, as reflected by the numerous citation to relevant papers, the many facial landmark detection benchmarks @cite_25 @cite_18 @cite_10 @cite_34 @cite_45 , and popular international events dedicated to this problem. With all this effort, a rigorous survey of the many applications of facial landmarks is outside the scope of this paper. In lieu of such a survey, and to get some idea of why this problem attracts so much attention, we offer the following cursory study. | {
"cite_N": [
"@cite_18",
"@cite_45",
"@cite_34",
"@cite_10",
"@cite_25"
],
"mid": [
"",
"2047508432",
"2284800790",
"1796263212",
"1963599662"
],
"abstract": [
"",
"We present a unified model for face detection, pose estimation, and landmark estimation in real-world, cluttered images. Our model is based on a mixtures of trees with a shared pool of parts; we model every facial landmark as a part and use global mixtures to capture topological changes due to viewpoint. We show that tree-structured models are surprisingly effective at capturing global elastic deformation, while being easy to optimize unlike dense graph structures. We present extensive results on standard face benchmarks, as well as a new “in the wild” annotated dataset, that suggests our system advances the state-of-the-art, sometimes considerably, for all three tasks. Though our model is modestly trained with hundreds of faces, it compares favorably to commercial systems trained with billions of examples (such as Google Picasa and face.com).",
"Computer Vision has recently witnessed great research advance towards automatic facial points detection. Numerous methodologies have been proposed during the last few years that achieve accurate and efficient performance. However, fair comparison between these methodologies is infeasible mainly due to two issues. (a) Most existing databases, captured under both constrained and unconstrained (in-the-wild) conditions have been annotated using different mark-ups and, in most cases, the accuracy of the annotations is low. (b) Most published works report experimental results using different training testing sets, different error metrics and, of course, landmark points with semantically different locations. In this paper, we aim to overcome the aforementioned problems by (a) proposing a semi-automatic annotation technique that was employed to re-annotate most existing facial databases under a unified protocol, and (b) presenting the 300 Faces In-The-Wild Challenge (300-W), the first facial landmark localization challenge that was organized twice, in 2013 and 2015. To the best of our knowledge, this is the first effort towards a unified annotation scheme of massive databases and a fair experimental comparison of existing facial landmark localization systems. The images and annotations of the new testing database that was used in the 300-W challenge are available from http: ibug.doc.ic.ac.uk resources 300-W_IMAVIS .",
"We address the problem of interactive facial feature localization from a single image. Our goal is to obtain an accurate segmentation of facial features on high-resolution images under a variety of pose, expression, and lighting conditions. Although there has been significant work in facial feature localization, we are addressing a new application area, namely to facilitate intelligent high-quality editing of portraits, that brings requirements not met by existing methods. We propose an improvement to the Active Shape Model that allows for greater independence among the facial components and improves on the appearance fitting step by introducing a Viterbi optimization process that operates along the facial contours. Despite the improvements, we do not expect perfect results in all cases. We therefore introduce an interaction model whereby a user can efficiently guide the algorithm towards a precise solution. We introduce the Helen Facial Feature Dataset consisting of annotated portrait images gathered from Flickr that are more diverse and challenging than currently existing datasets. We present experiments that compare our automatic method to published results, and also a quantitative evaluation of the effectiveness of our interactive method.",
"We present a novel approach to localizing parts in images of human faces. The approach combines the output of local detectors with a nonparametric set of global models for the part locations based on over 1,000 hand-labeled exemplar images. By assuming that the global models generate the part locations as hidden variables, we derive a Bayesian objective function. This function is optimized using a consensus of models for these hidden variables. The resulting localizer handles a much wider range of expression, pose, lighting, and occlusion than prior ones. We show excellent performance on real-world face datasets such as Labeled Faces in the Wild (LFW) and a new Labeled Face Parts in the Wild (LFPW) and show that our localizer achieves state-of-the-art performance on the less challenging BioID dataset."
]
} |
1708.07517 | 2748489050 | We show how a simple convolutional neural network (CNN) can be trained to accurately and robustly regress 6 degrees of freedom (6DoF) 3D head pose, directly from image intensities. We further explain how this FacePoseNet (FPN) can be used to align faces in 2D and 3D as an alternative to explicit facial landmark detection for these tasks. We claim that in many cases the standard means of measuring landmark detector accuracy can be misleading when comparing different face alignments. Instead, we compare our FPN with existing methods by evaluating how they affect face recognition accuracy on the IJB-A and IJB-B benchmarks: using the same recognition pipeline, but varying the face alignment method. Our results show that (a) better landmark detection accuracy measured on the 300W benchmark does not necessarily imply better face recognition accuracy. (b) Our FPN provides superior 2D and 3D face alignment on both benchmarks. Finally, (c), FPN aligns faces at a small fraction of the computational cost of comparably accurate landmark detectors. For many purposes, FPN is thus a far faster and far more accurate face alignment method than using facial landmark detectors. | We consider two of the most widely cited face landmark detector papers of the last decade, the tree based approach @cite_45 and supervised descent method @cite_38 . At the time of writing, based on Google Scholar, the latter accumulated nearly a thousand citations and the former well over a thousand. We found 23 application names appearing frequently (more than ten times) in the titles of the papers that cite these two and counted the number of times these applications were mentioned. The relative frequencies of these applications are reported in Fig. . | {
"cite_N": [
"@cite_38",
"@cite_45"
],
"mid": [
"2157285372",
"2047508432"
],
"abstract": [
"Many computer vision problems (e.g., camera calibration, image alignment, structure from motion) are solved through a nonlinear optimization method. It is generally accepted that 2nd order descent methods are the most robust, fast and reliable approaches for nonlinear optimization of a general smooth function. However, in the context of computer vision, 2nd order descent methods have two main drawbacks: (1) The function might not be analytically differentiable and numerical approximations are impractical. (2) The Hessian might be large and not positive definite. To address these issues, this paper proposes a Supervised Descent Method (SDM) for minimizing a Non-linear Least Squares (NLS) function. During training, the SDM learns a sequence of descent directions that minimizes the mean of NLS functions sampled at different points. In testing, SDM minimizes the NLS objective using the learned descent directions without computing the Jacobian nor the Hessian. We illustrate the benefits of our approach in synthetic and real examples, and show how SDM achieves state-of-the-art performance in the problem of facial feature detection. The code is available at www.humansensing.cs. cmu.edu intraface.",
"We present a unified model for face detection, pose estimation, and landmark estimation in real-world, cluttered images. Our model is based on a mixtures of trees with a shared pool of parts; we model every facial landmark as a part and use global mixtures to capture topological changes due to viewpoint. We show that tree-structured models are surprisingly effective at capturing global elastic deformation, while being easy to optimize unlike dense graph structures. We present extensive results on standard face benchmarks, as well as a new “in the wild” annotated dataset, that suggests our system advances the state-of-the-art, sometimes considerably, for all three tasks. Though our model is modestly trained with hundreds of faces, it compares favorably to commercial systems trained with billions of examples (such as Google Picasa and face.com)."
]
} |
1708.07517 | 2748489050 | We show how a simple convolutional neural network (CNN) can be trained to accurately and robustly regress 6 degrees of freedom (6DoF) 3D head pose, directly from image intensities. We further explain how this FacePoseNet (FPN) can be used to align faces in 2D and 3D as an alternative to explicit facial landmark detection for these tasks. We claim that in many cases the standard means of measuring landmark detector accuracy can be misleading when comparing different face alignments. Instead, we compare our FPN with existing methods by evaluating how they affect face recognition accuracy on the IJB-A and IJB-B benchmarks: using the same recognition pipeline, but varying the face alignment method. Our results show that (a) better landmark detection accuracy measured on the 300W benchmark does not necessarily imply better face recognition accuracy. (b) Our FPN provides superior 2D and 3D face alignment on both benchmarks. Finally, (c), FPN aligns faces at a small fraction of the computational cost of comparably accurate landmark detectors. For many purposes, FPN is thus a far faster and far more accurate face alignment method than using facial landmark detectors. | Of course, this simple survey is by no means accurate: the same term is counted twice if the paper using it in its title cites both @cite_45 and @cite_38 and many paper titles do not clearly state the application they describe (e.g., @cite_15 describes a method for face alignment in 3D but does not mention alignment'' in the title). Nevertheless, with around two thousand papers included in this survey, the result is quite clear: Alignment, face recognition and pose estimation -- also considered alignment -- are overwhelmingly more popular than any other application. This, of course, excluding other landmark detection papers. | {
"cite_N": [
"@cite_38",
"@cite_15",
"@cite_45"
],
"mid": [
"2157285372",
"2136863438",
"2047508432"
],
"abstract": [
"Many computer vision problems (e.g., camera calibration, image alignment, structure from motion) are solved through a nonlinear optimization method. It is generally accepted that 2nd order descent methods are the most robust, fast and reliable approaches for nonlinear optimization of a general smooth function. However, in the context of computer vision, 2nd order descent methods have two main drawbacks: (1) The function might not be analytically differentiable and numerical approximations are impractical. (2) The Hessian might be large and not positive definite. To address these issues, this paper proposes a Supervised Descent Method (SDM) for minimizing a Non-linear Least Squares (NLS) function. During training, the SDM learns a sequence of descent directions that minimizes the mean of NLS functions sampled at different points. In testing, SDM minimizes the NLS objective using the learned descent directions without computing the Jacobian nor the Hessian. We illustrate the benefits of our approach in synthetic and real examples, and show how SDM achieves state-of-the-art performance in the problem of facial feature detection. The code is available at www.humansensing.cs. cmu.edu intraface.",
"We present a data-driven method for estimating the 3D shapes of faces viewed in single, unconstrained photos (aka \"in-the-wild\"). Our method was designed with an emphasis on robustness and efficiency - with the explicit goal of deployment in real-world applications which reconstruct and display faces in 3D. Our key observation is that for many practical applications, warping the shape of a reference face to match the appearance of a query, is enough to produce realistic impressions of the query's 3D shape. Doing so, however, requires matching visual features between the (possibly very different) query and reference images, while ensuring that a plausible face shape is produced. To this end, we describe an optimization process which seeks to maximize the similarity of appearances and depths, jointly, to those of a reference model. We describe our system for monocular face shape reconstruction and present both qualitative and quantitative experiments, comparing our method against alternative systems, and demonstrating its capabilities. Finally, as a testament to its suitability for real-world applications, we offer an open, on-line implementation of our system, providing unique means of instant 3D viewing of faces appearing in web photos.",
"We present a unified model for face detection, pose estimation, and landmark estimation in real-world, cluttered images. Our model is based on a mixtures of trees with a shared pool of parts; we model every facial landmark as a part and use global mixtures to capture topological changes due to viewpoint. We show that tree-structured models are surprisingly effective at capturing global elastic deformation, while being easy to optimize unlike dense graph structures. We present extensive results on standard face benchmarks, as well as a new “in the wild” annotated dataset, that suggests our system advances the state-of-the-art, sometimes considerably, for all three tasks. Though our model is modestly trained with hundreds of faces, it compares favorably to commercial systems trained with billions of examples (such as Google Picasa and face.com)."
]
} |
1708.07517 | 2748489050 | We show how a simple convolutional neural network (CNN) can be trained to accurately and robustly regress 6 degrees of freedom (6DoF) 3D head pose, directly from image intensities. We further explain how this FacePoseNet (FPN) can be used to align faces in 2D and 3D as an alternative to explicit facial landmark detection for these tasks. We claim that in many cases the standard means of measuring landmark detector accuracy can be misleading when comparing different face alignments. Instead, we compare our FPN with existing methods by evaluating how they affect face recognition accuracy on the IJB-A and IJB-B benchmarks: using the same recognition pipeline, but varying the face alignment method. Our results show that (a) better landmark detection accuracy measured on the 300W benchmark does not necessarily imply better face recognition accuracy. (b) Our FPN provides superior 2D and 3D face alignment on both benchmarks. Finally, (c), FPN aligns faces at a small fraction of the computational cost of comparably accurate landmark detectors. For many purposes, FPN is thus a far faster and far more accurate face alignment method than using facial landmark detectors. | What does it mean to align a face? The term alignment almost always appears in the titles of papers which present facial landmark detection methods @cite_33 @cite_49 @cite_14 (and most others) implying that the two terms are used interchangeability. This reflects an interpretation of alignment as forming correspondences between particular spatial locations in one face image and another . A different interpretation of alignment , and the one used here, refers not only to establishing these correspondences but also to warping the two face images, thus making them easier to compare and match. Face warping with estimated 2D (in-plane) or 3D transformations is well known to have a profound impact on the performance of face recognition systems @cite_42 @cite_16 . | {
"cite_N": [
"@cite_14",
"@cite_33",
"@cite_42",
"@cite_49",
"@cite_16"
],
"mid": [
"1998294030",
"2121684305",
"1916406603",
"1990937109",
"2158844819"
],
"abstract": [
"This paper presents a highly efficient, very accurate regression approach for face alignment. Our approach has two novel components: a set of local binary features, and a locality principle for learning those features. The locality principle guides us to learn a set of highly discriminative local binary features for each facial landmark independently. The obtained local binary features are used to jointly learn a linear regression for the final output. Our approach achieves the state-of-the-art results when tested on the current most challenging benchmarks. Furthermore, because extracting and regressing local binary features is computationally very cheap, our system is much faster than previous methods. It achieves over 3, 000 fps on a desktop or 300 fps on a mobile phone for locating a few dozens of landmarks.",
"The development of facial databases with an abundance of annotated facial data captured under unconstrained 'in-the-wild' conditions have made discriminative facial deformable models the de facto choice for generic facial landmark localization. Even though very good performance for the facial landmark localization has been shown by many recently proposed discriminative techniques, when it comes to the applications that require excellent accuracy, such as facial behaviour analysis and facial motion capture, the semi-automatic person-specific or even tedious manual tracking is still the preferred choice. One way to construct a person-specific model automatically is through incremental updating of the generic model. This paper deals with the problem of updating a discriminative facial deformable model, a problem that has not been thoroughly studied in the literature. In particular, we study for the first time, to the best of our knowledge, the strategies to update a discriminative model that is trained by a cascade of regressors. We propose very efficient strategies to update the model and we show that is possible to automatically construct robust discriminative person and imaging condition specific models 'in-the-wild' that outperform state-of-the-art generic face alignment strategies.",
"“Frontalization” is the process of synthesizing frontal facing views of faces appearing in single unconstrained photos. Recent reports have suggested that this process may substantially boost the performance of face recognition systems. This, by transforming the challenging problem of recognizing faces viewed from unconstrained viewpoints to the easier problem of recognizing faces in constrained, forward facing poses. Previous frontalization methods did this by attempting to approximate 3D facial shapes for each query image. We observe that 3D face shape estimation from unconstrained photos may be a harder problem than frontalization and can potentially introduce facial misalignments. Instead, we explore the simpler approach of using a single, unmodified, 3D surface as an approximation to the shape of all input faces. We show that this leads to a straightforward, efficient and easy to implement method for frontalization. More importantly, it produces aesthetic new frontal views and is surprisingly effective when used for face recognition and gender estimation.",
"We present a very efficient, highly accurate, \"Explicit Shape Regression\" approach for face alignment. Unlike previous regression-based approaches, we directly learn a vectorial regression function to infer the whole facial shape (a set of facial landmarks) from the image and explicitly minimize the alignment errors over the training data. The inherent shape constraint is naturally encoded into the regressor in a cascaded learning framework and applied from coarse to fine during the test, without using a fixed parametric shape model as in most previous methods. To make the regression more effective and efficient, we design a two-level boosted regression, shape indexed features and a correlation-based feature selection method. This combination enables us to learn accurate models from large training data in a short time (20 min for 2,000 training images), and run regression extremely fast in test (15 ms for a 87 landmarks shape). Experiments on challenging data show that our approach significantly outperforms the state-of-the-art in terms of both accuracy and efficiency.",
"Many recognition algorithms depend on careful positioning of an object into a canonical pose, so the position of features relative to a fixed coordinate system can be examined. Currently, this positioning is done either manually or by training a class-specialized learning algorithm with samples of the class that have been hand-labeled with parts or poses. In this paper, we describe a novel method to achieve this positioning using poorly aligned examples of a class with no additional labeling. Given a set of unaligned examplars of a class, such as faces, we automatically build an alignment mechanism, without any additional labeling of parts or poses in the data set. Using this alignment mechanism, new members of the class, such as faces resulting from a face detector, can be precisely aligned for the recognition process. Our alignment method improves performance on a face recognition task, both over unaligned images and over images aligned with a face alignment algorithm specifically developed for and trained on hand-labeled face images. We also demonstrate its use on an entirely different class of objects (cars), again without providing any information about parts or pose to the learning algorithm."
]
} |
1708.07517 | 2748489050 | We show how a simple convolutional neural network (CNN) can be trained to accurately and robustly regress 6 degrees of freedom (6DoF) 3D head pose, directly from image intensities. We further explain how this FacePoseNet (FPN) can be used to align faces in 2D and 3D as an alternative to explicit facial landmark detection for these tasks. We claim that in many cases the standard means of measuring landmark detector accuracy can be misleading when comparing different face alignments. Instead, we compare our FPN with existing methods by evaluating how they affect face recognition accuracy on the IJB-A and IJB-B benchmarks: using the same recognition pipeline, but varying the face alignment method. Our results show that (a) better landmark detection accuracy measured on the 300W benchmark does not necessarily imply better face recognition accuracy. (b) Our FPN provides superior 2D and 3D face alignment on both benchmarks. Finally, (c), FPN aligns faces at a small fraction of the computational cost of comparably accurate landmark detectors. For many purposes, FPN is thus a far faster and far more accurate face alignment method than using facial landmark detectors. | Although sometimes alignments involve non-parametric or part-based warps @cite_15 @cite_32 , often, global 2D or 3D (parametric) transformation are all that is required for this purpose. Such aligned faces are then further processed in systems for face recognition @cite_46 @cite_50 @cite_26 , emotion recognition @cite_4 @cite_51 , age and gender estimation @cite_17 @cite_2 @cite_3 , and more. In fact, it was recently claimed that a global alignment is both more robust and far faster to warp than non-parametric transformations @cite_27 @cite_41 . This paper focuses on such global transformations, showing how they can be estimated quickly and accurately using a deep neural network. | {
"cite_N": [
"@cite_26",
"@cite_4",
"@cite_15",
"@cite_41",
"@cite_32",
"@cite_3",
"@cite_27",
"@cite_50",
"@cite_2",
"@cite_46",
"@cite_51",
"@cite_17"
],
"mid": [
"",
"2049310893",
"2136863438",
"2304348237",
"1607528569",
"",
"2727093475",
"1901075642",
"2483369080",
"2115651492",
"2251198138",
"1965804146"
],
"abstract": [
"",
"Local binary pattern from three orthogonal planes (LBP-TOP) has been widely used in emotion recognition in the wild. However, it suffers from illumination and pose changes. This paper mainly focuses on the robustness of LBP-TOP to unconstrained environment. Recent proposed method, spatiotemporal local monogenic binary pattern (STLMBP), was verified to work promisingly in different illumination conditions. Thus this paper proposes an improved spatiotemporal feature descriptor based on STLMBP. The improved descriptor uses not only magnitude and orientation, but also the phase information, which provide complementary information. In detail, the magnitude, orientation and phase images are obtained by using an effective monogenic filter, and multiple feature vectors are finally fused by multiple kernel learning. STLMBP and the proposed method are evaluated in the Acted Facial Expression in the Wild as part of the 2014 Emotion Recognition in the Wild Challenge. They achieve competitive results, with an accuracy gain of 6.35 and 7.65 above the challenge baseline (LBP-TOP) over video.",
"We present a data-driven method for estimating the 3D shapes of faces viewed in single, unconstrained photos (aka \"in-the-wild\"). Our method was designed with an emphasis on robustness and efficiency - with the explicit goal of deployment in real-world applications which reconstruct and display faces in 3D. Our key observation is that for many practical applications, warping the shape of a reference face to match the appearance of a query, is enough to produce realistic impressions of the query's 3D shape. Doing so, however, requires matching visual features between the (possibly very different) query and reference images, while ensuring that a plausible face shape is produced. To this end, we describe an optimization process which seeks to maximize the similarity of appearances and depths, jointly, to those of a reference model. We describe our system for monocular face shape reconstruction and present both qualitative and quantitative experiments, comparing our method against alternative systems, and demonstrating its capabilities. Finally, as a testament to its suitability for real-world applications, we offer an open, on-line implementation of our system, providing unique means of instant 3D viewing of faces appearing in web photos.",
"Face recognition capabilities have recently made extraordinary leaps. Though this progress is at least partially due to ballooning training set sizes – huge numbers of face images downloaded and labeled for identity – it is not clear if the formidable task of collecting so many images is truly necessary. We propose a far more accessible means of increasing training data sizes for face recognition systems: Domain specific data augmentation. We describe novel methods of enriching an existing dataset with important facial appearance variations by manipulating the faces it contains. This synthesis is also used when matching query images represented by standard convolutional neural networks. The effect of training and testing with synthesized images is tested on the LFW and IJB-A (verification and identification) benchmarks and Janus CS2. The performances obtained by our approach match state of the art results reported by systems trained on millions of downloaded images.",
"We describe a non-parametric, \"example-based\" method for estimating the depth of an object, viewed in a single photo. Our method consults a database of example 3D geometries, searching for those which look similar to the object in the photo. The known depths of the selected database objects act as shape priors which constrain the process of estimating the object's depth. We show how this process can be performed by optimizing a well defined target likelihood function, via a hard-EM procedure. We address the problem of representing the (possibly infinite) variability of viewing conditions with a finite (and often very small) example set, by proposing an on-the-fly example update scheme. We further demonstrate the importance of non-stationarity in avoiding misleading examples when estimating structured shapes. We evaluate our method and present both qualitative as well as quantitative results for challenging object classes. Finally, we show how this same technique may be readily applied to a number of related problems. These include the novel task of estimating the occluded depth of an object's backside and the task of tailoring custom fitting image-maps for input depths.",
"",
"Recent work demonstrated that computer graphics techniques can be used to improve face recognition performances by synthesizing multiple new views of faces available in existing face collections. By so doing, more images and more appearance variations are available for training, thereby improving the deep models trained on these images. Similar rendering techniques were also applied at test time to align faces in 3D and reduce appearance variations when comparing faces. These previous results, however, did not consider the computational cost of rendering: At training, rendering millions of face images can be prohibitive; at test time, rendering can quickly become a bottleneck, particularly when multiple images represent a subject. This paper builds on a number of observations which, under certain circumstances, allow rendering new 3D views of faces at a computational cost which is equivalent to simple 2D image warping. We demonstrate this by showing that the run-time of an optimized OpenGL rendering engine is slower than the simple Python implementation we designed for the same purpose. The proposed rendering is used in a face recognition pipeline and tested on the challenging IJB-A and Janus CS2 benchmarks. Our results show that our rendering is not only fast, but improves recognition accuracy.",
"To perform unconstrained face recognition robust to variations in illumination, pose and expression, this paper presents a new scheme to extract “Multi-Directional Multi-Level Dual-Cross Patterns” (MDML-DCPs) from face images. Specifically, the MDML-DCPs scheme exploits the first derivative of Gaussian operator to reduce the impact of differences in illumination and then computes the DCP feature at both the holistic and component levels. DCP is a novel face image descriptor inspired by the unique textural structure of human faces. It is computationally efficient and only doubles the cost of computing local binary patterns, yet is extremely robust to pose and expression variations. MDML-DCPs comprehensively yet efficiently encodes the invariant characteristics of a face image from multiple levels into patterns that are highly discriminative of inter-personal differences but robust to intra-personal variations. Experimental results on the FERET, CAS-PERL-R1, FRGC 2.0, and LFW databases indicate that DCP outperforms the state-of-the-art local descriptors (e.g., LBP, LTP, LPQ, POEM, tLBP, and LGXP) for both face identification and face verification tasks. More impressively, the best performance is achieved on the challenging LFW and FRGC 2.0 databases by deploying MDML-DCPs in a simple recognition scheme.",
"We propose a two-level system for apparent age estimation from facial images. Our system first classifies samples into overlapping age groups. Within each group, the apparent age is estimated with local regressors, whose outputs are then fused for the final estimate. We use a deformable parts model based face detector, and features from a pretrained deep convolutional network. Kernel extreme learning machines are used for classification. We evaluate our system on the ChaLearn Looking at People 2016 - Apparent Age Estimation challenge dataset, and report 0.3740 normal score on the sequestered test set.",
"Recently, promising results have been shown on face recognition researches. However, face recognition and retrieval across age is still challenging. Unlike prior methods using complex models with strong parametric assumptions to model the aging process, we use a data-driven method to address this problem. We propose a novel coding framework called Cross-Age Reference Coding (CARC). By leveraging a large-scale image dataset freely available on the Internet as a reference set, CARC is able to encode the low-level feature of a face image with an age-invariant reference space. In the testing phase, the proposed method only requires a linear projection to encode the feature and therefore it is highly scalable. To thoroughly evaluate our work, we introduce a new large-scale dataset for face recognition and retrieval across age called Cross-Age Celebrity Dataset (CACD). The dataset contains more than 160,000 images of 2,000 celebrities with age ranging from 16 to 62. To the best of our knowledge, it is by far the largest publicly available cross-age face dataset. Experimental results show that the proposed method can achieve state-of-the-art performance on both our dataset as well as the other widely used dataset for face recognition across age, MORPH dataset.",
"We present a novel method for classifying emotions from static facial images. Our approach leverages on the recent success of Convolutional Neural Networks (CNN) on face recognition problems. Unlike the settings often assumed there, far less labeled data is typically available for training emotion classification systems. Our method is therefore designed with the goal of simplifying the problem domain by removing confounding factors from the input images, with an emphasis on image illumination variations. This, in an effort to reduce the amount of data required to effectively train deep CNN models. To this end, we propose novel transformations of image intensities to 3D spaces, designed to be invariant to monotonic photometric transformations. These are applied to CASIA Webface images which are then used to train an ensemble of multiple architecture CNNs on multiple representations. Each model is then fine-tuned with limited emotion labeled training data to obtain final classification models. Our method was tested on the Emotion Recognition in the Wild Challenge (EmotiW 2015), Static Facial Expression Recognition sub-challenge (SFEW) and shown to provide a substantial, 15.36 improvement over baseline results (40 gain in performance).",
"This paper concerns the estimation of facial attributes—namely, age and gender—from images of faces acquired in challenging, in the wild conditions. This problem has received far less attention than the related problem of face recognition, and in particular, has not enjoyed the same dramatic improvement in capabilities demonstrated by contemporary face recognition systems. Here, we address this problem by making the following contributions. First, in answer to one of the key problems of age estimation research—absence of data—we offer a unique data set of face images, labeled for age and gender, acquired by smart-phones and other mobile devices, and uploaded without manual filtering to online image repositories. We show the images in our collection to be more challenging than those offered by other face-photo benchmarks. Second, we describe the dropout-support vector machine approach used by our system for face attribute estimation, in order to avoid over-fitting. This method, inspired by the dropout learning techniques now popular with deep belief networks, is applied here for training support vector machines, to the best of our knowledge, for the first time. Finally, we present a robust face alignment technique, which explicitly considers the uncertainties of facial feature detectors. We report extensive tests analyzing both the difficulty levels of contemporary benchmarks as well as the capabilities of our own system. These show our method to outperform state-of-the-art by a wide margin."
]
} |
1708.07517 | 2748489050 | We show how a simple convolutional neural network (CNN) can be trained to accurately and robustly regress 6 degrees of freedom (6DoF) 3D head pose, directly from image intensities. We further explain how this FacePoseNet (FPN) can be used to align faces in 2D and 3D as an alternative to explicit facial landmark detection for these tasks. We claim that in many cases the standard means of measuring landmark detector accuracy can be misleading when comparing different face alignments. Instead, we compare our FPN with existing methods by evaluating how they affect face recognition accuracy on the IJB-A and IJB-B benchmarks: using the same recognition pipeline, but varying the face alignment method. Our results show that (a) better landmark detection accuracy measured on the 300W benchmark does not necessarily imply better face recognition accuracy. (b) Our FPN provides superior 2D and 3D face alignment on both benchmarks. Finally, (c), FPN aligns faces at a small fraction of the computational cost of comparably accurate landmark detectors. For many purposes, FPN is thus a far faster and far more accurate face alignment method than using facial landmark detectors. | Deep pose estimation. This work describes a deep network trained to estimate the 6DoF of 3D faces viewed in single images. Deep learning is increasingly used for similar purposes, though typically focusing on general object classes @cite_39 @cite_19 @cite_0 . Some recently addressed faces in particular, though their methods are designed to estimate 2D landmarks along with 3D face shapes @cite_11 @cite_47 @cite_35 . Unlike our proposed pose estimation, they regress poses by using iterative methods which involve computationally costly face rendering. We regress 6DoF directly from image intensities without such rendering steps. | {
"cite_N": [
"@cite_35",
"@cite_39",
"@cite_0",
"@cite_19",
"@cite_47",
"@cite_11"
],
"mid": [
"2964014798",
"2341204628",
"1591870335",
"2523096747",
"2589255576",
""
],
"abstract": [
"Face alignment, which fits a face model to an image and extracts the semantic meanings of facial pixels, has been an important topic in CV community. However, most algorithms are designed for faces in small to medium poses (below 45), lacking the ability to align faces in large poses up to 90. The challenges are three-fold: Firstly, the commonly used landmark-based face model assumes that all the landmarks are visible and is therefore not suitable for profile views. Secondly, the face appearance varies more dramatically across large poses, ranging from frontal view to profile view. Thirdly, labelling landmarks in large poses is extremely challenging since the invisible landmarks have to be guessed. In this paper, we propose a solution to the three problems in an new alignment framework, called 3D Dense Face Alignment (3DDFA), in which a dense 3D face model is fitted to the image via convolutional neutral network (CNN). We also propose a method to synthesize large-scale training samples in profile views to solve the third problem of data labelling. Experiments on the challenging AFLW database show that our approach achieves significant improvements over state-of-the-art methods.",
"We introduce an approach that leverages surface normal predictions, along with appearance cues, to retrieve 3D models for objects depicted in 2D still images from a large CAD object library. Critical to the success of our approach is the ability to recover accurate surface normals for objects in the depicted scene. We introduce a skip-network model built on the pre-trained Oxford VGG convolutional neural network (CNN) for surface normal prediction. Our model achieves state-of-the-art accuracy on the NYUv2 RGB-D dataset for surface normal prediction, and recovers fine object detail compared to previous methods. Furthermore, we develop a two-stream network over the input image and predicted surface normals that jointly learns pose and style for CAD model retrieval. When using the predicted surface normals, our two-stream network matches prior work using surface normals computed from RGB-D images on the task of pose prediction, and achieves state of the art when using RGB-D input. Finally, our two-stream network allows us to retrieve CAD models that better match the style and pose of a depicted object compared with baseline approaches.",
"Object viewpoint estimation from 2D images is an essential task in computer vision. However, two issues hinder its progress: scarcity of training data with viewpoint annotations, and a lack of powerful features. Inspired by the growing availability of 3D models, we propose a framework to address both issues by combining render-based image synthesis and CNNs (Convolutional Neural Networks). We believe that 3D models have the potential in generating a large number of images of high variation, which can be well exploited by deep CNN with a high learning capacity. Towards this goal, we propose a scalable and overfit-resistant image synthesis pipeline, together with a novel CNN specifically tailored for the viewpoint estimation task. Experimentally, we show that the viewpoint estimation from our pipeline can significantly outperform state-of-the-art methods on PASCAL 3D+ benchmark.",
"For applications in navigation and robotics, estimating the 3D pose of objects is as important as detection. Many approaches to pose estimation rely on detecting or tracking parts or keypoints [11, 21]. In this paper we build on a recent state-of-the-art convolutional network for slidingwindow detection [10] to provide detection and rough pose estimation in a single shot, without intermediate stages of detecting parts or initial bounding boxes. While not the first system to treat pose estimation as a categorization problem, this is the first attempt to combine detection and pose estimation at the same level using a deep learning approach. The key to the architecture is a deep convolutional network where scores for the presence of an object category, the offset for its location, and the approximate pose are all estimated on a regular grid of locations in the image. The resulting system is as accurate as recent work on pose estimation (42.4 8 View mAVP on Pascal 3D+ [21] ) and significantly faster (46 frames per second (FPS) on a TITAN X GPU). This approach to detection and rough pose estimation is fast and accurate enough to be widely applied as a pre-processing step for tasks including high-accuracy pose estimation, object tracking and localization, and vSLAM.",
"Keypoint detection is one of the most importantpre-processing steps in tasks such as face modeling, recognitionand verification. In this paper, we present an iterative methodfor Keypoint Estimation and Pose prediction of unconstrainedfaces by Learning Efficient H-CNN Regressors (KEPLER) foraddressing the face alignment problem. Recent state of the artmethods have shown improvements in face keypoint detectionby employing Convolution Neural Networks (CNNs). Althougha simple feed forward neural network can learn the mappingbetween input and output spaces, it cannot learn the inherentstructural dependencies. We present a novel architecture calledH-CNN (Heatmap-CNN) which captures structured global andlocal features and thus favors accurate keypoint detecion. H-CNNis jointly trained on the visibility, fiducials and 3D-pose of theface. As the iterations proceed, the error decreases making thegradients small and thus requiring efficient training of DCNNs tomitigate this. KEPLER performs global corrections in pose andfiducials for the first four iterations followed by local correctionsin a subsequent stage. As a by-product, KEPLER also provides3D pose (pitch, yaw and roll) of the face accurately. In thispaper, we show that without using any 3D information, KEPLERoutperforms state of the art methods for alignment on challengingdatasets such as AFW [38] and AFLW [17].",
""
]
} |
1708.07517 | 2748489050 | We show how a simple convolutional neural network (CNN) can be trained to accurately and robustly regress 6 degrees of freedom (6DoF) 3D head pose, directly from image intensities. We further explain how this FacePoseNet (FPN) can be used to align faces in 2D and 3D as an alternative to explicit facial landmark detection for these tasks. We claim that in many cases the standard means of measuring landmark detector accuracy can be misleading when comparing different face alignments. Instead, we compare our FPN with existing methods by evaluating how they affect face recognition accuracy on the IJB-A and IJB-B benchmarks: using the same recognition pipeline, but varying the face alignment method. Our results show that (a) better landmark detection accuracy measured on the 300W benchmark does not necessarily imply better face recognition accuracy. (b) Our FPN provides superior 2D and 3D face alignment on both benchmarks. Finally, (c), FPN aligns faces at a small fraction of the computational cost of comparably accurate landmark detectors. For many purposes, FPN is thus a far faster and far more accurate face alignment method than using facial landmark detectors. | In all these cases, absence of training data was cited as a major obstacle for training effective models. In response, some turned to larger 3D object data sets @cite_1 @cite_22 or using synthetically generated examples @cite_5 . We propose a far simpler alternative and show it to result in robust and accurate face alignment. | {
"cite_N": [
"@cite_5",
"@cite_1",
"@cite_22"
],
"mid": [
"2951152493",
"1991264156",
"2519379752"
],
"abstract": [
"Reconstructing the detailed geometric structure of a face from a given image is a key to many computer vision and graphics applications, such as motion capture and reenactment. The reconstruction task is challenging as human faces vary extensively when considering expressions, poses, textures, and intrinsic geometries. While many approaches tackle this complexity by using additional data to reconstruct the face of a single subject, extracting facial surface from a single image remains a difficult problem. As a result, single-image based methods can usually provide only a rough estimate of the facial geometry. In contrast, we propose to leverage the power of convolutional neural networks to produce a highly detailed face reconstruction from a single image. For this purpose, we introduce an end-to-end CNN framework which derives the shape in a coarse-to-fine fashion. The proposed architecture is composed of two main blocks, a network that recovers the coarse facial geometry (CoarseNet), followed by a CNN that refines the facial features of that geometry (FineNet). The proposed networks are connected by a novel layer which renders a depth image given a mesh in 3D. Unlike object recognition and detection problems, there are no suitable datasets for training CNNs to perform face geometry reconstruction. Therefore, our training regime begins with a supervised phase, based on synthetic images, followed by an unsupervised phase that uses only unconstrained facial images. The accuracy and robustness of the proposed model is demonstrated by both qualitative and quantitative evaluation tests.",
"3D object detection and pose estimation methods have become popular in recent years since they can handle ambiguities in 2D images and also provide a richer description for objects compared to 2D object detectors. However, most of the datasets for 3D recognition are limited to a small amount of images per category or are captured in controlled environments. In this paper, we contribute PASCAL3D+ dataset, which is a novel and challenging dataset for 3D object detection and pose estimation. PASCAL3D+ augments 12 rigid categories of the PASCAL VOC 2012 [4] with 3D annotations. Furthermore, more images are added for each category from ImageNet [3]. PASCAL3D+ images exhibit much more variability compared to the existing 3D datasets, and on average there are more than 3,000 object instances per category. We believe this dataset will provide a rich testbed to study 3D detection and pose estimation and will help to significantly push forward research in this area. We provide the results of variations of DPM [6] on our new dataset for object detection and viewpoint estimation in different scenarios, which can be used as baselines for the community. Our benchmark is available online at http: cvgl.stanford.edu projects pascal3d",
"We contribute a large scale database for 3D object recognition, named ObjectNet3D, that consists of 100 categories, 90,127 images, 201,888 objects in these images and 44,147 3D shapes. Objects in the 2D images in our database are aligned with the 3D shapes, and the alignment provides both accurate 3D pose annotation and the closest 3D shape annotation for each 2D object. Consequently, our database is useful for recognizing the 3D pose and 3D shape of objects from 2D images. We also provide baseline experiments on four tasks: region proposal generation, 2D object detection, joint 2D detection and 3D object pose estimation, and image-based 3D shape retrieval, which can serve as baselines for future research using our database. Our database is available online at http: cvgl.stanford.edu projects objectnet3d."
]
} |
1708.07715 | 2745712772 | The tendency to overestimate immediate utility is a common cognitive bias. As a result people behave inconsistently over time and fail to reach long-term goals. Behavioral economics tries to help affected individuals by implementing external incentives. However, designing robust incentives is often difficult due to imperfect knowledge of the parameter @math quantifying a person's present bias. Using the graphical model of Kleinberg and Oren, we approach this problem from an algorithmic perspective. Based on the assumption that the only information about @math is its membership in some set @math , we distinguish between two models of uncertainty: one in which @math is fixed and one in which it varies over time. As our main result we show that the conceptual loss of efficiency incurred by incentives in the form of penalty fees is at most @math in the former and @math in the latter model. We also give asymptotically matching lower bounds and approximation algorithms. | A significant amount of research in this field has been devoted to temporal discounting in general and quasi-hyperbolic discounting in particular, see @cite_5 for a survey. The quasi-hyperbolic discounting model proposed by Laibson @cite_12 is characterized by two parameters: the present bias @math and the exponential discount rate @math . People who plan according to this model have an accurate perception of the present, but scale down any costs and rewards realized @math time units in the future by a factor of @math . To keep our work clearly delineated in scope, we adopt Akerlof's model of quasi-hyperbolic discounting @cite_6 and make the following two assumptions: First, we focus on the present bias @math and set the exponential discount rate to @math . Secondly, we assume people to be naive in the sense that they are unaware of their present bias and only optimize their current perceived utility when making a decision. Note that Alice and Bob from the previous example behave like agents in Akerlof's model for a present bias of @math and @math respectively. | {
"cite_N": [
"@cite_5",
"@cite_6",
"@cite_12"
],
"mid": [
"2145107370",
"",
"2118052532"
],
"abstract": [
"This paper discusses the discounted utility (DU) model: its historical development, underlying assumptions, and \"anomalies\" - the empirical regularities that are inconsistent with its theoretical predictions. We then summarize the alternate theoretical formulations that have been advanced to address these anomalies. We also review three decades of empirical research on intertemporal choice, and discuss reasons for the spectacular variation in implicit discount rates across studies. Throughout the paper, we stress the importance of distinguishing time preference, per se, from many other considerations that also influence intertemporal choices.",
"",
"Hyperbolic discount functions induce dynamically inconsistent preferences, implying a motive for consumers to constrain their own future choices. This paper analyzes the decisions of a hyperbolic consumer who has access to an imperfect commitment technology: an illiquid asset whose sale must be initiated one period before the sale proceeds are received. The model predicts that consumption tracks income, and the model explains why consumers have asset-specific marginal propensities to consume. The model suggests that financial innovation may have caused the ongoing decline in U. S. savings rates, since financial innovation in- creases liquidity, eliminating commitment opportunities. Finally, the model implies that financial market innovation may reduce welfare by providing “too much†liquidity."
]
} |
1708.07715 | 2745712772 | The tendency to overestimate immediate utility is a common cognitive bias. As a result people behave inconsistently over time and fail to reach long-term goals. Behavioral economics tries to help affected individuals by implementing external incentives. However, designing robust incentives is often difficult due to imperfect knowledge of the parameter @math quantifying a person's present bias. Using the graphical model of Kleinberg and Oren, we approach this problem from an algorithmic perspective. Based on the assumption that the only information about @math is its membership in some set @math , we distinguish between two models of uncertainty: one in which @math is fixed and one in which it varies over time. As our main result we show that the conceptual loss of efficiency incurred by incentives in the form of penalty fees is at most @math in the former and @math in the latter model. We also give asymptotically matching lower bounds and approximation algorithms. | The graphical model is of particular interest to us as it provides a natural framework for a design problem frequently encountered in behavioral economics. Given a certain social or economic setting, the problem is to improve a time-inconsistent person's performance via various sorts of incentives , such as monetary rewards, deadlines or penalty fees, see e.g. @cite_7 . Using the graphical model, Kleinberg and Oren demonstrate how a strategic choice reduction can incentivize people to reach predefined goals @cite_4 . To implement their incentives, they simply remove the corresponding edges from the task graph. However, there is a computational drawback to this approach. As we have shown in previous work, an optimal set of edges to remove from a task graph with @math nodes is NP-hard to approximate within a factor less than @math @cite_8 . A more general form of incentives avoiding these harsh complexity theoretic limitations are penalty fees. In the graphical model penalty fees are at least as powerful as choice reduction and admit a polynomial time @math -approximation @cite_3 . | {
"cite_N": [
"@cite_3",
"@cite_4",
"@cite_7",
"@cite_8"
],
"mid": [
"2594830350",
"1971014051",
"1583523562",
"2225272567"
],
"abstract": [
"People tend to behave inconsistently over time due to an inherent present bias. As this may impair performance, social and economic settings need to be adapted accordingly. Common tools to reduce the impact of time-inconsistent behavior are penalties and prohibition. Such tools are called commitment devices. In recent work Kleinberg and Oren [EC, 2014] connect the design of a prohibition-based commitment device to a combinatorial problem in which edges are removed from a task graph G with n nodes. However, this problem is NP-hard to approximate within a ratio less than n^(1 2) 3 [Albers and Kraft, WINE, 2016]. To address this issue, we propose a penalty-based commitment device that does not delete edges, but raises their cost. The benefits of our approach are twofold. On the conceptual side, we show that penalties are up to 1 beta times more efficient than prohibition, where 0 < beta <= 1 parameterizes the present bias. On the computational side, we improve approximability by presenting a 2-approximation algorithm for allocating penalties. To complement this result, we prove that optimal penalties are NP-hard to approximate within a ratio of 1.08192.",
"In many settings, people exhibit behavior that is inconsistent across time ' we allocate a block of time to get work done and then procrastinate, or put effort into a project and then later fail to complete it. An active line of research in behavioral economics and related fields has developed and analyzed models for this type of time-inconsistent behavior. Here we propose a graph-theoretic model of tasks and goals, in which dependencies among actions are represented by a directed graph, and a time-inconsistent agent constructs a path through this graph. We first show how instances of this path-finding problem on different input graphs can reconstruct a wide range of qualitative phenomena observed in the literature on time-inconsistency, including procrastination, abandonment of long-range tasks, and the benefits of reduced sets of choices. We then explore a set of analyses that quantify over the set of all graphs; among other results, we find that in any graph, there can be only polynomially many distinct forms of time-inconsistent behavior; and any graph in which a time-inconsistent agent incurs significantly more cost than an optimal agent must contain a large 'procrastination' structure as a minor. Finally, we use this graph-theoretic model to explore ways in which tasks can be designed to help motivate agents to reach designated goals.",
"",
"We study the complexity of motivating time-inconsistent agents to complete long term projects in a graph-based planning model as proposed by Kleinberg and Oreni¾?[5]. Given a task graph G with n nodes, our objective is to guide an agent towards a target node t under certain budget constraints. The crux is that the agent may change its strategy over time due to its present-bias. We consider two strategies to guide the agent. First, a single reward is placed at t and arbitrary edges can be removed from G. Secondly, rewards can be placed at arbitrary nodes of G but no edges must be deleted. In both cases we show that it is NP-complete to decide if a given budget is sufficient to guide the agent. For the first setting, we give complementing upper and lower bounds on the approximability of the minimum required budget. In particular, we devise a @math 1+n-approximation algorithm and prove NP-hardness for ratios greater than @math n 3. Finally, we argue that the second setting does not permit any efficient approximation unless @math P=NP."
]
} |
1708.07279 | 2746222821 | Neural network models have recently received heated research attention in the natural language processing community. Compared with traditional models with discrete features, neural models have two main advantages. First, they take low-dimensional, real-valued embedding vectors as inputs, which can be trained over large raw data, thereby addressing the issue of feature sparsity in discrete models. Second, deep neural networks can be used to automatically combine input features, and including non-local features that capture semantic patterns that cannot be expressed using discrete indicator features. As a result, neural network models have achieved competitive accuracies compared with the best discrete models for a range of NLP tasks. On the other hand, manual feature templates have been carefully investigated for most NLP tasks over decades and typically cover the most useful indicator pattern for solving the problems. Such information can be complementary the features automatically induced from neural networks, and therefore combining discrete and neural features can potentially lead to better accuracy compared with models that leverage discrete or neural features only. In this paper, we systematically investigate the effect of discrete and neural feature combination for a range of fundamental NLP tasks based on sequence labeling, including word segmentation, POS tagging and named entity recognition for Chinese and English, respectively. Our results on standard benchmarks show that state-of-the-art neural models can give accuracies comparable to the best discrete models in the literature for most tasks and combing discrete and neural features unanimously yield better results. | There has been two main kinds of methods for word segmentation. Xue @cite_10 treat it as a sequence labeling task, using B(egin) I(nternal) E(nding) S(ingle-character word) tags on each character in the input to indicate its segmentation status. The method was followed by @cite_29 , who use CRF to improve the accuracies. Most subsequent work follows @cite_29 @cite_12 @cite_64 @cite_9 @cite_1 and feature engineering has been out of the key research questions. This kind of research is commonly referred to as the character-based method. Recently, neural networks have been applied to character-based segmentation @cite_22 @cite_8 @cite_49 , giving results comparable to discrete methods. On the other hand, the second kind of work studies word-based segmentation, scoring outputs based on word features directly @cite_30 @cite_3 @cite_37 . We focus on the character-based method, which is a typical sequence labeling problem. | {
"cite_N": [
"@cite_30",
"@cite_37",
"@cite_64",
"@cite_22",
"@cite_8",
"@cite_9",
"@cite_29",
"@cite_1",
"@cite_3",
"@cite_49",
"@cite_10",
"@cite_12"
],
"mid": [
"2165664509",
"2143026224",
"",
"2251811146",
"2252225757",
"2159406587",
"2036516910",
"2137281231",
"1512811194",
"",
"2163377725",
"2063925051"
],
"abstract": [
"Standard approaches to Chinese word segmentation treat the problem as a tagging task, assigning labels to the characters in the sequence indicating whether the character marks a word boundary. Discriminatively trained models based on local character features are used to make the tagging decisions, with Viterbi decoding finding the highest scoring segmentation. In this paper we propose an alternative, word-based segmentor, which uses features based on complete words and word sequences. The generalized perceptron algorithm is used for discriminative training, and we use a beamsearch decoder. Closed tests on the first and second SIGHAN bakeoffs show that our system is competitive with the best in the literature, achieving the highest reported F-scores for a number of corpora.",
"We report an empirical investigation on type-supervised domain adaptation for joint Chinese word segmentation and POS-tagging, making use of domainspecific tag dictionaries and only unlabeled target domain data to improve target-domain accuracies, given a set of annotated source domain sentences. Previous work on POS-tagging of other languages showed that type-supervision can be a competitive alternative to tokensupervision, while semi-supervised techniques such as label propagation are important to the effectiveness of typesupervision. We report similar findings using a novel approach for joint Chinese segmentation and POS-tagging, under a cross-domain setting. With the help of unlabeled sentences and a lexicon of 3,000 words, we obtain 33 error reduction in target-domain tagging. In addition, combined type- and token-supervision can lead to improved cost-effectiveness.",
"",
"This study explores the feasibility of performing Chinese word segmentation (CWS) and POS tagging by deep learning. We try to avoid task-specific feature engineering, and use deep layers of neural networks to discover relevant features to the tasks. We leverage large-scale unlabeled data to improve internal representation of Chinese characters, and use these improved representations to enhance supervised word segmentation and POS tagging models. Our networks achieved close to state-of-theart performance with minimal computational cost. We also describe a perceptron-style algorithm for training the neural networks, as an alternative to maximum-likelihood method, to speed up the training process and make the learning algorithm easier to be implemented.",
"Recently, neural network models for natural language processing tasks have been increasingly focused on for their ability to alleviate the burden of manual feature engineering. In this paper, we propose a novel neural network model for Chinese word segmentation called Max-Margin Tensor Neural Network (MMTNN). By exploiting tag embeddings and tensorbased transformation, MMTNN has the ability to model complicated interactions between tags and context characters. Furthermore, a new tensor factorization approach is proposed to speed up the model and avoid overfitting. Experiments on the benchmark dataset show that our model achieves better performances than previous neural network models and that our model can achieve a competitive performance with minimal feature engineering. Despite Chinese word segmentation being a specific case, MMTNN can be easily generalized and applied to other sequence labeling tasks.",
"The large combined search space of joint word segmentation and Part-of-Speech (POS) tagging makes efficient decoding very hard. As a result, effective high order features representing rich contexts are inconvenient to use. In this work, we propose a novel stacked subword model for this task, concerning both efficiency and effectiveness. Our solution is a two step process. First, one word-based segmenter, one character-based segmenter and one local character classifier are trained to produce coarse segmentation and POS information. Second, the outputs of the three predictors are merged into sub-word sequences, which are further bracketed and labeled with POS tags by a fine-grained sub-word tagger. The coarse-to-fine search scheme is efficient, while in the sub-word tagging step rich contextual features can be approximately derived. Evaluation on the Penn Chinese Tree-bank shows that our model yields improvements over the best system reported in the literature.",
"Chinese word segmentation is a difficult, important and widely-studied sequence modeling problem. This paper demonstrates the ability of linear-chain conditional random fields (CRFs) to perform robust and accurate Chinese word segmentation by providing a principled framework that easily supports the integration of domain knowledge in the form of multiple lexicons of characters and words. We also present a probabilistic new word detection method, which further improves performance. Our system is evaluated on four datasets used in a recent comprehensive Chinese word segmentation competition. State-of-the-art performance is obtained.",
"Supervised methods have been the dominant approach for Chinese word segmentation. The performance can drop significantly when the test domain is different from the training domain. In this paper, we study the problem of obtaining partial annotation from freely available data to help Chinese word segmentation on different domains. Different sources of free annotations are transformed into a unified form of partial annotation and a variant CRF model is used to leverage both fully and partially annotated data consistently. Experimental results show that the Chinese word segmentation model benefits from free partially annotated data. On the SIGHAN Bakeoff 2010 data, we achieve results that are competitive to the best reported in the literature.",
"We present a theoretical and empirical comparative analysis of the two dominant categories of approaches in Chinese word segmentation: word-based models and character-based models. We show that, in spite of similar performance overall, the two models produce different distribution of segmentation errors, in a way that can be explained by theoretical properties of the two models. The analysis is further exploited to improve segmentation accuracy by integrating a word-based segmenter and a character-based segmenter. A Bootstrap Aggregating model is proposed. By letting multiple segmenters vote, our model improves segmentation consistently on the four different data sets from the second SIGHAN bakeoff.",
"",
"In this paper we report results of a supervised machine-learning approach to Chinese word segmentation. A maximum entropy tagger is trained on manually annotated data to automatically assign to Chinese characters, or hanzi, tags that indicate the position of a hanzi within a word. The tagged output is then converted into segmented text for evaluation. Preliminary results show that this approach is competitive against other supervised machine-learning segmenters reported in previous studies, achieving precision and recall rates of 95.01 and 94.94 respectively, trained on a 237K-word training set.",
"We investigate the possibility of exploiting character-based dependency for Chinese information processing. As Chinese text is made up of character sequences rather than word sequences, word in Chinese is not so natural a concept as in English, nor is word easy to be defined without argument for such a language. Therefore we propose a character-level dependency scheme to represent primary linguistic relationships within a Chinese sentence. The usefulness of character dependencies are verified through two specialized dependency parsing tasks. The first is to handle trivial character dependencies that are equally transformed from traditional word boundaries. The second furthermore considers the case that annotated internal character dependencies inside a word are involved. Both of these results from character-level dependency parsing are positive. This study provides an alternative way to formularize basic character-and word-level representation for Chinese."
]
} |
1708.07279 | 2746222821 | Neural network models have recently received heated research attention in the natural language processing community. Compared with traditional models with discrete features, neural models have two main advantages. First, they take low-dimensional, real-valued embedding vectors as inputs, which can be trained over large raw data, thereby addressing the issue of feature sparsity in discrete models. Second, deep neural networks can be used to automatically combine input features, and including non-local features that capture semantic patterns that cannot be expressed using discrete indicator features. As a result, neural network models have achieved competitive accuracies compared with the best discrete models for a range of NLP tasks. On the other hand, manual feature templates have been carefully investigated for most NLP tasks over decades and typically cover the most useful indicator pattern for solving the problems. Such information can be complementary the features automatically induced from neural networks, and therefore combining discrete and neural features can potentially lead to better accuracy compared with models that leverage discrete or neural features only. In this paper, we systematically investigate the effect of discrete and neural feature combination for a range of fundamental NLP tasks based on sequence labeling, including word segmentation, POS tagging and named entity recognition for Chinese and English, respectively. Our results on standard benchmarks show that state-of-the-art neural models can give accuracies comparable to the best discrete models in the literature for most tasks and combing discrete and neural features unanimously yield better results. | POS-tagging has been investigated as a classic sequence labeling problem @cite_63 @cite_50 @cite_6 , for which a well-established set of features are used. These handcrafted features basically include words, the context of words, word morphologies and word shapes. Various neural network models have also been used for this task. In order to include word morphology and word shape knowledge, a convolutional neural network (CNN) for automatically learning character-level representations is investigated in @cite_35 . @cite_54 built a CNN neural network for multiple sequence labeling tasks, which gives state-of-the-art POS results. Recurrent neural network models have also been used for this task @cite_36 @cite_18 . @cite_18 combines bidirectional LSTM with a CRF layer, their model is robust and has less dependence on word embedding. | {
"cite_N": [
"@cite_35",
"@cite_18",
"@cite_36",
"@cite_54",
"@cite_6",
"@cite_50",
"@cite_63"
],
"mid": [
"2101609803",
"1940872118",
"",
"",
"1515847863",
"2008652694",
"1773803948"
],
"abstract": [
"Distributed word representations have recently been proven to be an invaluable resource for NLP. These representations are normally learned using neural networks and capture syntactic and semantic information about words. Information about word morphology and shape is normally ignored when learning word representations. However, for tasks like part-of-speech tagging, intra-word information is extremely useful, specially when dealing with morphologically rich languages. In this paper, we propose a deep neural network that learns character-level representation of words and associate them with usual word representations to perform POS tagging. Using the proposed approach, while avoiding the use of any handcrafted feature, we produce state-of-the-art POS taggers for two languages: English, with 97.32 accuracy on the Penn Treebank WSJ corpus; and Portuguese, with 97.47 accuracy on the Mac-Morpho corpus, where the latter represents an error reduction of 12.2 on the best previous known result.",
"In this paper, we propose a variety of Long Short-Term Memory (LSTM) based models for sequence tagging. These models include LSTM networks, bidirectional LSTM (BI-LSTM) networks, LSTM with a Conditional Random Field (CRF) layer (LSTM-CRF) and bidirectional LSTM with a CRF layer (BI-LSTM-CRF). Our work is the first to apply a bidirectional LSTM CRF (denoted as BI-LSTM-CRF) model to NLP benchmark sequence tagging data sets. We show that the BI-LSTM-CRF model can efficiently use both past and future input features thanks to a bidirectional LSTM component. It can also use sentence level tag information thanks to a CRF layer. The BI-LSTM-CRF model can produce state of the art (or close to) accuracy on POS, chunking and NER data sets. In addition, it is robust and has less dependence on word embedding as compared to previous observations.",
"",
"",
"I examine what would be necessary to move part-of-speech tagging performance from its current level of about 97.3 token accuracy (56 sentence accuracy) to close to 100 accuracy. I suggest that it must still be possible to greatly increase tagging performance and examine some useful improvements that have recently been made to the Stanford Part-of-Speech Tagger. However, an error analysis of some of the remaining errors suggests that there is limited further mileage to be had either from better machine learning or better features in a discriminative sequence classifier. The prospects for further gains from semisupervised learning also seem quite limited. Rather, I suggest and begin to demonstrate that the largest opportunity for further progress comes from improving the taxonomic basis of the linguistic resources from which taggers are trained. That is, from improved descriptive linguistics. However, I conclude by suggesting that there are also limits to this process. The status of some words may not be able to be adequately captured by assigning them to one of a small number of categories. While conventions can be used in such cases to improve tagging consistency, they lack a strong linguistic basis.",
"We describe new algorithms for training tagging models, as an alternative to maximum-entropy models or conditional random fields (CRFs). The algorithms rely on Viterbi decoding of training examples, combined with simple additive updates. We describe theory justifying the algorithms through a modification of the proof of convergence of the perceptron algorithm for classification problems. We give experimental results on part-of-speech tagging and base noun phrase chunking, in both cases showing improvements over results for a maximum-entropy tagger.",
"This paper presents a statistical model which trains from a corpus annotated with Part Of Speech tags and assigns them to previously unseen text with state of the art accuracy The model can be classi ed as a Maximum Entropy model and simultaneously uses many contextual features to predict the POS tag Furthermore this paper demonstrates the use of specialized fea tures to model di cult tagging decisions discusses the corpus consistency problems discovered during the implementation of these features and proposes a training strategy that mitigates these problems"
]
} |
1708.07279 | 2746222821 | Neural network models have recently received heated research attention in the natural language processing community. Compared with traditional models with discrete features, neural models have two main advantages. First, they take low-dimensional, real-valued embedding vectors as inputs, which can be trained over large raw data, thereby addressing the issue of feature sparsity in discrete models. Second, deep neural networks can be used to automatically combine input features, and including non-local features that capture semantic patterns that cannot be expressed using discrete indicator features. As a result, neural network models have achieved competitive accuracies compared with the best discrete models for a range of NLP tasks. On the other hand, manual feature templates have been carefully investigated for most NLP tasks over decades and typically cover the most useful indicator pattern for solving the problems. Such information can be complementary the features automatically induced from neural networks, and therefore combining discrete and neural features can potentially lead to better accuracy compared with models that leverage discrete or neural features only. In this paper, we systematically investigate the effect of discrete and neural feature combination for a range of fundamental NLP tasks based on sequence labeling, including word segmentation, POS tagging and named entity recognition for Chinese and English, respectively. Our results on standard benchmarks show that state-of-the-art neural models can give accuracies comparable to the best discrete models in the literature for most tasks and combing discrete and neural features unanimously yield better results. | Named entity recognition is also a classical sequence labeling task in the NLP community. Similar to other tasks, most works access NER problem through feature engineering. McCallum & Li @cite_17 use CRF model for NER task and exploit Web lexicon as feature enhancement. Chieu & Ng @cite_5 , Krishnan & Manning @cite_28 and @cite_39 tackle this task through non-local features @cite_40 . Besides, many neural models, which are free from handcrafted features, have been proposed in recent years. In @cite_54 model we referred before, NER task has also been included. @cite_23 boost the neural model by adding character embedding on Collobert's structure. @cite_21 take the lead by employing LSTM for NER tasks. @cite_55 use CNN model to extract character embedding and attach it with word embedding and afterwards feed them into Bi-directional LSTM model. Through adding lexicon features, Chiu's NER system get state-of-the-art performance on both CoNLL2003 and OntoNotes 5.0 NER datasets. | {
"cite_N": [
"@cite_28",
"@cite_54",
"@cite_21",
"@cite_55",
"@cite_39",
"@cite_40",
"@cite_23",
"@cite_5",
"@cite_17"
],
"mid": [
"",
"",
"2042188227",
"2179519966",
"2251435463",
"2004763266",
"",
"",
"2141099517"
],
"abstract": [
"",
"",
"In this approach to named entity recognition, a recurrent neural network, known as Long Short-Term Memory, is applied. The network is trained to perform 2 passes on each sentence, outputting its decisions on the second pass. The first pass is used to acquire information for disambiguation during the second pass. SARDNET, a self-organising map for sequences is used to generate representations for the lexical items presented to the LSTM network, whilst orthogonal representations are used to represent the part of speech and chunk tags.",
"Named entity recognition is a challenging task that has traditionally required large amounts of knowledge in the form of feature engineering and lexicons to achieve high performance. In this paper, we present a novel neural network architecture that automatically detects word- and character-level features using a hybrid bidirectional LSTM and CNN architecture, eliminating the need for most feature engineering. We also propose a novel method of encoding partial lexicon matches in neural networks and compare it to existing approaches. Extensive evaluation shows that, given only tokenized text and publicly available word embeddings, our system is competitive on the CoNLL-2003 dataset and surpasses the previously reported state of the art performance on the OntoNotes 5.0 dataset by 2.13 F1 points. By using two lexicons constructed from publicly-available sources, we establish new state of the art performance with an F1 score of 91.62 on CoNLL-2003 and 86.28 on OntoNotes, surpassing systems that employ heavy feature engineering, proprietary lexicons, and rich entity linking information.",
"Different languages contain complementary cues about entities, which can be used to improve Named Entity Recognition (NER) systems. We propose a method that formulates the problem of exploring such signals on unannotated bilingual text as a simple Integer Linear Program, which encourages entity tags to agree via bilingual constraints. Bilingual NER experiments on the large OntoNotes 4.0 Chinese-English corpus show that the proposed method can improve strong baselines for both Chinese and English. In particular, Chinese performance improves by over 5 absolute F1 score. We can then annotate a large amount of bilingual text (80k sentence pairs) using our method, and add it as uptraining data to the original monolingual NER training corpus. The Chinese model retrained on this new combined dataset outperforms the strong baseline by over 3 F1 score.",
"We analyze some of the fundamental design challenges and misconceptions that underlie the development of an efficient and robust NER system. In particular, we address issues such as the representation of text chunks, the inference approach needed to combine local NER decisions, the sources of prior knowledge and how to use them within an NER system. In the process of comparing several solutions to these challenges we reach some surprising conclusions, as well as develop an NER system that achieves 90.8 F1 score on the CoNLL-2003 NER shared task, the best reported result for this dataset.",
"",
"",
"Models for many natural language tasks benefit from the flexibility to use overlapping, non-independent features. For example, the need for labeled data can be drastically reduced by taking advantage of domain knowledge in the form of word lists, part-of-speech tags, character n-grams, and capitalization patterns. While it is difficult to capture such inter-dependent features with a generative probabilistic model, conditionally-trained models, such as conditional maximum entropy models, handle them well. There has been significant work with such models for greedy sequence modeling in NLP (Ratnaparkhi, 1996; , 1998)."
]
} |
1708.07671 | 2749924182 | Let @math be the orientable surface of genus @math . We prove that the component structure of a graph chosen uniformly at random from the class @math of all graphs on vertex set @math with @math edges embeddable on @math features two phase transitions. The first phase transition mirrors the classical phase transition in the Erdős--Renyi random graph @math chosen uniformly at random from all graphs with vertex set @math and @math edges. It takes place at @math , when a unique largest component, the so-called , emerges. The second phase transition occurs at @math , when the giant component covers almost all vertices of the graph. This kind of phenomenon is strikingly different from @math and has only been observed for graphs on surfaces. Moreover, we derive an asymptotic estimation of the number of graphs in @math throughout the regimes of these two phase transitions. | The order of the largest component of the random graph @math at the time of the phase transition has been extensively studied @cite_17 @cite_28 @cite_60 @cite_41 @cite_62 . Most of the results have been proved using purely probabilistic arguments (e.g. random walks, martingales), leading to even stronger results than the ones stated in thm:ER , e.g. about the limiting distribution of the order and size of the largest component @cite_34 @cite_19 @cite_28 @cite_16 . In the case of @math , the additional condition of the graph being embeddable on @math makes it virtually impossible to use the same techniques in order to derive such strong results. | {
"cite_N": [
"@cite_62",
"@cite_60",
"@cite_28",
"@cite_41",
"@cite_16",
"@cite_19",
"@cite_34",
"@cite_17"
],
"mid": [
"2036312115",
"2040419749",
"2088110451",
"2012184276",
"2530299459",
"2079728415",
"2161355281",
""
],
"abstract": [
"The theme of this work is an \"inside-out\" approach to the enumeration of graphs. It is based on a well-known decomposition of a graph into its 2-core, i.e. the largest subgraph of minimum degree 2 or more, and a forest of trees attached. Using our earlier (asymptotic) formulae for the total number of 2-cores with a given number of vertices and edges, we solve the corresponding enumeration problem for the connected 2-cores. For a subrange of the parameters, we also enumerate those 2-cores by using a deeper inside-out notion of a kernel of a connected 2-core.Using this enumeration result in combination with Caley's formula for forests, we obtain an alternative and simpler proof of the asymptotic formula of Bender, Canfield and McKay for the number of connected graphs with n vertices and m edges, with improved error estimate for a range of m values.As another application, we study the limit joint distribution of three parameters of the giant component of a random graph with n vertices in the supercritical phase, when the difference between average vertex degree and 1 far exceeds n-1 3. The three parameters are defined in terms of the 2-core of the giant component, i.e. its largest subgraph of minimum degree 2 or more. They are the number of vertices in the 2-core, the excess (#edges - #vertices) of the 2-core, and the number of vertices not in the 2-core. We show that the limit distribution is jointly Gaussian throughout the whole supercritical phase. In particular, for the first time, the 2-core size is shown to be asymptotically normal, in the widest possible range of the average vertex degree.",
"We study the behavior of a random graph process (G(n, M))02 n for M(n) = n 2 + s and ∣s∣3n−;2 → ∞. Among others we find the number of components in G(n, M) and estimate the number of vertices and edges in the kth largest component of G(n, M), for any natural number k, Moreover, it is shown that, with probability 1 –o(1), when M(n) = n 2 + s, s3n−2 →−∞, then during a random graph process in some step M1 > M a “new” largest component will emerge, whereas when s3n−2→∞, the largest component of G(n, M) remains largest until the very end of the process. © 1990 Wiley Periodicals, Inc.",
"In this paper we give a simple new proof of a result of Pittel and Wormald concerning the asymptotic value and (suitably rescaled) limiting distribution of the number of vertices in the giant component of G(n,p) above the scaling window of the phase transition. Nachmias and Peres used martingale arguments to study Karp@?s exploration process, obtaining a simple proof of a weak form of this result. We use slightly different martingale arguments to obtain a much sharper result with little extra work.",
"Consider the random graph models G(n, #edges = M) and G(n, Prob(edge) = p) with M = M(n) = (1 + )n-1 3)n 2 and p = p(n) = (1 +An-1 3) n . For 1 > -1 define an i-component of a random graph as a component which has exactly I more edges than vertices. Call an i-component with I > 1 a complex component. For both models, we show that when A is constant, the expected number of complex components is bounded, almost surely (a.s.) each of these components (if any exist) has size of order n2 3, and the maximum value of I is bounded in probability. We prove that a.s. the largest suspended tree in each complex component has size of order n2 3, and deletion of all suspended trees results in a \"smoothed\" graph of size of order n 3, with the maximum vertex degree 3. The total number of branching vertices, i.e., of degree 3, is bounded in probability. Thus, each complex component is almost surely topologically equivalent to a 3-regular multigraph of a uniformly bounded size. Lengths of the shortest cycle and of the shortest path between two branching vertices of a smoothed graph are each of order n 3. We find a relatively simple integral formula for the limit distribution of the numbers of complex components, which implies, in particular, that all values of the \"complexity spectrum\" have positive limiting probabilities. We also answer questions raised by Erdos and Renyi back in 1960. It is proven that there exists p(A), the limiting planarity probability, with 0 < p(A) < 1, p(-oo) = 1, p(oo) = 0. In particular, G(n, M) (G(n, p), resp.) is almost surely nonplanar iff (M n 2)n-213 -* 00 ((np I)n-1 3) -+ oo, resp.).",
"Recently, in [Random Struct Algorithm 41 2012, 441-450] we adapted exploration and martingale arguments of Nachmias and Peres [ALEA Lat Am J Probab Math Stat 3 2007, 133-142], in turn based on ideas of Martin-Lof [J Appl Probab 23 1986, 265-282], Karp [Random Struct Alg 1 1990, 73-93] and Aldous [Ann Probab 25 1997, 812-854], to prove asymptotic normality of the number L1 of vertices in the largest component L1 of the random r-uniform hypergraph in the supercritical regime. In this paper we take these arguments further to prove two new results: strong tail bounds on the distribution of L1, and joint asymptotic normality of L1 and the number M1 of edges of L1 in the sparsely supercritical case. These results are used in [Combin Probab Comput 25 2016, 21-75], where we enumerate sparsely connected hypergraphs asymptotically. © 2016 Wiley Periodicals, Inc. Random Struct. Alg., 50, 325-352, 2017",
"Let H d (n,p) signify a random d -uniform hypergraph with n vertices in which each of the @math possible edges is present with probability p = p(n) independently, and let H d (n,m) denote a uniformly distributed d -uniform hypergraph with n vertices and m edges. We derive local limit theorems for the joint distribution of the number of vertices and the number of edges in the largest component of H d (n,p) and H d (n,m) in the regime @math , resp. d(d−1)m n >1+ϵ, where ϵ>0 is arbitrarily small but fixed as n → ∞. The proofs are based on a purely probabilistic approach.",
"We establish central and local limit theorems for the number of vertices in the largest component of a random d-uniform hypergraph Hd(n,p) with edge probability p = c- @math , where c > (d - 1)-1 is a constant. The proof relies on a new, purely probabilistic approach. © 2009 Wiley Periodicals, Inc. Random Struct. Alg., 2010",
""
]
} |
1708.07671 | 2749924182 | Let @math be the orientable surface of genus @math . We prove that the component structure of a graph chosen uniformly at random from the class @math of all graphs on vertex set @math with @math edges embeddable on @math features two phase transitions. The first phase transition mirrors the classical phase transition in the Erdős--Renyi random graph @math chosen uniformly at random from all graphs with vertex set @math and @math edges. It takes place at @math , when a unique largest component, the so-called , emerges. The second phase transition occurs at @math , when the giant component covers almost all vertices of the graph. This kind of phenomenon is strikingly different from @math and has only been observed for graphs on surfaces. Moreover, we derive an asymptotic estimation of the number of graphs in @math throughout the regimes of these two phase transitions. | Planar graphs and graphs embeddable on @math have been investigated separately for the sparse' regime @math @cite_49 and for the dense' regime @math with @math @cite_47 @cite_53 . From a random graph point of view, in particular when the giant component is considered, the sparse regime is the more interesting regime. In the sparse regime, Kang and uczak @cite_49 supplied new resourceful proof methods---some of which we apply in a somewhat similar fashion in this paper---combining probabilistic and graph theoretic methods with techniques from enumerative and analytic combinatorics. On the other hand, minor mistakes in @cite_49 led to results that featured order terms that claimed to be stronger than what has actually been proved. One contribution of this paper is to correct and strengthen these results from @cite_49 . | {
"cite_N": [
"@cite_47",
"@cite_49",
"@cite_53"
],
"mid": [
"",
"2057857650",
"2018165282"
],
"abstract": [
"",
"Let P (n;M) be a graph chosen uniformly at random from the family of all labeled planar graphs with n vertices and M edges. In the paper we study the component structure of P (n;M). Combining counting arguments with analytic techniques, we show that there are two critical periods in the evolution of P (n;M). The first one, of width ( n 2=3 ), is analogous to the phase transition observed in the standard random graph models and takes place for M = n=2 +O(n 2=3 ), when the largest complex component is formed. Then, for M = n + O(n 3=5 ), when the complex components covers nearly all vertices, the second critical period of width n 3=5 occurs. Starting from that moment increasing of M mostly affects the density of the complex components, not its size.",
"A graph is planar if it can be embedded in the plane, or in the sphere, so that no two edges cross at an interior point. A planar graph together with a particular embedding is called a map. There is a rich theory of counting maps, started by Tutte in the 1960's. However, in this paper we are interested in counting graphs as combinatorial objects, regardless of how many nonequivalent topological embeddings they may have. As we are going to see, this makes the counting considerably more difficult."
]
} |
1708.07239 | 2748606298 | The volume and velocity of information that gets generated online limits current journalistic practices to fact-check claims at the same rate. Computational approaches for fact checking may be the key to help mitigate the risks of massive misinformation spread. Such approaches can be designed to not only be scalable and effective at assessing veracity of dubious claims, but also to boost a human fact checker's productivity by surfacing relevant facts and patterns to aid their analysis. To this end, we present a novel, unsupervised network-flow based approach to determine the truthfulness of a statement of fact expressed in the form of a (subject, predicate, object) triple. We view a knowledge graph of background information about real-world entities as a flow network, and knowledge as a fluid, abstract commodity. We show that computational fact checking of such a triple then amounts to finding a "knowledge stream" that emanates from the subject node and flows toward the object node through paths connecting them. Evaluation on a range of real-world and hand-crafted datasets of facts related to entertainment, business, sports, geography and more reveals that this network-flow model can be very effective in discerning true statements from false ones, outperforming existing algorithms on many test cases. Moreover, the model is expressive in its ability to automatically discover several useful path patterns and surface relevant facts that may help a human fact checker corroborate or refute a claim. | Other approaches focus on checking the content of a claim based on prior knowledge, typically found in a knowledge base or knowledge graph. We can distinguish two broad classes of methods based on how easy it is to interpret their results. On the one hand, we have approaches inspired by logical reasoning (e.g., ILP @cite_27 and AMIE @cite_25 ), which mine first-order Horn clauses and are thus easy to interpret. On the other, there are statistical learning models (e.g., RESCAL @cite_6 , TransE @cite_30 , TransH @cite_9 , TransR @cite_26 , and ProjE @cite_28 ) that create vector embeddings for entities and relations, which can be used to assign similarity scores. Statistical approaches can be particularly hard to interpret, but they are great at handling uncertainties and capturing global regularities in the KG. Unfortunately, scalability is an issue for both types of approaches, as many of the algorithms mentioned above struggle to perform in the face of large-scale KGs, due to either large search spaces or high model complexity. Nickel @cite_6 review a number of these models. | {
"cite_N": [
"@cite_30",
"@cite_26",
"@cite_28",
"@cite_9",
"@cite_6",
"@cite_27",
"@cite_25"
],
"mid": [
"2127795553",
"2184957013",
"2556343638",
"2283196293",
"1529533208",
"",
"2151502664"
],
"abstract": [
"We consider the problem of embedding entities and relationships of multi-relational data in low-dimensional vector spaces. Our objective is to propose a canonical model which is easy to train, contains a reduced number of parameters and can scale up to very large databases. Hence, we propose TransE, a method which models relationships by interpreting them as translations operating on the low-dimensional embeddings of the entities. Despite its simplicity, this assumption proves to be powerful since extensive experiments show that TransE significantly outperforms state-of-the-art methods in link prediction on two knowledge bases. Besides, it can be successfully trained on a large scale data set with 1M entities, 25k relationships and more than 17M training samples.",
"Knowledge graph completion aims to perform link prediction between entities. In this paper, we consider the approach of knowledge graph embeddings. Recently, models such as TransE and TransH build entity and relation embeddings by regarding a relation as translation from head entity to tail entity. We note that these models simply put both entities and relations within the same semantic space. In fact, an entity may have multiple aspects and various relations may focus on different aspects of entities, which makes a common space insufficient for modeling. In this paper, we propose TransR to build entity and relation embeddings in separate entity space and relation spaces. Afterwards, we learn embeddings by first projecting entities from entity space to corresponding relation space and then building translations between projected entities. In experiments, we evaluate our models on three tasks including link prediction, triple classification and relational fact extraction. Experimental results show significant and consistent improvements compared to state-of-the-art baselines including TransE and TransH. The source code of this paper can be obtained from https: github.com mrlyk423 relation_extraction.",
"With the large volume of new information created every day, determining the validity of information in a knowledge graph and filling in its missing parts are crucial tasks for many researchers and practitioners. To address this challenge, a number of knowledge graph completion methods have been developed using low-dimensional graph embeddings. Although researchers continue to improve these models using an increasingly complex feature space, we show that simple changes in the architecture of the underlying model can outperform state-of-the-art models without the need for complex feature engineering. In this work, we present a shared variable neural network model called ProjE that fills-in missing information in a knowledge graph by learning joint embeddings of the knowledge graph's entities and edges, and through subtle, but important, changes to the standard loss function. In doing so, ProjE has a parameter size that is smaller than 11 out of 15 existing methods while performing @math better than the current-best method on standard datasets. We also show, via a new fact checking task, that ProjE is capable of accurately determining the veracity of many declarative statements.",
"We deal with embedding a large scale knowledge graph composed of entities and relations into a continuous vector space. TransE is a promising method proposed recently, which is very efficient while achieving state-of-the-art predictive performance. We discuss some mapping properties of relations which should be considered in embedding, such as reflexive, one-to-many, many-to-one, and many-to-many. We note that TransE does not do well in dealing with these properties. Some complex models are capable of preserving these mapping properties but sacrifice efficiency in the process. To make a good trade-off between model capacity and efficiency, in this paper we propose TransH which models a relation as a hyperplane together with a translation operation on it. In this way, we can well preserve the above mapping properties of relations with almost the same model complexity of TransE. Additionally, as a practical knowledge graph is often far from completed, how to construct negative examples to reduce false negative labels in training is very important. Utilizing the one-to-many many-to-one mapping property of a relation, we propose a simple trick to reduce the possibility of false negative labeling. We conduct extensive experiments on link prediction, triplet classification and fact extraction on benchmark datasets like WordNet and Freebase. Experiments show TransH delivers significant improvements over TransE on predictive accuracy with comparable capability to scale up.",
"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.",
"",
"Recent advances in information extraction have led to huge knowledge bases (KBs), which capture knowledge in a machine-readable format. Inductive Logic Programming (ILP) can be used to mine logical rules from the KB. These rules can help deduce and add missing knowledge to the KB. While ILP is a mature field, mining logical rules from KBs is different in two aspects: First, current rule mining systems are easily overwhelmed by the amount of data (state-of-the art systems cannot even run on today's KBs). Second, ILP usually requires counterexamples. KBs, however, implement the open world assumption (OWA), meaning that absent data cannot be used as counterexamples. In this paper, we develop a rule mining model that is explicitly tailored to support the OWA scenario. It is inspired by association rule mining and introduces a novel measure for confidence. Our extensive experiments show that our approach outperforms state-of-the-art approaches in terms of precision and coverage. Furthermore, our system, AMIE, mines rules orders of magnitude faster than state-of-the-art approaches."
]
} |
1708.07239 | 2748606298 | The volume and velocity of information that gets generated online limits current journalistic practices to fact-check claims at the same rate. Computational approaches for fact checking may be the key to help mitigate the risks of massive misinformation spread. Such approaches can be designed to not only be scalable and effective at assessing veracity of dubious claims, but also to boost a human fact checker's productivity by surfacing relevant facts and patterns to aid their analysis. To this end, we present a novel, unsupervised network-flow based approach to determine the truthfulness of a statement of fact expressed in the form of a (subject, predicate, object) triple. We view a knowledge graph of background information about real-world entities as a flow network, and knowledge as a fluid, abstract commodity. We show that computational fact checking of such a triple then amounts to finding a "knowledge stream" that emanates from the subject node and flows toward the object node through paths connecting them. Evaluation on a range of real-world and hand-crafted datasets of facts related to entertainment, business, sports, geography and more reveals that this network-flow model can be very effective in discerning true statements from false ones, outperforming existing algorithms on many test cases. Moreover, the model is expressive in its ability to automatically discover several useful path patterns and surface relevant facts that may help a human fact checker corroborate or refute a claim. | Only a few approaches fall somewhere in the middle of this interpretability spectrum. Ciampaglia @cite_0 propose an approach that relies on a single short, specific path to differentiate a true fact from a false one. Although intuitive, their algorithm fails to account for the semantics of the target predicate. | {
"cite_N": [
"@cite_0"
],
"mid": [
"1532503642"
],
"abstract": [
"Traditional fact checking by expert journalists cannot keep up with the enormous volume of information that is now generated online. Computational fact checking may significantly enhance our ability to evaluate the veracity of dubious information. Here we show that the complexities of human fact checking can be approximated quite well by finding the shortest path between concept nodes under properly defined semantic proximity metrics on knowledge graphs. Framed as a network problem this approach is feasible with efficient computational techniques. We evaluate this approach by examining tens of thousands of claims related to history, entertainment, geography, and biographical information using a public knowledge graph extracted from Wikipedia. Statements independently known to be true consistently receive higher support via our method than do false ones. These findings represent a significant step toward scalable computational fact-checking methods that may one day mitigate the spread of harmful misinformation."
]
} |
1708.07239 | 2748606298 | The volume and velocity of information that gets generated online limits current journalistic practices to fact-check claims at the same rate. Computational approaches for fact checking may be the key to help mitigate the risks of massive misinformation spread. Such approaches can be designed to not only be scalable and effective at assessing veracity of dubious claims, but also to boost a human fact checker's productivity by surfacing relevant facts and patterns to aid their analysis. To this end, we present a novel, unsupervised network-flow based approach to determine the truthfulness of a statement of fact expressed in the form of a (subject, predicate, object) triple. We view a knowledge graph of background information about real-world entities as a flow network, and knowledge as a fluid, abstract commodity. We show that computational fact checking of such a triple then amounts to finding a "knowledge stream" that emanates from the subject node and flows toward the object node through paths connecting them. Evaluation on a range of real-world and hand-crafted datasets of facts related to entertainment, business, sports, geography and more reveals that this network-flow model can be very effective in discerning true statements from false ones, outperforming existing algorithms on many test cases. Moreover, the model is expressive in its ability to automatically discover several useful path patterns and surface relevant facts that may help a human fact checker corroborate or refute a claim. | PRA @cite_4 and PredPath @cite_32 mine the KG in search of paths connecting the subject to the object of a triple, and use the predicate labels found along these paths to identify features for a supervised learning framework. Labeled examples of true and false triples are therefore needed at the stage of feature selection and during model training. Both approaches spend significant computational resources on feature generation and selection. And even though they rely on massive amounts of features, most provide only a very weak signal. Nevertheless, they have been shown to be very effective on fact-checking test cases @cite_32 , and in large-scale machine reading projects @cite_3 @cite_16 . They also offer some interpretability due to the features (or rules) they learn. Our methods achieve comparable or better performance while offering greater interpretability and expressiveness in terms of supporting facts, and without training --- except for learning'' the edge capacities (see sec:relational-similarity ). | {
"cite_N": [
"@cite_16",
"@cite_4",
"@cite_32",
"@cite_3"
],
"mid": [
"2016753842",
"2029249040",
"2339329611",
"1756422141"
],
"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.",
"Scientific literature with rich metadata can be represented as a labeled directed graph. This graph representation enables a number of scientific tasks such as ad hoc retrieval or named entity recognition (NER) to be formulated as typed proximity queries in the graph. One popular proximity measure is called Random Walk with Restart (RWR), and much work has been done on the supervised learning of RWR measures by associating each edge label with a parameter. In this paper, we describe a novel learnable proximity measure which instead uses one weight per edge label sequence: proximity is defined by a weighted combination of simple \"path experts\", each corresponding to following a particular sequence of labeled edges. Experiments on eight tasks in two subdomains of biology show that the new learning method significantly outperforms the RWR model (both trained and untrained). We also extend the method to support two additional types of experts to model intrinsic properties of entities: query-independent experts, which generalize the PageRank measure, and popular entity experts which allow rankings to be adjusted for particular entities that are especially important.",
"Traditional fact checking by experts and analysts cannot keep pace with the volume of newly created information. It is important and necessary, therefore, to enhance our ability to computationally determine whether some statement of fact is true or false. We view this problem as a link-prediction task in a knowledge graph, and present a discriminative path-based method for fact checking in knowledge graphs that incorporates connectivity, type information, and predicate interactions. Given a statement S of the form (subject, predicate, object), for example, (Chicago, capitalOf, Illinois), our approach mines discriminative paths that alternatively define the generalized statement (U.S. city, predicate, U.S. state) and uses the mined rules to evaluate the veracity of statement S . We evaluate our approach by examining thousands of claims related to history, geography, biology, and politics using a public, million node knowledge graph extracted from Wikipedia and PubMedDB. Not only does our approach significantly outperform related models, we also find that the discriminative predicate path model is easily interpretable and provides sensible reasons for the final determination.",
"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."
]
} |
1708.06839 | 2749679580 | We describe a new cardinality estimation algorithm that is extremely space-efficient. It applies one of three novel estimators to the compressed state of the Flajolet-Martin-85 coupon collection process. In an apples-to-apples empirical comparison against compressed HyperLogLog sketches, the new algorithm simultaneously wins on all three dimensions of the time space accuracy tradeoff. Our prototype uses the zstd compression library, and produces sketches that are smaller than the entropy of HLL, so no possible implementation of compressed HLL can match its space efficiency. The paper's technical contributions include analyses and simulations of the three new estimators, accurate values for the entropies of FM85 and HLL, and a non-trivial method for estimating a double asymptotic limit via simulation. | Aside from HLL @cite_13 , the other cardinality estimation sketch that is in widespread today is KMV @cite_6 @cite_19 @cite_4 @cite_9 @cite_7 , which achieves a Standard Error of @math by tracking the @math numerically smallest 64-bit hashes that have been seen so far. Although KMV sketches consume 10 times as much memory as HLL sketches, they are more accurate for set intersections, and it has been argued that KMV provides more operational flexibility in large-scale systems environments. | {
"cite_N": [
"@cite_4",
"@cite_7",
"@cite_9",
"@cite_6",
"@cite_19",
"@cite_13"
],
"mid": [
"1979819093",
"2005731313",
"2963699114",
"1785933978",
"2083966857",
"2144982963"
],
"abstract": [
"A Bottom-sketch is a summary of a set of items with nonnegative weights that supports approximate query processing. A sketch is obtained by associating with each item in a ground set an independent random rank drawn from a probability distribution that depends on the weight of the item and including the k items with smallest rank value. Bottom-k sketches are an alternative to k-mins sketches[9], which consist of the k minimum ranked items in k independent rank assignments,and of min-hash [5] sketches, where hash functions replace random rank assignments. Sketches support approximate aggregations, including weight and selectivity of a subpopulation. Coordinated sketches of multiple subsets over the same ground set support subset-relation queries such as Jaccard similarity or the weight of the union. All-distances sketches are applicable for datasets where items lie in some metric space such as data streams (time) or networks. These sketches compactly encode the respective plain sketches of all neighborhoods of a location. These sketches support queries posed over time windows or neighborhoods and time spatially decaying aggregates. An important advantage of bottom-k sketches, established in a line of recent work, is much tighter estimators for several basic aggregates. To materialize this benefit, we must adapt traditional k-mins applications to use bottom-k sketches. We propose all-distances bottom-k sketches and develop and analyze data structures that incrementally construct bottom-k sketches and all-distances bottom-k sketches. Another advantage of bottom-k sketches is that when the data is represented explicitly, they can be obtained much more efficiently than k-mins sketches. We show that k-mins sketches can be derived from respective bottom-k sketches, which enables the use of bottom-k sketches with off-the-shelf k-mins estimators. (In fact, we obtain tighter estimators since each bottom-k sketch is adistribution over k-mins sketches).",
"A new class of algorithms to estimate the cardinality of very large multisets using constant memory and doing only one pass on the data is introduced here. It is based on order statistics rather than on bit patterns in binary representations of numbers. Three families of estimators are analyzed. They attain a standard error of 1M using M units of storage, which places them in the same class as the best known algorithms so far. The algorithms have a very simple internal loop, which gives them an advantage in terms of processing speed. For instance, a memory of only 12 kB and only few seconds are sufficient to process a multiset with several million elements and to build an estimate with accuracy of order 2 percent. The algorithms are validated both by mathematical analysis and by experimentations on real internet traffic.",
"Given m distributed data streams A_1,..., A_m, we consider the problem of estimating the number of unique identifiers in streams defined by set expressions over A_1,..., A_m. We identify a broad class of algorithms for solving this problem, and show that the estimators output by any algorithm in this class are perfectly unbiased and satisfy strong variance bounds. Our analysis unifies and generalizes a variety of earlier results in the literature. To demonstrate its generality, we describe several novel sampling algorithms in our class, and show that they achieve a novel tradeoff between accuracy, space usage, update speed, and applicability.",
"We present three algorithms to count the number of distinct elements in a data stream to within a factor of 1 ± ?. Our algorithms improve upon known algorithms for this problem, and offer a spectrum of time space tradeoffs.",
"The task of estimating the number of distinct values (DVs) in a large dataset arises in a wide variety of settings in computer science and elsewhere. We provide DV estimation techniques for the case in which the dataset of interest is split into partitions. We create for each partition a synopsis that can be used to estimate the number of DVs in the partition. By combining and extending a number of results in the literature, we obtain both suitable synopses and DV estimators. The synopses can be created in parallel, and can be easily combined to yield synopses and DV estimates for \"compound\" partitions that are created from the base partitions via arbitrary multiset union, intersection, or difference operations. Our synopses can also handle deletions of individual partition elements. We prove that our DV estimators are unbiased, provide error bounds, and show how to select synopsis sizes in order to achieve a desired estimation accuracy. Experiments and theory indicate that our synopses and estimators lead to lower computational costs and more accurate DV estimates than previous approaches.",
"This extended abstract describes and analyses a near-optimal probabilistic algorithm, HYPERLOGLOG, dedicated to estimating the number of phdistinct elements (the cardinality) of very large data ensembles. Using an auxiliary memory of m units (typically, \"short bytes''), HYPERLOGLOG performs a single pass over the data and produces an estimate of the cardinality such that the relative accuracy (the standard error) is typically about @math . This improves on the best previously known cardinality estimator, LOGLOG, whose accuracy can be matched by consuming only 64 of the original memory. For instance, the new algorithm makes it possible to estimate cardinalities well beyond @math with a typical accuracy of 2 while using a memory of only 1.5 kilobytes. The algorithm parallelizes optimally and adapts to the sliding window model."
]
} |
1708.06839 | 2749679580 | We describe a new cardinality estimation algorithm that is extremely space-efficient. It applies one of three novel estimators to the compressed state of the Flajolet-Martin-85 coupon collection process. In an apples-to-apples empirical comparison against compressed HyperLogLog sketches, the new algorithm simultaneously wins on all three dimensions of the time space accuracy tradeoff. Our prototype uses the zstd compression library, and produces sketches that are smaller than the entropy of HLL, so no possible implementation of compressed HLL can match its space efficiency. The paper's technical contributions include analyses and simulations of the three new estimators, accurate values for the entropies of FM85 and HLL, and a non-trivial method for estimating a double asymptotic limit via simulation. | The HLL family of sketches began with the FM85 paper @cite_1 , which proposed an estimator based on the average position of the leftmost uncollected coupon in each row. Because the sketch required @math words of memory, the authors speculated that a lossy 8-column sliding window into the coupon tableau might suffice. With the benefit of hindsight we know that this specific idea didn't pan out, but this shows that at the very beginning of the field it was understood that sketches can be smaller than naive implementations would suggest. | {
"cite_N": [
"@cite_1"
],
"mid": [
"2025051251"
],
"abstract": [
"This paper introduces a class of probabilistic counting algorithms with which one can estimate the number of distinct elements in a large collection of data (typically a large file stored on disk) in a single pass using only a small additional storage (typically less than a hundred binary words) and only a few operations per element scanned. The algorithms are based on statistical observations made on bits of hashed values of records. They are by construction totally insensitive to the replicative structure of elements in the file; they can be used in the context of distributed systems without any degradation of performances and prove especially useful in the context of data bases query optimisation."
]
} |
1708.06839 | 2749679580 | We describe a new cardinality estimation algorithm that is extremely space-efficient. It applies one of three novel estimators to the compressed state of the Flajolet-Martin-85 coupon collection process. In an apples-to-apples empirical comparison against compressed HyperLogLog sketches, the new algorithm simultaneously wins on all three dimensions of the time space accuracy tradeoff. Our prototype uses the zstd compression library, and produces sketches that are smaller than the entropy of HLL, so no possible implementation of compressed HLL can match its space efficiency. The paper's technical contributions include analyses and simulations of the three new estimators, accurate values for the entropies of FM85 and HLL, and a non-trivial method for estimating a double asymptotic limit via simulation. | In the LogLog'' papers that followed, the authors switched to estimators based on the average position of the rightmost collected coupon in each row, allowing the sketch to shrink to @math bits. The estimator in the HyperLogLog paper @cite_13 employed a harmonic average, and spliced in the bitmap estimator from @cite_3 for the small- @math regime. There was still a spike in error at the crossover point, which the HLL++ @cite_11 implementation reduced to a small bump by applying an empirical bias correction. @cite_12 showed that a Maximum Likelihood estimator yields a similar error curve which lacks the bump [compare the blue curves in top two plots of the current paper's Figure .] | {
"cite_N": [
"@cite_13",
"@cite_12",
"@cite_3",
"@cite_11"
],
"mid": [
"2144982963",
"2593397259",
"2008365755",
"1982092405"
],
"abstract": [
"This extended abstract describes and analyses a near-optimal probabilistic algorithm, HYPERLOGLOG, dedicated to estimating the number of phdistinct elements (the cardinality) of very large data ensembles. Using an auxiliary memory of m units (typically, \"short bytes''), HYPERLOGLOG performs a single pass over the data and produces an estimate of the cardinality such that the relative accuracy (the standard error) is typically about @math . This improves on the best previously known cardinality estimator, LOGLOG, whose accuracy can be matched by consuming only 64 of the original memory. For instance, the new algorithm makes it possible to estimate cardinalities well beyond @math with a typical accuracy of 2 while using a memory of only 1.5 kilobytes. The algorithm parallelizes optimally and adapts to the sliding window model.",
"This paper presents new methods to estimate the cardinalities of data sets recorded by HyperLogLog sketches. A theoretically motivated extension to the original estimator is presented that eliminates the bias for small and large cardinalities. Based on the maximum likelihood principle a second unbiased method is derived together with a robust and efficient numerical algorithm to calculate the estimate. The maximum likelihood approach can also be applied to more than a single HyperLogLog sketch. In particular, it is shown that it gives more precise cardinality estimates for union, intersection, or relative complements of two sets that are both represented by HyperLogLog sketches compared to the conventional technique using the inclusion-exclusion principle. All the new methods are demonstrated and verified by extensive simulations.",
"We present a probabilistic algorithm for counting the number of unique values in the presence of duplicates. This algorithm has O ( q ) time complexity, where q is the number of values including duplicates, and produces an estimation with an arbitrary accuracy prespecified by the user using only a small amount of space. Traditionally, accurate counts of unique values were obtained by sorting, which has O ( q log q ) time complexity. Our technique, called linear counting , is based on hashing. We present a comprehensive theoretical and experimental analysis of linear counting. The analysis reveals an interesting result: A load factor (number of unique values hash table size) much larger than 1.0 (e.g., 12) can be used for accurate estimation (e.g., 1 of error). We present this technique with two important applications to database problems: namely, (1) obtaining the column cardinality (the number of unique values in a column of a relation) and (2) obtaining the join selectivity (the number of unique values in the join column resulting from an unconditional join divided by the number of unique join column values in the relation to he joined). These two parameters are important statistics that are used in relational query optimization and physical database design.",
"Cardinality estimation has a wide range of applications and is of particular importance in database systems. Various algorithms have been proposed in the past, and the HyperLogLog algorithm is one of them. In this paper, we present a series of improvements to this algorithm that reduce its memory requirements and significantly increase its accuracy for an important range of cardinalities. We have implemented our proposed algorithm for a system at Google and evaluated it empirically, comparing it to the original HyperLogLog algorithm. Like HyperLogLog, our improved algorithm parallelizes perfectly and computes the cardinality estimate in a single pass."
]
} |
1708.06839 | 2749679580 | We describe a new cardinality estimation algorithm that is extremely space-efficient. It applies one of three novel estimators to the compressed state of the Flajolet-Martin-85 coupon collection process. In an apples-to-apples empirical comparison against compressed HyperLogLog sketches, the new algorithm simultaneously wins on all three dimensions of the time space accuracy tradeoff. Our prototype uses the zstd compression library, and produces sketches that are smaller than the entropy of HLL, so no possible implementation of compressed HLL can match its space efficiency. The paper's technical contributions include analyses and simulations of the three new estimators, accurate values for the entropies of FM85 and HLL, and a non-trivial method for estimating a double asymptotic limit via simulation. | HIP estimators were discovered independently by @cite_18 and @cite_16 , and both papers used HLL sketches to illustrate how this idea can be applied. | {
"cite_N": [
"@cite_18",
"@cite_16"
],
"mid": [
"2090914728",
"2002231822"
],
"abstract": [
"Graph datasets with billions of edges, such as social and Web graphs, are prevalent. To be feasible, computation on such large graphs should scale linearly with graph size. All-distances sketches (ADSs) are emerging as a powerful tool for scalable computation of some basic properties of individual nodes or the whole graph. ADSs were first proposed two decades ago (Cohen 1994) and more recent algorithms include ANF (Palmer, Gibbons, and Faloutsos 2002) and hyperANF (Boldi, Rosa, and Vigna 2011). A sketch of logarithmic size is computed for each node in the graph and the computation in total requires only a near linear number of edge relaxations. From the ADS of a node, we can estimate neighborhood cardinalities (the number of nodes within some query distance) and closeness centrality. More generally we can estimate the distance distribution, effective diameter, similarities, and other parameters of the full graph. We make several contributions which facilitate a more effective use of ADSs for scalable analysis of massive graphs. We provide, for the first time, a unified exposition of ADS algorithms and applications. We present the Historic Inverse Probability (HIP) estimators which are applied to the ADS of a node to estimate a large natural class of queries including neighborhood cardinalities and closeness centralities. We show that our HIP estimators have at most half the variance of previous neighborhood cardinality estimators and that this is essentially optimal. Moreover, HIP obtains a polynomial improvement over state of the art for more general domain queries and the estimators are simple, flexible, unbiased, and elegant. The ADS generalizes Min-Hash sketches, used for approximating cardinality (distinct count) on data streams. We obtain lower bounds on Min-Hash cardinality estimation using classic estimation theory. We illustrate the power of HIP, both in terms of ease of application and estimation quality, by comparing it to the HyperLogLog algorithm ( 2007), demonstrating a significant improvement over this state-of-the-art practical algorithm. We also study the quality of ADS estimation of distance ranges, generalizing the near-linear time factor-2 approximation of the diameter.",
"Counting the number of distinct elements in a large dataset is a common task in web applications and databases. This problem is difficult in limited memory settings where storing a large hash table table is intractable. This paper advances the state of the art in probabilistic methods for estimating the number of distinct elements in a streaming setting New streaming algorithms are given that provably beat the \"optimal\" errors for Min-count and HyperLogLog while using the same sketch. This paper also contributes to the understanding and theory of probabilistic cardinality estimation introducing the concept of an area cutting process and the martingale estimator. These ideas lead to theoretical analyses of both old and new sketches and estimators and show the new estimators are optimal for several streaming settings while also providing accurate error bounds that match those obtained via simulation. Furthermore, the area cutting process provides a geometric intuition behind all methods for counting distinct elements which are not affected by duplicates. This intuition leads to a new sketch, Discrete Max-count, and the analysis of a class of sketches, self-similar area cutting decompositions that have attractive properties and unbiased estimators for both streaming and non-streaming settings. Together, these contributions lead to multi-faceted advances in sketch construction, cardinality and error estimation, the theory, and intuition for the problem of approximate counting of distinct elements for both the streaming and non-streaming cases."
]
} |
1708.06839 | 2749679580 | We describe a new cardinality estimation algorithm that is extremely space-efficient. It applies one of three novel estimators to the compressed state of the Flajolet-Martin-85 coupon collection process. In an apples-to-apples empirical comparison against compressed HyperLogLog sketches, the new algorithm simultaneously wins on all three dimensions of the time space accuracy tradeoff. Our prototype uses the zstd compression library, and produces sketches that are smaller than the entropy of HLL, so no possible implementation of compressed HLL can match its space efficiency. The paper's technical contributions include analyses and simulations of the three new estimators, accurate values for the entropies of FM85 and HLL, and a non-trivial method for estimating a double asymptotic limit via simulation. | Finally, the KNW sketch @cite_8 is space-optimal in a theoretical sense, but the constant factors are unknown, and to the best of our knowledge it has never been used in a real system. | {
"cite_N": [
"@cite_8"
],
"mid": [
"2103126020"
],
"abstract": [
"We give the first optimal algorithm for estimating the number of distinct elements in a data stream, closing a long line of theoretical research on this problem begun by Flajolet and Martin in their seminal paper in FOCS 1983. This problem has applications to query optimization, Internet routing, network topology, and data mining. For a stream of indices in 1,...,n , our algorithm computes a (1 ± e)-approximation using an optimal O(1 e-2 + log(n)) bits of space with 2 3 success probability, where 0 We also give an algorithm to estimate the Hamming norm of a stream, a generalization of the number of distinct elements, which is useful in data cleaning, packet tracing, and database auditing. Our algorithm uses nearly optimal space, and has optimal O(1) update and reporting times."
]
} |
1708.06977 | 2747206364 | Despite their success for object detection, convolutional neural networks are ill-equipped for incremental learning, i.e., adapting the original model trained on a set of classes to additionally detect objects of new classes, in the absence of the initial training data. They suffer from "catastrophic forgetting" - an abrupt degradation of performance on the original set of classes, when the training objective is adapted to the new classes. We present a method to address this issue, and learn object detectors incrementally, when neither the original training data nor annotations for the original classes in the new training set are available. The core of our proposed solution is a loss function to balance the interplay between predictions on the new classes and a new distillation loss which minimizes the discrepancy between responses for old classes from the original and the updated networks. This incremental learning can be performed multiple times, for a new set of classes in each step, with a moderate drop in performance compared to the baseline network trained on the ensemble of data. We present object detection results on the PASCAL VOC 2007 and COCO datasets, along with a detailed empirical analysis of the approach. | An alternative to transfer knowledge from one network to another is distillation @cite_4 @cite_37 . This was originally proposed to transfer knowledge between different neural networks---from a large network to a smaller one for efficient deployment. The method in @cite_4 encouraged the large (old) and the small (new) networks to produce similar responses. It has found several applications in domain adaptation and model compression @cite_18 @cite_40 @cite_30 . Overall, transfer learning and domain adaptation methods require at least unlabeled data for both the tasks or domains, and in its absence, the new network quickly forgets all the knowledge acquired in the source domain @cite_28 @cite_17 @cite_13 @cite_38 . In contrast, our approach addresses the challenging case where no training data is available for the original task (i.e., detecting objects belonging to the original classes), by building on the concept of knowledge distillation @cite_4 . | {
"cite_N": [
"@cite_30",
"@cite_38",
"@cite_37",
"@cite_18",
"@cite_4",
"@cite_28",
"@cite_40",
"@cite_13",
"@cite_17"
],
"mid": [
"2953226914",
"2113839990",
"",
"2951874610",
"1821462560",
"2060277733",
"",
"2036963181",
""
],
"abstract": [
"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.",
"Catastrophic forgetting is a problem faced by many machine learning models and algorithms. When trained on one task, then trained on a second task, many machine learning models \"forget\" how to perform the first task. This is widely believed to be a serious problem for neural networks. Here, we investigate the extent to which the catastrophic forgetting problem occurs for modern neural networks, comparing both established and recent gradient-based training algorithms and activation functions. We also examine the effect of the relationship between the first task and the second task on catastrophic forgetting. We find that it is always best to train using the dropout algorithm--the dropout algorithm is consistently best at adapting to the new task, remembering the old task, and has the best tradeoff curve between these two extremes. We find that different tasks and relationships between tasks result in very different rankings of activation function performance. This suggests the choice of activation function should always be cross-validated.",
"",
"In this work we propose a technique that transfers supervision between images from different modalities. We use learned representations from a large labeled modality as a supervisory signal for training representations for a new unlabeled paired modality. Our method enables learning of rich representations for unlabeled modalities and can be used as a pre-training procedure for new modalities with limited labeled data. We show experimental results where we transfer supervision from labeled RGB images to unlabeled depth and optical flow images and demonstrate large improvements for both these cross modal supervision transfers. Code, data and pre-trained models are available at this https URL",
"A very simple way to improve the performance of almost any machine learning algorithm is to train many different models on the same data and then to average their predictions. Unfortunately, making predictions using a whole ensemble of models is cumbersome and may be too computationally expensive to allow deployment to a large number of users, especially if the individual models are large neural nets. Caruana and his collaborators have shown that it is possible to compress the knowledge in an ensemble into a single model which is much easier to deploy and we develop this approach further using a different compression technique. We achieve some surprising results on MNIST and we show that we can significantly improve the acoustic model of a heavily used commercial system by distilling the knowledge in an ensemble of models into a single model. We also introduce a new type of ensemble composed of one or more full models and many specialist models which learn to distinguish fine-grained classes that the full models confuse. Unlike a mixture of experts, these specialist models can be trained rapidly and in parallel.",
"Abstract All natural cognitive systems, and, in particular, our own, gradually forget previously learned information. Plausible models of human cognition should therefore exhibit similar patterns of gradual forgetting of old information as new information is acquired. Only rarely does new learning in natural cognitive systems completely disrupt or erase previously learned information; that is, natural cognitive systems do not, in general, forget ‘catastrophically'. Unfortunately, though, catastrophic forgetting does occur under certain circumstances in distributed connectionist networks. The very features that give these networks their remarkable abilities to generalize, to function in the presence of degraded input, and so on, are found to be the root cause of catastrophic forgetting. The challenge in this field is to discover how to keep the advantages of distributed connectionist networks while avoiding the problem of catastrophic forgetting. In this article the causes, consequences and numerous solutions to the problem of catastrophic forgetting in neural networks are examined. The review will consider how the brain might have overcome this problem and will also explore the consequences of this solution for distributed connectionist networks.",
"",
"Multilayer connectionist models of memory based on the encoder model using the backpropagation learning rule are evaluated. The models are applied to standard recognition memory procedures in which items are studied sequentially and then tested for retention. Sequential learning in these models leads to 2 major problems. First, well-learned information is forgotten rapidly as new information is learned. Second, discrimination between studied items and new items either decreases or is nonmonotonic as a function of learning. To address these problems, manipulations of the network within the multilayer model and several variants of the multilayer model were examined, including a model with prelearned memory and a context model, but none solved the problems. The problems discussed provide limitations on connectionist models applied to human memory and in tasks where information to be learned is not all available during learning. The first stage of the connectionist revolution in psychology is reaching maturity and perhaps drawing to an end. This stage has been concerned with the exploration of classes of models, and the criteria that have been used to evaluate the success of an application have been necessarily loose. In the early stages of development of a new approach, lax acceptability criteria are appropriate because of the large range of models to be examined. However, there comes a second stage when the models serve as competitors to existing models developed within other theoretical frameworks, and they have to be competitively evaluated according to more stringent criteria. A few notable connectionist models have reached these standards, whereas others have not. The second stage of development also requires that the connectionist models be evaluated in areas where their potential for success is not immediately obvious. One such area is recognition memory. The work presented in this article evaluates several variants of the multilayer connectionist model as accounts of empirical results in this area. I mainly discuss multilayer models using the error-correcting backpropagation algorithm and do not address other architectures such as adaptive resonance schemes (Carpenter & Grossberg, 1987). Before launching into the modeling of recognition memory, I need to specify the aims and rules under which this project was carried out. This is important in a new area of inquiry because there are many divergent views about what needs to be",
""
]
} |
1708.06977 | 2747206364 | Despite their success for object detection, convolutional neural networks are ill-equipped for incremental learning, i.e., adapting the original model trained on a set of classes to additionally detect objects of new classes, in the absence of the initial training data. They suffer from "catastrophic forgetting" - an abrupt degradation of performance on the original set of classes, when the training objective is adapted to the new classes. We present a method to address this issue, and learn object detectors incrementally, when neither the original training data nor annotations for the original classes in the new training set are available. The core of our proposed solution is a loss function to balance the interplay between predictions on the new classes and a new distillation loss which minimizes the discrepancy between responses for old classes from the original and the updated networks. This incremental learning can be performed multiple times, for a new set of classes in each step, with a moderate drop in performance compared to the baseline network trained on the ensemble of data. We present object detection results on the PASCAL VOC 2007 and COCO datasets, along with a detailed empirical analysis of the approach. | This phenomenon of forgetting is believed to be caused by two factors @cite_26 @cite_32 . First, the internal representations in hidden layers are often overlapping, and a small change in a single neuron can affect multiple representations at the same time @cite_26 . Second, all the parameters in feedforward networks are involved in computations for every data point, and a backpropagation update affects all of them in each training step @cite_32 . The problem of addressing these issues in neural networks has its origin in classical connectionist networks several years ago @cite_28 @cite_17 @cite_26 @cite_11 @cite_14 , but needs to be adapted to today's large deep neural network architectures for vision tasks @cite_35 @cite_6 . | {
"cite_N": [
"@cite_35",
"@cite_26",
"@cite_14",
"@cite_28",
"@cite_32",
"@cite_17",
"@cite_6",
"@cite_11"
],
"mid": [
"2949808626",
"2100602534",
"2169969772",
"2060277733",
"2147800946",
"",
"",
""
],
"abstract": [
"When building a unified vision system or gradually adding new capabilities to a system, the usual assumption is that training data for all tasks is always available. However, as the number of tasks grows, storing and retraining on such data becomes infeasible. A new problem arises where we add new capabilities to a Convolutional Neural Network (CNN), but the training data for its existing capabilities are unavailable. We propose our Learning without Forgetting method, which uses only new task data to train the network while preserving the original capabilities. Our method performs favorably compared to commonly used feature extraction and fine-tuning adaption techniques and performs similarly to multitask learning that uses original task data we assume unavailable. A more surprising observation is that Learning without Forgetting may be able to replace fine-tuning with similar old and new task datasets for improved new task performance.",
"It is well known that when a connectionist network is trained on one set of patterns and then attempts to add new patterns to its repertoire, catastrophic interference may result. The use of sparse, orthogonal hidden-layer representations has been shown to reduce catastrophic interference. The author demonstrates that the use of sparse representations may, in certain cases, actually result in worse performance on catastrophic interference. This paper argues for the necessity of maintaining hidden-layer representations that are both as highly distributed and as highly orthogonal as possible. The author presents a learning algorithm, called context-biasing, that dynamically solves the problem of constraining hiddenlayer representations to simultaneously produce good orthogonality and distributedness. On the data tested for this study, context-biasing is shown to reduce catastrophic interference by more than 50 compared to standard backpropagation. In particular, this technique succeeds in reducing catastrophic interference on data where sparse, orthogonal distributions failed to produce any improvement.",
"While humans forget gradually, highly distributed connectionist networks forget catastrophically: newly learned information often completely erases previously learned information. This is not just implausible cognitively, but disastrous practically. However, it is not easy in connectionist cognitive modelling to keep away from highly distributed neural networks, if only because of their ability to generalize. A realistic and effective system that solves the problem of catastrophic interference in sequential learning of ‘static’ (i.e. non-temporally ordered) patterns has been proposed recently (Robins 1995, Connection Science, 7: 123–146, 1996, Connection Science, 8: 259–275, Ans and Rousset 1997, CR Academie des Sciences Paris, Life Sciences, 320: 989–997, French 1997, Connection Science, 9: 353–379, 1999, Trends in Cognitive Sciences, 3: 128–135, Ans and Rousset 2000, Connection Science, 12: 1–19). The basic principle is to learn new external patterns interleaved with internally generated ‘pseudopatterns...",
"Abstract All natural cognitive systems, and, in particular, our own, gradually forget previously learned information. Plausible models of human cognition should therefore exhibit similar patterns of gradual forgetting of old information as new information is acquired. Only rarely does new learning in natural cognitive systems completely disrupt or erase previously learned information; that is, natural cognitive systems do not, in general, forget ‘catastrophically'. Unfortunately, though, catastrophic forgetting does occur under certain circumstances in distributed connectionist networks. The very features that give these networks their remarkable abilities to generalize, to function in the presence of degraded input, and so on, are found to be the root cause of catastrophic forgetting. The challenge in this field is to discover how to keep the advantages of distributed connectionist networks while avoiding the problem of catastrophic forgetting. In this article the causes, consequences and numerous solutions to the problem of catastrophic forgetting in neural networks are examined. The review will consider how the brain might have overcome this problem and will also explore the consequences of this solution for distributed connectionist networks.",
"The ability of learning networks to generalize can be greatly enhanced by providing constraints from the task domain. This paper demonstrates how such constraints can be integrated into a backpropagation network through the architecture of the network. This approach has been successfully applied to the recognition of handwritten zip code digits provided by the U.S. Postal Service. A single network learns the entire recognition operation, going from the normalized image of the character to the final classification.",
"",
"",
""
]
} |
1708.06977 | 2747206364 | Despite their success for object detection, convolutional neural networks are ill-equipped for incremental learning, i.e., adapting the original model trained on a set of classes to additionally detect objects of new classes, in the absence of the initial training data. They suffer from "catastrophic forgetting" - an abrupt degradation of performance on the original set of classes, when the training objective is adapted to the new classes. We present a method to address this issue, and learn object detectors incrementally, when neither the original training data nor annotations for the original classes in the new training set are available. The core of our proposed solution is a loss function to balance the interplay between predictions on the new classes and a new distillation loss which minimizes the discrepancy between responses for old classes from the original and the updated networks. This incremental learning can be performed multiple times, for a new set of classes in each step, with a moderate drop in performance compared to the baseline network trained on the ensemble of data. We present object detection results on the PASCAL VOC 2007 and COCO datasets, along with a detailed empirical analysis of the approach. | Li and Hoiem @cite_35 use knowledge distillation for one of the classical vision tasks, image classification, formulated in a deep learning framework. However, their evaluation is limited to the case where the old network is trained on a dataset, while the new network is trained on a different one, e.g., Places365 for the old and PASCAL VOC for the new, ImageNet for the old and PASCAL VOC for the new, etc. While this is interesting, it is a simpler task, because: (i) different datasets often contain dissimilar classes, (ii) there is little confusion between datasets---it is in fact possible to identify a dataset simply from an image @cite_25 . | {
"cite_N": [
"@cite_35",
"@cite_25"
],
"mid": [
"2949808626",
"2031342017"
],
"abstract": [
"When building a unified vision system or gradually adding new capabilities to a system, the usual assumption is that training data for all tasks is always available. However, as the number of tasks grows, storing and retraining on such data becomes infeasible. A new problem arises where we add new capabilities to a Convolutional Neural Network (CNN), but the training data for its existing capabilities are unavailable. We propose our Learning without Forgetting method, which uses only new task data to train the network while preserving the original capabilities. Our method performs favorably compared to commonly used feature extraction and fine-tuning adaption techniques and performs similarly to multitask learning that uses original task data we assume unavailable. A more surprising observation is that Learning without Forgetting may be able to replace fine-tuning with similar old and new task datasets for improved new task performance.",
"Datasets are an integral part of contemporary object recognition research. They have been the chief reason for the considerable progress in the field, not just as source of large amounts of training data, but also as means of measuring and comparing performance of competing algorithms. At the same time, datasets have often been blamed for narrowing the focus of object recognition research, reducing it to a single benchmark performance number. Indeed, some datasets, that started out as data capture efforts aimed at representing the visual world, have become closed worlds unto themselves (e.g. the Corel world, the Caltech-101 world, the PASCAL VOC world). With the focus on beating the latest benchmark numbers on the latest dataset, have we perhaps lost sight of the original purpose? The goal of this paper is to take stock of the current state of recognition datasets. We present a comparison study using a set of popular datasets, evaluated based on a number of criteria including: relative data bias, cross-dataset generalization, effects of closed-world assumption, and sample value. The experimental results, some rather surprising, suggest directions that can improve dataset collection as well as algorithm evaluation protocols. But more broadly, the hope is to stimulate discussion in the community regarding this very important, but largely neglected issue."
]
} |
1708.06977 | 2747206364 | Despite their success for object detection, convolutional neural networks are ill-equipped for incremental learning, i.e., adapting the original model trained on a set of classes to additionally detect objects of new classes, in the absence of the initial training data. They suffer from "catastrophic forgetting" - an abrupt degradation of performance on the original set of classes, when the training objective is adapted to the new classes. We present a method to address this issue, and learn object detectors incrementally, when neither the original training data nor annotations for the original classes in the new training set are available. The core of our proposed solution is a loss function to balance the interplay between predictions on the new classes and a new distillation loss which minimizes the discrepancy between responses for old classes from the original and the updated networks. This incremental learning can be performed multiple times, for a new set of classes in each step, with a moderate drop in performance compared to the baseline network trained on the ensemble of data. We present object detection results on the PASCAL VOC 2007 and COCO datasets, along with a detailed empirical analysis of the approach. | Very recently, Rebuffi al @cite_6 address some of the drawbacks in @cite_35 with their incremental learning approach for image classification. They also use knowledge distillation, but decouple the classifier and the representation learning. Additionally, they rely on a subset of the original training data to preserve the performance on the old classes. In comparison, our approach is an end-to-end learning framework, where the representation and the classifier are learned jointly, and we do not use any of the original training samples to avoid catastrophic forgetting. Alternatives to distillation are: growing the capacity of the network with new layers @cite_19 , applying strong per-parameter regularization selectively @cite_33 . The downside to these methods is the rapid increase in the number of new parameters to be learned @cite_19 , and their limited evaluation on the easier task of image classification @cite_33 . | {
"cite_N": [
"@cite_19",
"@cite_35",
"@cite_33",
"@cite_6"
],
"mid": [
"2426267443",
"2949808626",
"2560647685",
""
],
"abstract": [
"Methods and systems for performing a sequence of machine learning tasks. One system includes a sequence of deep neural networks (DNNs), including: a first DNN corresponding to a first machine learning task, wherein the first DNN comprises a first plurality of indexed layers, and each layer in the first plurality of indexed layers is configured to receive a respective layer input and process the layer input to generate a respective layer output; and one or more subsequent DNNs corresponding to one or more respective machine learning tasks, wherein each subsequent DNN comprises a respective plurality of indexed layers, and each layer in a respective plurality of indexed layers with index greater than one receives input from a preceding layer of the respective subsequent DNN, and one or more preceding layers of respective preceding DNNs, wherein a preceding layer is a layer whose index is one less than the current index.",
"When building a unified vision system or gradually adding new capabilities to a system, the usual assumption is that training data for all tasks is always available. However, as the number of tasks grows, storing and retraining on such data becomes infeasible. A new problem arises where we add new capabilities to a Convolutional Neural Network (CNN), but the training data for its existing capabilities are unavailable. We propose our Learning without Forgetting method, which uses only new task data to train the network while preserving the original capabilities. Our method performs favorably compared to commonly used feature extraction and fine-tuning adaption techniques and performs similarly to multitask learning that uses original task data we assume unavailable. A more surprising observation is that Learning without Forgetting may be able to replace fine-tuning with similar old and new task datasets for improved new task performance.",
"Abstract The ability to learn tasks in a sequential fashion is crucial to the development of artificial intelligence. Until now neural networks have not been capable of this and it has been widely thought that catastrophic forgetting is an inevitable feature of connectionist models. We show that it is possible to overcome this limitation and train networks that can maintain expertise on tasks that they have not experienced for a long time. Our approach remembers old tasks by selectively slowing down learning on the weights important for those tasks. We demonstrate our approach is scalable and effective by solving a set of classification tasks based on a hand-written digit dataset and by learning several Atari 2600 games sequentially.",
""
]
} |
1708.07104 | 2749784378 | The proliferation of misleading information in everyday access media outlets such as social media feeds, news blogs, and online newspapers have made it challenging to identify trustworthy news sources, thus increasing the need for computational tools able to provide insights into the reliability of online content. In this paper, we focus on the automatic identification of fake content in online news. Our contribution is twofold. First, we introduce two novel datasets for the task of fake news detection, covering seven different news domains. We describe the collection, annotation, and validation process in detail and present several exploratory analysis on the identification of linguistic differences in fake and legitimate news content. Second, we conduct a set of learning experiments to build accurate fake news detectors. In addition, we provide comparative analyses of the automatic and manual identification of fake news. | To date, there are three important lines of research into the automated classification of genuine and fake news items. First, on a conceptual level, a distinction has been made between 'three types of fake news' @cite_11 : serious fabrications (i.e. news items about false and non-existing events or information such as celebrity gossip), hoaxes (i.e. providing false information via, for example, social media with the intention to be picked up by traditional news websites) and satire (i.e. humorous news items that mimic genuine news but contain irony and absurdity). Here, we focus on the first category, serious fabrication, in the two domains of general news (in six different categories), as well as on celebrity gossip. | {
"cite_N": [
"@cite_11"
],
"mid": [
"2252350410"
],
"abstract": [
"A fake news detection system aims to assist users in detecting and filtering out varieties of potentially deceptive news. The prediction of the chances that a particular news item is intentionally deceptive is based on the analysis of previously seen truthful and deceptive news. A scarcity of deceptive news, available as corpora for predictive modeling, is a major stumbling block in this field of natural language processing (NLP) and deception detection. This paper discusses three types of fake news, each in contrast to genuine serious reporting, and weighs their pros and cons as a corpus for text analytics and predictive modeling. Filtering, vetting, and verifying online information continues to be essential in library and information science (LIS), as the lines between traditional news and online information are blurring."
]
} |
1708.07104 | 2749784378 | The proliferation of misleading information in everyday access media outlets such as social media feeds, news blogs, and online newspapers have made it challenging to identify trustworthy news sources, thus increasing the need for computational tools able to provide insights into the reliability of online content. In this paper, we focus on the automatic identification of fake content in online news. Our contribution is twofold. First, we introduce two novel datasets for the task of fake news detection, covering seven different news domains. We describe the collection, annotation, and validation process in detail and present several exploratory analysis on the identification of linguistic differences in fake and legitimate news content. Second, we conduct a set of learning experiments to build accurate fake news detectors. In addition, we provide comparative analyses of the automatic and manual identification of fake news. | Second, attempts to differentiate satire from real news yielded promising results @cite_8 . The authors built a corpus of satire news (from and ) and real news ( and ) in four domains (civics, science, business, soft news), resulting in a total of 240 news articles. The best classification performances were achieved with feature sets representing absurdity, punctuation, and grammar (each with an F1 score of 0.87). | {
"cite_N": [
"@cite_8"
],
"mid": [
"2516338333"
],
"abstract": [
"Satire is an attractive subject in deception detection research: it is a type of deception that intentionally incorporates cues revealing its own deceptiveness. Whereas other types of fabrications aim to instill a false sense of truth in the reader, a successful satirical hoax must eventually be exposed as a jest. This paper provides a conceptual overview of satire and humor, elaborating and illustrating the unique features of satirical news, which mimics the format and style of journalistic reporting. Satirical news stories were carefully matched and examined in contrast with their legitimate news counterparts in 12 contemporary news topics in 4 domains (civics, science, business, and “soft” news). Building on previous work in satire detection, we proposed an SVMbased algorithm, enriched with 5 predictive features (Absurdity, Humor, Grammar, Negative Affect, and Punctuation) and tested their combinations on 360 news articles. Our best predicting feature combination (Absurdity, Grammar and Punctuation) detects satirical news with a 90 precision and 84 recall (F-score=87 ). Our work in algorithmically identifying satirical news pieces can aid in minimizing the potential deceptive impact of satire."
]
} |
1708.07104 | 2749784378 | The proliferation of misleading information in everyday access media outlets such as social media feeds, news blogs, and online newspapers have made it challenging to identify trustworthy news sources, thus increasing the need for computational tools able to provide insights into the reliability of online content. In this paper, we focus on the automatic identification of fake content in online news. Our contribution is twofold. First, we introduce two novel datasets for the task of fake news detection, covering seven different news domains. We describe the collection, annotation, and validation process in detail and present several exploratory analysis on the identification of linguistic differences in fake and legitimate news content. Second, we conduct a set of learning experiments to build accurate fake news detectors. In addition, we provide comparative analyses of the automatic and manual identification of fake news. | Also related to our research is work done on the automatic identification of deceptive content, which has explored domains such as forums, consumer reviews websites, online advertising, online dating, and crowdfounding platforms @cite_5 @cite_9 @cite_12 @cite_4 @cite_1 . Linguistic clues such as self references or positive and negative words have been used to profile true tellers from liars @cite_13 . Other work has focused on analyzing the number of words, sentences, self references, affect, spatial and temporal information associated with deceptive content @cite_2 . Expressivity, informality, diversity and non-immediacy have also been explored to identify deceitful behaviors @cite_1 . | {
"cite_N": [
"@cite_4",
"@cite_9",
"@cite_1",
"@cite_2",
"@cite_5",
"@cite_13",
"@cite_12"
],
"mid": [
"",
"",
"2341908715",
"2106165737",
"2158094649",
"2091034860",
"2155841152"
],
"abstract": [
"",
"",
"Crowdfunding sites with recent explosive growth are equally attractive platforms for swindlers or scammers. Though the growing number of articles on crowdfunding scams indicate that the fraud threats are accelerating, there has been little knowledge on the scamming practices and patterns. The key contribution of this research is to discover the hidden clues in the text by exploring linguistic features to distinguish scam campaigns from non-scams. Our results indicate that by providing less information and writing more carefully (and less informally), scammers deliberately try to deceive people; (i) they use less number of words, verbs, and sentences in their campaign pages. (ii) scammers make less typographical errors, 4.5-4.7 times lower than non-scammers.(iii) Expressivity of scams is 2.6-8.5 times lower as well.",
"Modality is an important context factor in deception, which is context-dependent. In order to build a reliable and flexible tool for automatic-deception-detection (ADD), we investigated the characteristics of verbal cues to deceptive behavior in three modalities: text, audio and face-to-face communication. Seven categories of verbal cues (21 cues) were studied: quantity, complexity, diversity, verb nonimmediacy, uncertainty, specificity and affect. After testing the interaction effects between modality and condition (deception or truth), we found significance only with specificity and observed that differences between deception and truth were in general consistent across the three modalities. However, modality had strong effects on verbal cues. For example, messages delivered face-to-face were largest in quantity (number of words, verbs, and sentences), followed by the audio modality. Text had the sparsest examples. These modality effects are an important factor in building baselines in ADD tools, because they make it possible to use them to adjust the baseline for an unknown modality according to a known baseline, thereby simplifying the process of ADD. The paper discusses in detail the implications of these findings on modality effects in three modalities.",
"This article explores the operation of warrants, connections between online and real-world identities, on deceptive behavior in computer-mediated communication. A survey of 132 participants assessed three types of warrants (the use of a real name, a photo, and the presence of real-world acquaintances) in five different media: IM, Forums, Chat, Social Networking Sites (SNS) and Email. The effect of warrants on lies about demographic information (e.g., age, gender, education, etc.), one's interests (e.g., religion, music preferences, etc.), and the seriousness of lies was assessed. Overall, deception was observed most frequently in Chat and least often in SNS and Email. The relationship between warrants and deception was negative and linear, with warrants suppressing the frequency and seriousness of deception regardless of medium, although real-world acquaintances were especially powerful in constraining deception in SNS and emails.",
"Telling lies often requires creating a story about an experience or attitude that does not exist. As a result, false stories may be qualitatively different from true stories. The current project investigated the features of linguistic style that distinguish between true and false stories. In an analysis of five independent samples, a computer-based text analysis program correctly classified liars and truth-tellers at a rate of 67 when the topic was constant and a rate of 61 overall. Compared to truth-tellers, liars showed lower cognitive complexity, used fewer self-references and other-references, and used more negative emotion words.",
"With the rapid growth of the Internet, online advertisement plays a more and more important role in the advertising market. One of the current and widely used revenue models for online advertising involves charging for each click based on the popularity of keywords and the number of competing advertisers. This pay-per-click model leaves room for individuals or rival companies to generate false clicks (i.e., click fraud), which pose serious problems to the development of healthy online advertising market. To detect click fraud, an important issue is to detect duplicate clicks over decaying window models, such as jumping windows and sliding windows. Decaying window models can be very helpful in defining and determining click fraud. However, although there are available algorithms to detect duplicates, there is still a lack of practical and effective solutions to detect click fraud in pay-per-click streams over decaying window models. In this paper, we address the problem of detecting duplicate clicks in pay-per-click streams over jumping windows and sliding windows, and are the first that propose two innovative algorithms that make only one pass over click streams and require significantly less memory space and operations. GBF algorithm is built on group Bloom filters which can process click streams over jumping windows with small number of sub-windows, while TBF algorithm is based on a new data structure called timing Bloom filter that detects click fraud over sliding windows and jumping windows with large number of sub-windows. Both GBF algorithm and TBF algorithm have zero false negative. Furthermore, both theoretical analysis and experimental results show that our algorithms can achieve low false positive rate when detecting duplicate clicks in pay-per-click streams over jumping windows and sliding windows."
]
} |
1708.06860 | 2749727272 | Increasingly, software developers are using a wide array of social collaborative platforms for software development and learning. In this work, we examined the similarities in developer's interests within and across GitHub and Stack Overflow. Our study finds that developers share common interests in GitHub and Stack Overflow, on average, 39 of the GitHub repositories and Stack Overflow questions that a developer had participated fall in the common interests. Also, developers do share similar interests with other developers who co-participated activities in the two platforms. In particular, developers who co-commit and co-pull-request same GitHub repositories and co-answer same Stack Overflow questions, share more common interests compare to other developers who co-participate in other platform activities. | Stack Overflow and GitHub have also been studied for empirical works on developer interests. For example, there were research works that focused on analyzing topics asked by developers in Stack Overflow @cite_1 @cite_5 @cite_8 @cite_4 @cite_15 @cite_16 . Similarly, there were also works on analyzing programming languages used by developers in GitHub and their relationships to GitHub contributions @cite_7 @cite_10 @cite_0 @cite_9 @cite_14 @cite_13 @cite_12 . There are also studies characterizing social network properties of GitHub and Stack Overflow @cite_3 @cite_16 . Our work extends this group of research by comparing developer interests in the two social collaborative platforms. To our best of knowledge, our work is the first inter-network study that examines cross-site developer interests in GitHub and Stack Overflow. | {
"cite_N": [
"@cite_13",
"@cite_14",
"@cite_4",
"@cite_7",
"@cite_8",
"@cite_9",
"@cite_1",
"@cite_3",
"@cite_0",
"@cite_5",
"@cite_15",
"@cite_16",
"@cite_10",
"@cite_12"
],
"mid": [
"",
"1994598608",
"2093400716",
"1966769823",
"",
"",
"2056894403",
"2118174895",
"2022248256",
"2108545456",
"",
"1981795678",
"",
""
],
"abstract": [
"",
"Given the increasing number of unsuccessful pull requests in GitHub projects, insights into the success and failure of these requests are essential for the developers. In this paper, we provide a comparative study between successful and unsuccessful pull requests made to 78 GitHub base projects by 20,142 developers from 103,192 forked projects. In the study, we analyze pull request discussion texts, project specific information (e.g., domain, maturity), and developer specific information (e.g., experience) in order to report useful insights, and use them to contrast between successful and unsuccessful pull requests. We believe our study will help developers overcome the issues with pull requests in GitHub, and project administrators with informed decision making.",
"The popularity of mobile devices has been steadily growing in recent years. These devices heavily depend on software from the underlying operating systems to the applications they run. Prior research showed that mobile software is different than traditional, large software systems. However, to date most of our research has been conducted on traditional software systems. Very little work has focused on the issues that mobile developers face. Therefore, in this paper, we use data from the popular online Q&A site, Stack Overflow, and analyze 13,232,821 posts to examine what mobile developers ask about. We employ Latent Dirichlet allocation-based topic models to help us summarize the mobile-related questions. Our findings show that developers are asking about app distribution, mobile APIs, data management, sensors and context, mobile tools, and user interface development. We also determine what popular mobile-related issues are the most difficult, explore platform specific issues, and investigate the types (e.g., what, how, or why) of questions mobile developers ask. Our findings help highlight the challenges facing mobile developers that require more attention from the software engineering research and development communities in the future and establish a novel approach for analyzing questions asked on Q&A forums.",
"Feedback from software users constitutes a vital part in the evolution of software projects. By filing issue reports, users help identify and fix bugs, document software code, and enhance the software via feature requests. Many studies have explored issue reports, proposed approaches to enable the submission of higher-quality reports, and presented techniques to sort, categorize and leverage issues for software engineering needs. Who, however, cares about filing issues? What kind of issues are reported in issue trackers? What kind of correlation exist between issue reporting and the success of software projects? In this study, we address the need for answering such questions by performing an empirical study on a hundred thousands of open source projects. After filtering relevant trackers, the study used about 20,000 projects. We investigate and answer various research questions on the popularity and impact of issue trackers.",
"",
"",
"Programming question and answer (QA questions in some topics lead to discussions in other topics; and the topics gaining the most popularity over time are web development (especially jQuery), mobile applications (especially Android), Git, and MySQL.",
"Social coding enables a different experience of software development as the activities and interests of one developer are easily advertised to other developers. Developers can thus track the activities relevant to various projects in one umbrella site. Such a major change in collaborative software development makes an investigation of networkings on social coding sites valuable. Furthermore, project hosting platforms promoting this development paradigm have been thriving, among which GitHub has arguably gained the most momentum. In this paper, we contribute to the body of knowledge on social coding by investigating the network structure of social coding in GitHub. We collect 100,000 projects and 30,000 developers from GitHub, construct developer-developer and project-project relationship graphs, and compute various characteristics of the graphs. We then identify influential developers and projects on this sub network of GitHub by using PageRank. Understanding how developers and projects are actually related to each other on a social coding site is the first step towards building tool supports to aid social programmers in performing their tasks more efficiently.",
"Users on GitHub can watch repositories to receive notifications about project activity. This introduces a new type of passive project membership. In this paper, we investigate the behavior of watchers and their contribution to the projects they watch. We find that a subset of project watchers begin contributing to the project and those contributors account for a significant percentage of contributors on the project. As contributors, watchers are more confident and contribute over a longer period of time in a more varied way than other contributors. This is likely attributable to the knowledge gained through project notifications.",
"Modern web applications consist of a significant amount of client- side code, written in JavaScript, HTML, and CSS. In this paper, we present a study of common challenges and misconceptions among web developers, by mining related questions asked on Stack Over- flow. We use unsupervised learning to categorize the mined questions and define a ranking algorithm to rank all the Stack Overflow questions based on their importance. We analyze the top 50 questions qualitatively. The results indicate that (1) the overall share of web development related discussions is increasing among developers, (2) browser related discussions are prevalent; however, this share is decreasing with time, (3) form validation and other DOM related discussions have been discussed consistently over time, (4) web related discussions are becoming more prevalent in mobile development, and (5) developers face implementation issues with new HTML5 features such as Canvas. We examine the implications of the results on the development, research, and standardization communities.",
"",
"StackOverflow provides a popular platform where developers post and answer questions. Recently, manually label 385 questions in StackOverflow and group them into 10 categories based on their contents. They also analyze how tags are used in StackOverflow. In this study, we extend their work to obtain a deeper understanding on how developers interact with one another on such a question and answer web site. First, we analyze the distributions of developers who ask and answer questions. We also investigate if there is a segregation of the StackOverflow community into questioners and answerers. We also perform automated text mining to find the various kinds of topics asked by developers. We use Latent Dirichlet Allocation (LDA), a well known topic modeling approach, to analyze the contents of tens of thousands of questions and answers, and produce five topics. Our topic modeling strategy provides an alternative perspective different from that of for categorizing StackOverflow questions. Each question can now be categorized into several topics with different probabilities, and the learned topic model could automatically assign a new question to several categories with varying probabilities. Last but not least, we show the distributions of questions and developers belonging to various topics generated by LDA.",
"",
""
]
} |
1708.07178 | 2746283391 | We present the Parallel, Forward-Backward with Pruning (PFBP) algorithm for feature selection (FS) in Big Data settings (high dimensionality and or sample size). To tackle the challenges of Big Data FS PFBP partitions the data matrix both in terms of rows (samples, training examples) as well as columns (features). By employing the concepts of p-values of conditional independence tests and meta-analysis techniques PFBP manages to rely only on computations local to a partition while minimizing communication costs. Then, it employs powerful and safe (asymptotically sound) heuristics to make early, approximate decisions, such as Early Dropping of features from consideration in subsequent iterations, Early Stopping of consideration of features within the same iteration, or Early Return of the winner in each iteration. PFBP provides asymptotic guarantees of optimal-ity for data distributions faithfully representable by a causal network (Bayesian network or maximal ancestral graph). Our empirical analysis confirms a super-linear speedup of the algorithm with increasing sample size, linear scalability with respect to the number of features and processing cores, while dominating other competitive algorithms in its class. | In this section we provide an overview of alternative parallel feature selection methods, focusing on methods for MapReduce-based systems (such as Spark), as well as causal-based methods, and compare them to PFBP. An overview of feature selection methods can be found in @cite_22 and @cite_35 . | {
"cite_N": [
"@cite_35",
"@cite_22"
],
"mid": [
"2260771218",
"2119479037"
],
"abstract": [
"Feature selection, as a data preprocessing strategy, has been proven to be effective and efficient in preparing data (especially high-dimensional data) for various data-mining and machine-learning problems. The objectives of feature selection include building simpler and more comprehensible models, improving data-mining performance, and preparing clean, understandable data. The recent proliferation of big data has presented some substantial challenges and opportunities to feature selection. In this survey, we provide a comprehensive and structured overview of recent advances in feature selection research. Motivated by current challenges and opportunities in the era of big data, we revisit feature selection research from a data perspective and review representative feature selection algorithms for conventional data, structured data, heterogeneous data and streaming data. Methodologically, to emphasize the differences and similarities of most existing feature selection algorithms for conventional data, we categorize them into four main groups: similarity-based, information-theoretical-based, sparse-learning-based, and statistical-based methods. To facilitate and promote the research in this community, we also present an open source feature selection repository that consists of most of the popular feature selection algorithms (http: featureselection.asu.edu ). Also, we use it as an example to show how to evaluate feature selection algorithms. At the end of the survey, we present a discussion about some open problems and challenges that require more attention in future research.",
"Variable and feature selection have become the focus of much research in areas of application for which datasets with tens or hundreds of thousands of variables are available. These areas include text processing of internet documents, gene expression array analysis, and combinatorial chemistry. The objective of variable selection is three-fold: improving the prediction performance of the predictors, providing faster and more cost-effective predictors, and providing a better understanding of the underlying process that generated the data. The contributions of this special issue cover a wide range of aspects of such problems: providing a better definition of the objective function, feature construction, feature ranking, multivariate feature selection, efficient search methods, and feature validity assessment methods."
]
} |
1708.07178 | 2746283391 | We present the Parallel, Forward-Backward with Pruning (PFBP) algorithm for feature selection (FS) in Big Data settings (high dimensionality and or sample size). To tackle the challenges of Big Data FS PFBP partitions the data matrix both in terms of rows (samples, training examples) as well as columns (features). By employing the concepts of p-values of conditional independence tests and meta-analysis techniques PFBP manages to rely only on computations local to a partition while minimizing communication costs. Then, it employs powerful and safe (asymptotically sound) heuristics to make early, approximate decisions, such as Early Dropping of features from consideration in subsequent iterations, Early Stopping of consideration of features within the same iteration, or Early Return of the winner in each iteration. PFBP provides asymptotic guarantees of optimal-ity for data distributions faithfully representable by a causal network (Bayesian network or maximal ancestral graph). Our empirical analysis confirms a super-linear speedup of the algorithm with increasing sample size, linear scalability with respect to the number of features and processing cores, while dominating other competitive algorithms in its class. | In extensive simulations it has been shown that causal-based feature selection methods are competitive with Lasso on classification and survival analysis tasks on many real datasets @cite_51 @cite_16 @cite_31 @cite_49 . Furthermore, the non-parallel version of PFBP (called Forward-Backward Selection with Early Dropping) as well as the standard Forward-Backward Selection algorithm have been shown to perform as well as Lasso if restricted to select the same number of variables @cite_12 . | {
"cite_N": [
"@cite_49",
"@cite_31",
"@cite_16",
"@cite_51",
"@cite_12"
],
"mid": [
"2950572649",
"",
"2171267265",
"2133091666",
"2619567144"
],
"abstract": [
"The statistically equivalent signature (SES) algorithm is a method for feature selection inspired by the principles of constrained-based learning of Bayesian Networks. Most of the currently available feature-selection methods return only a single subset of features, supposedly the one with the highest predictive power. We argue that in several domains multiple subsets can achieve close to maximal predictive accuracy, and that arbitrarily providing only one has several drawbacks. The SES method attempts to identify multiple, predictive feature subsets whose performances are statistically equivalent. Under that respect SES subsumes and extends previous feature selection algorithms, like the max-min parent children algorithm. SES is implemented in an homonym function included in the R package MXM, standing for mens ex machina, meaning 'mind from the machine' in Latin. The MXM implementation of SES handles several data-analysis tasks, namely classification, regression and survival analysis. In this paper we present the SES algorithm, its implementation, and provide examples of use of the SES function in R. Furthermore, we analyze three publicly available data sets to illustrate the equivalence of the signatures retrieved by SES and to contrast SES against the state-of-the-art feature selection method LASSO. Our results provide initial evidence that the two methods perform comparably well in terms of predictive accuracy and that multiple, equally predictive signatures are actually present in real world data.",
"",
"Motivation: Variable selection is a typical approach used for molecular-signature and biomarker discovery; however, its application to survival data is often complicated by censored samples. We propose a new algorithm for variable selection suitable for the analysis of high-dimensional, right-censored data called Survival Max–Min Parents and Children (SMMPC). The algorithm is conceptually simple, scalable, based on the theory of Bayesian networks (BNs) and the Markov blanket and extends the corresponding algorithm (MMPC) for classification tasks. The selected variables have a structural interpretation: if T is the survival time (in general the time-to-event), SMMPC returns the variables adjacent to T in the BN representing the data distribution. The selected variables also have a causal interpretation that we discuss. Results: We conduct an extensive empirical analysis of prototypical and state-of-the-art variable selection algorithms for survival data that are applicable to high-dimensional biological data. SMMPC selects on average the smallest variable subsets (less than a dozen per dataset), while statistically significantly outperforming all of the methods in the study returning a manageable number of genes that could be inspected by a human expert. Availability: Matlab and R code are freely available from http: www.mensxmachina.org Contact: vlagani@ics.forth.gr Supplementary information:Supplementary data are available at Bioinformatics online.",
"We present an algorithmic framework for learning local causal structure around target variables of interest in the form of direct causes effects and Markov blankets applicable to very large data sets with relatively small samples. The selected feature sets can be used for causal discovery and classification. The framework (Generalized Local Learning, or GLL) can be instantiated in numerous ways, giving rise to both existing state-of-the-art as well as novel algorithms. The resulting algorithms are sound under well-defined sufficient conditions. In a first set of experiments we evaluate several algorithms derived from this framework in terms of predictivity and feature set parsimony and compare to other local causal discovery methods and to state-of-the-art non-causal feature selection methods using real data. A second set of experimental evaluations compares the algorithms in terms of ability to induce local causal neighborhoods using simulated and resimulated data and examines the relation of predictivity with causal induction performance. Our experiments demonstrate, consistently with causal feature selection theory, that local causal feature selection methods (under broad assumptions encompassing appropriate family of distributions, types of classifiers, and loss functions) exhibit strong feature set parsimony, high predictivity and local causal interpretability. Although non-causal feature selection methods are often used in practice to shed light on causal relationships, we find that they cannot be interpreted causally even when they achieve excellent predictivity. Therefore we conclude that only local causal techniques should be used when insight into causal structure is sought. In a companion paper we examine in depth the behavior of GLL algorithms, provide extensions, and show how local techniques can be used for scalable and accurate global causal graph learning.",
"Forward-backward selection is one of the most basic and commonly-used feature selection algorithms available. It is also general and conceptually applicable to many different types of data. In this paper, we propose a heuristic that significantly improves its running time, while preserving predictive accuracy. The idea is to temporarily discard the variables that are conditionally independent with the outcome given the selected variable set. Depending on how those variables are reconsidered and reintroduced, this heuristic gives rise to a family of algorithms with increasingly stronger theoretical guarantees. In distributions that can be faithfully represented by Bayesian networks or maximal ancestral graphs, members of this algorithmic family are able to correctly identify the Markov blanket in the sample limit. In experiments we show that the proposed heuristic increases computational efficiency by about two orders of magnitude in high-dimensional problems, while selecting fewer variables and retaining predictive performance. Furthermore, we show that the proposed algorithm and feature selection with LASSO perform similarly when restricted to select the same number of variables, making the proposed algorithm an attractive alternative for problems where no (efficient) algorithm for LASSO exists."
]
} |
1708.07178 | 2746283391 | We present the Parallel, Forward-Backward with Pruning (PFBP) algorithm for feature selection (FS) in Big Data settings (high dimensionality and or sample size). To tackle the challenges of Big Data FS PFBP partitions the data matrix both in terms of rows (samples, training examples) as well as columns (features). By employing the concepts of p-values of conditional independence tests and meta-analysis techniques PFBP manages to rely only on computations local to a partition while minimizing communication costs. Then, it employs powerful and safe (asymptotically sound) heuristics to make early, approximate decisions, such as Early Dropping of features from consideration in subsequent iterations, Early Stopping of consideration of features within the same iteration, or Early Return of the winner in each iteration. PFBP provides asymptotic guarantees of optimal-ity for data distributions faithfully representable by a causal network (Bayesian network or maximal ancestral graph). Our empirical analysis confirms a super-linear speedup of the algorithm with increasing sample size, linear scalability with respect to the number of features and processing cores, while dominating other competitive algorithms in its class. | Lasso has been parallelized for single machines and shared-memory clusters @cite_64 @cite_25 @cite_9 . These approaches only parallelize over features and not samples (i.e. consider vertical partitioning). Naturally, ideas and techniques presented in those works could be adapted or extended for Spark or related systems. An implementation of Lasso linear regression is provided in the Spark MLlib library . A disadvantage of that implementation is that it requires extensive tuning of its hyper-parameters (like the regularization parameter @math and several parameters of the optimization procedure), rendering it impractical as typically many different hyper-parameter combinations have to be used. Unfortunately, we were not able to find any Spark implementation of Lasso for logistic regression, or any work dealing with the problem of efficient parallelization of Lasso on Spark. Finally, we note that in contrast to forward selection using conditional independence tests, Lasso is not easily extensible for different problems, and requires specialized algorithms for different data types , whose objective function may be non-convex or computationally demanding . | {
"cite_N": [
"@cite_9",
"@cite_64",
"@cite_25"
],
"mid": [
"2386452692",
"2114643899",
"2062770263"
],
"abstract": [
"Lasso regression is a widely used technique in data mining for model selection and feature extraction. In many applications, it remains challenging to apply the regression model to large-scale problems that have massive data samples with high-dimensional features. One popular and promising strategy is to solve the Lasso problem in parallel. Parallel solvers run multiple cores in parallel on a shared memory system to speedup the computation, while the practical usage is limited by the huge dimension in the feature space. Screening is a promising method to solve the problem of high dimensionality by discarding the inactive features and removing them from optimization. However, when integrating screening methods with parallel solvers, most of solvers cannot guarantee the convergence on the reduced feature matrix. In this paper, we propose a novel parallel framework by parallelizing screening methods and integrating it with our proposed parallel solver. We propose two parallel screening algorithms: Parallel Strong Rule (PSR) and Parallel Dual Polytope Projection (PDPP). For the parallel solver, we proposed an Asynchronous Grouped Coordinate Descent method (AGCD) to optimize the regression problem in parallel on the reduced feature matrix. AGCD is based on a grouped selection strategy to select the coordinate that has the maximum descent for the objective function in a group of candidates. Empirical studies on the real-world datasets demonstrate that the proposed parallel framework has a superior performance compared to the state-of-the-art parallel solvers.",
"We propose Shotgun, a parallel coordinate descent algorithm for minimizing L1-regularized losses. Though coordinate descent seems inherently sequential, we prove convergence bounds for Shotgun which predict linear speedups, up to a problem-dependent limit. We present a comprehensive empirical study of Shotgun for Lasso and sparse logistic regression. Our theoretical predictions on the potential for parallelism closely match behavior on real data. Shotgun outperforms other published solvers on a range of large problems, proving to be one of the most scalable algorithms for L1.",
"This paper proposes parallel and distributed algorithms for solving very large-scale sparse optimization problems on computer clusters and clouds. Modern datasets usually have a large number of features or training samples, and they are usually stored in a distributed manner. Motivated by the need of solving sparse optimization problems with large datasets, we propose two approaches including (i) distributed implementations of prox-linear algorithms and (ii) GRock, a parallel greedy block coordinate descent method. Different separability properties of the objective terms in the problem enable different data distributed schemes along with their corresponding algorithm implementations. We also establish the convergence of GRock and explain why it often performs exceptionally well for sparse optimization. Numerical results on a computer cluster and Amazon EC2 demonstrate the efficiency and elasticity of our algorithms."
]
} |
1708.07178 | 2746283391 | We present the Parallel, Forward-Backward with Pruning (PFBP) algorithm for feature selection (FS) in Big Data settings (high dimensionality and or sample size). To tackle the challenges of Big Data FS PFBP partitions the data matrix both in terms of rows (samples, training examples) as well as columns (features). By employing the concepts of p-values of conditional independence tests and meta-analysis techniques PFBP manages to rely only on computations local to a partition while minimizing communication costs. Then, it employs powerful and safe (asymptotically sound) heuristics to make early, approximate decisions, such as Early Dropping of features from consideration in subsequent iterations, Early Stopping of consideration of features within the same iteration, or Early Return of the winner in each iteration. PFBP provides asymptotic guarantees of optimal-ity for data distributions faithfully representable by a causal network (Bayesian network or maximal ancestral graph). Our empirical analysis confirms a super-linear speedup of the algorithm with increasing sample size, linear scalability with respect to the number of features and processing cores, while dominating other competitive algorithms in its class. | Several algorithms have appeared in the literature that apply tests of conditional independence to select features. The theoretical properties of these algorithms often rely on the theory of Bayesian networks and the Markov blanket. The GS @cite_66 @cite_71 and the IAMB @cite_63 algorithms, were some of the first to present the forward-backward selection algorithm in the context of Bayesian networks and the Markov blanket and prove correctness for faithful distributions. These algorithms perform tests of independence conditioning each time on the full set @math of selected features and can guarantee to identify the Markov blanket for faithful distributions asymptotically. However, the larger the conditioning set, the more samples are required to obtain valid results. Thus, these algorithms are not well-suited for problems with large Markov blankets relative to the available sample size. | {
"cite_N": [
"@cite_63",
"@cite_66",
"@cite_71"
],
"mid": [
"2161642932",
"2129564794",
"2154058606"
],
"abstract": [
"This paper presents a number of new algorithms for discovering the Markov Blanket of a target variable T from training data. The Markov Blanket can be used for variable selection for classification, for causal discovery, and for Bayesian Network learning. We introduce a low-order polynomial algorithm and several variants that soundly induce the Markov Blanket under certain broad conditions in datasets with thousands of variables and compare them to other state-of-the-art local and global methods with excellent results.",
"In recent years, Bayesian networks have become highly successful tool for diagnosis, analysis, and decision making in real-world domains. We present an efficient algorithm for learning Bayes networks from data. Our approach constructs Bayesian networks by first identifying each node's Markov blankets, then connecting nodes in a maximally consistent way. In contrast to the majority of work, which typically uses hill-climbing approaches that may produce dense and causally incorrect nets, our approach yields much more compact causal networks by heeding independencies in the data. Compact causal networks facilitate fast inference and are also easier to understand. We prove that under mild assumptions, our approach requires time polynomial in the size of the data and the number of nodes. A randomized variant, also presented here, yields comparable results at much higher speeds.",
"In this paper we address the problem of provably correct feature selection in arbitrary domains. An optimal solution to the problem is a Markov boundary, which is a minimal set of features that make the probability distribution of a target variable conditionally invariant to the state of all other features in the domain. While numerous algorithms for this problem have been proposed, their theoretical correctness and practical behavior under arbitrary probability distributions is unclear. We address this by introducing the Markov Boundary Theorem that precisely characterizes the properties of an ideal Markov boundary, and use it to develop algorithms that learn a more general boundary that can capture complex interactions that only appear when the values of multiple features are considered together. We introduce two algorithms: an exact, provably correct one as well a more practical randomized anytime version, and show that they perform well on artificial as well as benchmark and real-world data sets. Throughout the paper we make minimal assumptions that consist of only a general set of axioms that hold for every probability distribution, which gives these algorithms universal applicability."
]
} |
1708.07178 | 2746283391 | We present the Parallel, Forward-Backward with Pruning (PFBP) algorithm for feature selection (FS) in Big Data settings (high dimensionality and or sample size). To tackle the challenges of Big Data FS PFBP partitions the data matrix both in terms of rows (samples, training examples) as well as columns (features). By employing the concepts of p-values of conditional independence tests and meta-analysis techniques PFBP manages to rely only on computations local to a partition while minimizing communication costs. Then, it employs powerful and safe (asymptotically sound) heuristics to make early, approximate decisions, such as Early Dropping of features from consideration in subsequent iterations, Early Stopping of consideration of features within the same iteration, or Early Return of the winner in each iteration. PFBP provides asymptotic guarantees of optimal-ity for data distributions faithfully representable by a causal network (Bayesian network or maximal ancestral graph). Our empirical analysis confirms a super-linear speedup of the algorithm with increasing sample size, linear scalability with respect to the number of features and processing cores, while dominating other competitive algorithms in its class. | Another class of such algorithms includes HITON-PC @cite_5 , MMPC @cite_11 , and more recently SES @cite_49 for multiple solutions. The main difference in this class of algorithms is that they condition on subsets of the selected features @math , not the full set. They do not guarantee to identify the full Markov blanket, but only a superset of the parents and children of @math . Recent extensive experiments have concluded that they perform well in practice @cite_51 . These algorithms remove from consideration any features that become independent of @math conditioned on some subset of the selected features @math . This is similar to the Early Dropping heuristic and renders the algorithms quite computationally efficient and scalable to high-dimensional settings. | {
"cite_N": [
"@cite_5",
"@cite_51",
"@cite_49",
"@cite_11"
],
"mid": [
"2121273346",
"2133091666",
"2950572649",
""
],
"abstract": [
"We introduce a novel, sound, sample-efficient, and highly-scalable algorithm for variable selection for classification, regression and prediction called HITON. The algorithm works by inducing the Markov Blanket of the variable to be classified or predicted. A wide variety of biomedical tasks with different characteristics were used for an empirical evaluation. Namely, (i) bioactivity prediction for drug discovery, (ii) clinical diagnosis of arrhythmias, (iii) bibliographic text categorization, (iv) lung cancer diagnosis from gene expression array data, and (v) proteomics-based prostate cancer detection. State-of-the-art algorithms for each domain were selected for baseline comparison. Results: (1) HITON reduces the number of variables in the prediction models by three orders of magnitude relative to the original variable set while improving or maintaining accuracy. (2) HITON outperforms the baseline algorithms by selecting more than two orders-ofmagnitude smaller variable sets than the baselines, in the selected tasks and datasets.",
"We present an algorithmic framework for learning local causal structure around target variables of interest in the form of direct causes effects and Markov blankets applicable to very large data sets with relatively small samples. The selected feature sets can be used for causal discovery and classification. The framework (Generalized Local Learning, or GLL) can be instantiated in numerous ways, giving rise to both existing state-of-the-art as well as novel algorithms. The resulting algorithms are sound under well-defined sufficient conditions. In a first set of experiments we evaluate several algorithms derived from this framework in terms of predictivity and feature set parsimony and compare to other local causal discovery methods and to state-of-the-art non-causal feature selection methods using real data. A second set of experimental evaluations compares the algorithms in terms of ability to induce local causal neighborhoods using simulated and resimulated data and examines the relation of predictivity with causal induction performance. Our experiments demonstrate, consistently with causal feature selection theory, that local causal feature selection methods (under broad assumptions encompassing appropriate family of distributions, types of classifiers, and loss functions) exhibit strong feature set parsimony, high predictivity and local causal interpretability. Although non-causal feature selection methods are often used in practice to shed light on causal relationships, we find that they cannot be interpreted causally even when they achieve excellent predictivity. Therefore we conclude that only local causal techniques should be used when insight into causal structure is sought. In a companion paper we examine in depth the behavior of GLL algorithms, provide extensions, and show how local techniques can be used for scalable and accurate global causal graph learning.",
"The statistically equivalent signature (SES) algorithm is a method for feature selection inspired by the principles of constrained-based learning of Bayesian Networks. Most of the currently available feature-selection methods return only a single subset of features, supposedly the one with the highest predictive power. We argue that in several domains multiple subsets can achieve close to maximal predictive accuracy, and that arbitrarily providing only one has several drawbacks. The SES method attempts to identify multiple, predictive feature subsets whose performances are statistically equivalent. Under that respect SES subsumes and extends previous feature selection algorithms, like the max-min parent children algorithm. SES is implemented in an homonym function included in the R package MXM, standing for mens ex machina, meaning 'mind from the machine' in Latin. The MXM implementation of SES handles several data-analysis tasks, namely classification, regression and survival analysis. In this paper we present the SES algorithm, its implementation, and provide examples of use of the SES function in R. Furthermore, we analyze three publicly available data sets to illustrate the equivalence of the signatures retrieved by SES and to contrast SES against the state-of-the-art feature selection method LASSO. Our results provide initial evidence that the two methods perform comparably well in terms of predictive accuracy and that multiple, equally predictive signatures are actually present in real world data.",
""
]
} |
1708.06828 | 2735502200 | The electronic health record (EHR) contains a large amount of multi-dimensional and unstructured clinical data of significant operational and research value. Distinguished from previous studies, our approach embraces a double-annotated dataset and strays away from obscure "black-box" models to comprehensive deep learning models. In this paper, we present a novel neural attention mechanism that not only classifies clinically important findings. Specifically, convolutional neural networks (CNN) with attention analysis are used to classify radiology head computed tomography reports based on five categories that radiologists would account for in assessing acute and communicable findings in daily practice. The experiments show that our CNN attention models outperform non-neural models, especially when trained on a larger dataset. Our attention analysis demonstrates the intuition behind the classifier's decision by generating a heatmap that highlights attended terms used by the CNN model; this is valuable when potential downstream medical decisions are to be performed by human experts or the classifier information is to be used in cohort construction such as for epidemiological studies. | Methods of extracting unstructured information from the EHR traditionally focused on rule-based systems of NLP, machine learning and statistical analysis, a hybrid of these systems, or cohort identification systems @cite_2 . Regarding Machine Learning and Statistical analysis, @cite_0 reports promising results on predicting post-hospitalization venous thromboembolism (VTE) risk from EHRs by using general machine learning techniques such as Naive Bayes, Support Vector Machines (SVM), -nearest neighbor (-NN), and Random Forest. @cite_20 also successfully used @math -gram SVM to help clinical diagnosis classification in ICU. Applications of neural networks also gained tremendous momentum in clinical note extraction, especially in relation extraction and named entity recognition. CNN, although originally invented for the purpose of solving computer vision, has proven to work profoundly well in various NLP tasks, and used for supervised learning and automatically learning features for classification of relation extraction @cite_26 and named entity recognition @cite_8 . | {
"cite_N": [
"@cite_26",
"@cite_8",
"@cite_0",
"@cite_2",
"@cite_20"
],
"mid": [
"143521272",
"2296283641",
"30705845",
"1808652302",
"2106107776"
],
"abstract": [
"Deep Neural Network has been applied to many Natural Language Processing tasks. Instead of building hand-craft features, DNN builds features by automatic learning, fitting different domains well. In this paper, we propose a novel convolution network, incorporating lexical features, applied to Relation Extraction. Since many current deep neural networks use word embedding by word table, which, however, neglects semantic meaning among words, we import a new coding method, which coding input words by synonym dictionary to integrate semantic knowledge into the neural network. We compared our Convolution Neural Network CNN on relation extraction with the state-of-art tree kernel approach, including Typed Dependency Path Kernel and Shortest Dependency Path Kernel and Context-Sensitive tree kernel, resulting in a 9 improvement competitive performance on ACE2005 data set. Also, we compared the synonym coding with the one-hot coding, and our approach got 1.6 improvement. Moreover, we also tried other coding method, such as hypernym coding, and give some discussion according the result.",
"Comunicacio presentada a la 2016 Conference of the North American Chapter of the Association for Computational Linguistics, celebrada a San Diego (CA, EUA) els dies 12 a 17 de juny 2016.",
"We consider the task of predicting which patients are most at risk for post-hospitalization venothromboembolism (VTE) using information automatically elicited from an EHR. Given a set of cases and controls, we use machine-learning methods to induce models for making these predictions. Our empirical evaluation of this approach offers a number of interesting and important conclusions. We identify several risk factors for VTE that were not previously recognized. We show that machine-learning methods are able to induce models that identify high-risk patients with accuracy that exceeds previously developed scoring models for VTE. Additionally, we show that, even without having prior knowledge about relevant risk factors, we are able to learn accurate models for this task.",
"Objective To summarize literature describing approaches aimed at automatically identifying patients with a common phenotype. @PARASPLIT Materials and methods We performed a review of studies describing systems or reporting techniques developed for identifying cohorts of patients with specific phenotypes. Every full text article published in (1) Journal of American Medical Informatics Association , (2) Journal of Biomedical Informatics , (3) Proceedings of the Annual American Medical Informatics Association Symposium , and (4) Proceedings of Clinical Research Informatics Conference within the past 3 years was assessed for inclusion in the review. Only articles using automated techniques were included. @PARASPLIT Results Ninety-seven articles met our inclusion criteria. Forty-six used natural language processing (NLP)-based techniques, 24 described rule-based systems, 41 used statistical analyses, data mining, or machine learning techniques, while 22 described hybrid systems. Nine articles described the architecture of large-scale systems developed for determining cohort eligibility of patients. @PARASPLIT Discussion We observe that there is a rise in the number of studies associated with cohort identification using electronic medical records. Statistical analyses or machine learning, followed by NLP techniques, are gaining popularity over the years in comparison with rule-based systems. @PARASPLIT Conclusions There are a variety of approaches for classifying patients into a particular phenotype. Different techniques and data sources are used, and good performance is reported on datasets at respective institutions. However, no system makes comprehensive use of electronic medical records addressing all of their known weaknesses.",
"Background Existing risk adjustment models for intensive care unit (ICU) outcomes rely on manual abstraction of patient-level predictors from medical charts. Developing an automated method for abstracting these data from free text might reduce cost and data collection times. @PARASPLIT Objective To develop a support vector machine (SVM) classifier capable of identifying a range of procedures and diagnoses in ICU clinical notes for use in risk adjustment. @PARASPLIT Materials and methods We selected notes from 2001–2008 for 4191 neonatal ICU (NICU) and 2198 adult ICU patients from the MIMIC-II database from the Beth Israel Deaconess Medical Center. Using these notes, we developed an implementation of the SVM classifier to identify procedures (mechanical ventilation and phototherapy in NICU notes) and diagnoses (jaundice in NICU and intracranial hemorrhage (ICH) in adult ICU). On the jaundice classification task, we also compared classifier performance using n-gram features to unigrams with application of a negation algorithm (NegEx). @PARASPLIT Results Our classifier accurately identified mechanical ventilation (accuracy=0.982, F1=0.954) and phototherapy use (accuracy=0.940, F1=0.912), as well as jaundice (accuracy=0.898, F1=0.884) and ICH diagnoses (accuracy=0.938, F1=0.943). Including bigram features improved performance on the jaundice (accuracy=0.898 vs 0.865) and ICH (0.938 vs 0.927) tasks, and outperformed NegEx-derived unigram features (accuracy=0.898 vs 0.863) on the jaundice task. @PARASPLIT Discussion Overall, a classifier using n-gram support vectors displayed excellent performance characteristics. The classifier generalizes to diverse patient populations, diagnoses, and procedures. @PARASPLIT Conclusions SVM-based classifiers can accurately identify procedure status and diagnoses among ICU patients, and including n-gram features improves performance, compared to existing methods."
]
} |
1708.06828 | 2735502200 | The electronic health record (EHR) contains a large amount of multi-dimensional and unstructured clinical data of significant operational and research value. Distinguished from previous studies, our approach embraces a double-annotated dataset and strays away from obscure "black-box" models to comprehensive deep learning models. In this paper, we present a novel neural attention mechanism that not only classifies clinically important findings. Specifically, convolutional neural networks (CNN) with attention analysis are used to classify radiology head computed tomography reports based on five categories that radiologists would account for in assessing acute and communicable findings in daily practice. The experiments show that our CNN attention models outperform non-neural models, especially when trained on a larger dataset. Our attention analysis demonstrates the intuition behind the classifier's decision by generating a heatmap that highlights attended terms used by the CNN model; this is valuable when potential downstream medical decisions are to be performed by human experts or the classifier information is to be used in cohort construction such as for epidemiological studies. | CNN has also seen upsurge in popularity in document level text classification such as sentiment analysis and question answering @cite_10 @cite_22 @cite_15 . A more recent approach in clinical and biomedical document classification relies on a CNN model proposed by Kim @cite_10 , and leverages the CNN's convolution feature and its ability to effectively capture both semantic and syntactic information to gain a solid 3 The attention mechanism is a method of emphasizing or de-emphasizing features that are more or less important in neural network classification problems @cite_9 . Originally developed for image processing, attention mechanism has successfully been adopted in various NLP domains including question answering, sentiment analysis, machine translation, and document level classification @cite_11 @cite_21 @cite_12 @cite_14 . The attention mechanism introduced here is efficient and gives a comprehensive way of understanding the classification decision. | {
"cite_N": [
"@cite_14",
"@cite_22",
"@cite_9",
"@cite_21",
"@cite_15",
"@cite_10",
"@cite_12",
"@cite_11"
],
"mid": [
"2533667621",
"2049434052",
"2951527505",
"2172010943",
"2963538407",
"",
"2950635152",
"2179022885"
],
"abstract": [
"With the advent of word embeddings, lexicons are no longer fully utilized for sentiment analysis although they still provide important features in the traditional setting. This paper introduces a novel approach to sentiment analysis that integrates lexicon embeddings and an attention mechanism into Convolutional Neural Networks. Our approach performs separate convolutions for word and lexicon embeddings and provides a global view of the document using attention. Our models are experimented on both the SemEval'16 Task 4 dataset and the Stanford Sentiment Treebank, and show comparative or better results against the existing state-of-the-art systems. Our analysis shows that lexicon embeddings allow to build high-performing models with much smaller word embeddings, and the attention mechanism effectively dims out noisy words for sentiment analysis.",
"This paper describes our deep learning system for sentiment analysis of tweets. The main contribution of this work is a new model for initializing the parameter weights of the convolutional neural network, which is crucial to train an accurate model while avoiding the need to inject any additional features. Briefly, we use an unsupervised neural language model to train initial word embeddings that are further tuned by our deep learning model on a distant supervised corpus. At a final stage, the pre-trained parameters of the network are used to initialize the model. We train the latter on the supervised training data recently made available by the official system evaluation campaign on Twitter Sentiment Analysis organized by Semeval-2015. A comparison between the results of our approach and the systems participating in the challenge on the official test sets, suggests that our model could be ranked in the first two positions in both the phrase-level subtask A (among 11 teams) and on the message-level subtask B (among 40 teams). This is an important evidence on the practical value of our solution.",
"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.",
"Traditional convolutional neural networks (CNN) are stationary and feedforward. They neither change their parameters during evaluation nor use feedback from higher to lower layers. Real brains, however, do. So does our Deep Attention Selective Network (dasNet) architecture. DasNets feedback structure can dynamically alter its convolutional filter sensitivities during classification. It harnesses the power of sequential processing to improve classification performance, by allowing the network to iteratively focus its internal attention on some of its convolutional filters. Feedback is trained through direct policy search in a huge million-dimensional parameter space, through scalable natural evolution strategies (SNES). On the CIFAR-10 and CIFAR-100 datasets, dasNet outperforms the previous state-of-the-art model.",
"This paper presents a new selection-based question answering dataset, SelQA. The dataset consists of questions generated through crowdsourcing and sentence length answers that are drawn from the ten most prevalent topics in the English Wikipedia. We introduce a corpus annotation scheme that enhances the generation of large, diverse, and challenging datasets by explicitly aiming to reduce word co-occurrences between the question and answers. Our annotation scheme is composed of a series of crowdsourcing tasks with a view to more effectively utilize crowdsourcing in the creation of question answering datasets in various domains. Several systems are compared on the tasks of answer sentence selection and answer triggering, providing strong baseline results for future work to improve upon.",
"",
"In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.",
"We present a method that learns to answer visual questions by selecting image regions relevant to the text-based query. Our method exhibits significant improvements in answering questions such as \"what color,\" where it is necessary to evaluate a specific location, and \"what room,\" where it selectively identifies informative image regions. Our model is tested on the VQA dataset which is the largest human-annotated visual question answering dataset to our knowledge."
]
} |
1708.06670 | 2963781258 | Deep convolutional neural networks (CNNs) have revolutionized the computer vision research and have seen unprecedented adoption for multiple tasks, such as classification, detection, and caption generation. However, they offer little transparency into their inner workings and are often treated as black boxes that deliver excellent performance. In this paper, we aim at alleviating this opaqueness of CNNs by providing visual explanations for the network’s predictions. Our approach can analyze a variety of CNN-based models trained for computer vision applications, such as object recognition and caption generation. Unlike the existing methods, we achieve this via unraveling the forward pass operation. The proposed method exploits feature dependencies across the layer hierarchy and uncovers the discriminative image locations that guide the network’s predictions. We name these locations CNN fixations, loosely analogous to human eye fixations. Our approach is a generic method that requires no architectural changes, additional training, or gradient computation, and computes the important image locations (CNN fixations). We demonstrate through a variety of applications that our approach is able to localize the discriminative image locations across different network architectures, diverse vision tasks, and data modalities. | Our approach draws similarities to recent visualization works. A number of attempts (e.g. @cite_18 @cite_3 @cite_12 @cite_4 @cite_35 @cite_14 @cite_31 @cite_39 ) have been made in recent time to visualize the classifier decisions and deep learned features. | {
"cite_N": [
"@cite_35",
"@cite_18",
"@cite_14",
"@cite_4",
"@cite_3",
"@cite_39",
"@cite_31",
"@cite_12"
],
"mid": [
"2963382180",
"2113211233",
"2295107390",
"1849277567",
"1787224781",
"2503388974",
"2962858109",
""
],
"abstract": [
"",
"We present a method for explaining predictions for individual instances. The presented approach is general and can be used with all classification models that output probabilities. It is based on the decomposition of a model's predictions on individual contributions of each attribute. Our method works for the so-called black box models such as support vector machines, neural networks, and nearest neighbor algorithms, as well as for ensemble methods such as boosting and random forests. We demonstrate that the generated explanations closely follow the learned models and present a visualization technique that shows the utility of our approach and enables the comparison of different prediction methods.",
"In this work, we revisit the global average pooling layer proposed in [13], and shed light on how it explicitly enables the convolutional neural network (CNN) to have remarkable localization ability despite being trained on imagelevel labels. While this technique was previously proposed as a means for regularizing training, we find that it actually builds a generic localizable deep representation that exposes the implicit attention of CNNs on an image. Despite the apparent simplicity of global average pooling, we are able to achieve 37.1 top-5 error for object localization on ILSVRC 2014 without training on any bounding box annotation. We demonstrate in a variety of experiments that our network is able to localize the discriminative image regions despite just being trained for solving classification task1.",
"Large Convolutional Network models have recently demonstrated impressive classification performance on the ImageNet benchmark [18]. However there is no clear understanding of why they perform so well, or how they might be improved. In this paper we explore both issues. We introduce a novel visualization technique that gives insight into the function of intermediate feature layers and the operation of the classifier. Used in a diagnostic role, these visualizations allow us to find model architectures that outperform on the ImageNet classification benchmark. We also perform an ablation study to discover the performance contribution from different model layers. We show our ImageNet model generalizes well to other datasets: when the softmax classifier is retrained, it convincingly beats the current state-of-the-art results on Caltech-101 and Caltech-256 datasets.",
"Understanding and interpreting classification decisions of automated image classification systems is of high value in many applications, as it allows to verify the reasoning of the system and provides additional information to the human expert. Although machine learning methods are solving very successfully a plethora of tasks, they have in most cases the disadvantage of acting as a black box, not providing any information about what made them arrive at a particular decision. This work proposes a general solution to the problem of understanding classification decisions by pixel-wise decomposition of nonlinear classifiers. We introduce a methodology that allows to visualize the contributions of single pixels to predictions for kernel-based classifiers over Bag of Words features and for multilayered neural networks. These pixel contributions can be visualized as heatmaps and are provided to a human expert who can intuitively not only verify the validity of the classification decision, but also focus further analysis on regions of potential interest. We evaluate our method for classifiers trained on PASCAL VOC 2009 images, synthetic image data containing geometric shapes, the MNIST handwritten digits data set and for the pre-trained ImageNet model available as part of the Caffe open source package.",
"We aim to model the top-down attention of a convolutional neural network (CNN) classifier for generating task-specific attention maps. Inspired by a top-down human visual attention model, we propose a new backpropagation scheme, called Excitation Backprop, to pass along top-down signals downwards in the network hierarchy via a probabilistic Winner-Take-All process. Furthermore, we introduce the concept of contrastive attention to make the top-down attention maps more discriminative. We show a theoretic connection between the proposed contrastive attention formulation and the Class Activation Map computation. Efficient implementation of Excitation Backprop for common neural network layers is also presented. In experiments, we visualize the evidence of a model’s classification decision by computing the proposed top-down attention maps. For quantitative evaluation, we report the accuracy of our method in weakly supervised localization tasks on the MS COCO, PASCAL VOC07 and ImageNet datasets. The usefulness of our method is further validated in the text-to-region association task. On the Flickr30k Entities dataset, we achieve promising performance in phrase localization by leveraging the top-down attention of a CNN model that has been trained on weakly labeled web images. Finally, we demonstrate applications of our method in model interpretation and data annotation assistance for facial expression analysis and medical imaging tasks.",
"We propose a technique for producing ‘visual explanations’ for decisions from a large class of Convolutional Neural Network (CNN)-based models, making them more transparent. Our approach – Gradient-weighted Class Activation Mapping (Grad-CAM), uses the gradients of any target concept (say logits for ‘dog’ or even a caption), flowing into the final convolutional layer to produce a coarse localization map highlighting the important regions in the image for predicting the concept. Unlike previous approaches, Grad- CAM is applicable to a wide variety of CNN model-families: (1) CNNs with fully-connected layers (e.g. VGG), (2) CNNs used for structured outputs (e.g. captioning), (3) CNNs used in tasks with multi-modal inputs (e.g. visual question answering) or reinforcement learning, without architectural changes or re-training. We combine Grad-CAM with existing fine-grained visualizations to create a high-resolution class-discriminative visualization, Guided Grad-CAM, and apply it to image classification, image captioning, and visual question answering (VQA) models, including ResNet-based architectures. In the context of image classification models, our visualizations (a) lend insights into failure modes of these models (showing that seemingly unreasonable predictions have reasonable explanations), (b) outperform previous methods on the ILSVRC-15 weakly-supervised localization task, (c) are more faithful to the underlying model, and (d) help achieve model generalization by identifying dataset bias. For image captioning and VQA, our visualizations show even non-attention based models can localize inputs. Finally, we design and conduct human studies to measure if Grad-CAM explanations help users establish appropriate trust in predictions from deep networks and show that Grad-CAM helps untrained users successfully discern a ‘stronger’ deep network from a ‘weaker’ one even when both make identical predictions. Our code is available at https: github.com ramprs grad-cam along with a demo on CloudCV [2] and video at youtu.be COjUB9Izk6E.",
""
]
} |
1708.06670 | 2963781258 | Deep convolutional neural networks (CNNs) have revolutionized the computer vision research and have seen unprecedented adoption for multiple tasks, such as classification, detection, and caption generation. However, they offer little transparency into their inner workings and are often treated as black boxes that deliver excellent performance. In this paper, we aim at alleviating this opaqueness of CNNs by providing visual explanations for the network’s predictions. Our approach can analyze a variety of CNN-based models trained for computer vision applications, such as object recognition and caption generation. Unlike the existing methods, we achieve this via unraveling the forward pass operation. The proposed method exploits feature dependencies across the layer hierarchy and uncovers the discriminative image locations that guide the network’s predictions. We name these locations CNN fixations, loosely analogous to human eye fixations. Our approach is a generic method that requires no architectural changes, additional training, or gradient computation, and computes the important image locations (CNN fixations). We demonstrate through a variety of applications that our approach is able to localize the discriminative image locations across different network architectures, diverse vision tasks, and data modalities. | Most of these works are gradient based approaches that find out the image regions which can improve the predicted score for a chosen category. Simonyan @cite_20 measure sensitivity of the classification score for a given class with respect to a small change in pixel values. They compute partial derivative of the score in the pixel space and visualize them as saliency maps. They also show that this is closely related to visualizing using deconvolutions by Zeiler @cite_4 . The deconvolution @cite_4 approach visualizes the features (visual concepts) learned by the neurons across different layers. Guided backprop @cite_35 approach modifies the gradients to improve the visualizations qualitatively. | {
"cite_N": [
"@cite_35",
"@cite_4",
"@cite_20"
],
"mid": [
"2963382180",
"1849277567",
"2962851944"
],
"abstract": [
"",
"Large Convolutional Network models have recently demonstrated impressive classification performance on the ImageNet benchmark [18]. However there is no clear understanding of why they perform so well, or how they might be improved. In this paper we explore both issues. We introduce a novel visualization technique that gives insight into the function of intermediate feature layers and the operation of the classifier. Used in a diagnostic role, these visualizations allow us to find model architectures that outperform on the ImageNet classification benchmark. We also perform an ablation study to discover the performance contribution from different model layers. We show our ImageNet model generalizes well to other datasets: when the softmax classifier is retrained, it convincingly beats the current state-of-the-art results on Caltech-101 and Caltech-256 datasets.",
"This paper addresses the visualisation of image classification models, learnt using deep Convolutional Networks (ConvNets). We consider two visualisation techniques, based on computing the gradient of the class score with respect to the input image. The first one generates an image, which maximises the class score [5], thus visualising the notion of the class, captured by a ConvNet. The second technique computes a class saliency map, specific to a given image and class. We show that such maps can be employed for weakly supervised object segmentation using classification ConvNets. Finally, we establish the connection between the gradient-based ConvNet visualisation methods and deconvolutional networks [13]."
]
} |
1708.06670 | 2963781258 | Deep convolutional neural networks (CNNs) have revolutionized the computer vision research and have seen unprecedented adoption for multiple tasks, such as classification, detection, and caption generation. However, they offer little transparency into their inner workings and are often treated as black boxes that deliver excellent performance. In this paper, we aim at alleviating this opaqueness of CNNs by providing visual explanations for the network’s predictions. Our approach can analyze a variety of CNN-based models trained for computer vision applications, such as object recognition and caption generation. Unlike the existing methods, we achieve this via unraveling the forward pass operation. The proposed method exploits feature dependencies across the layer hierarchy and uncovers the discriminative image locations that guide the network’s predictions. We name these locations CNN fixations, loosely analogous to human eye fixations. Our approach is a generic method that requires no architectural changes, additional training, or gradient computation, and computes the important image locations (CNN fixations). We demonstrate through a variety of applications that our approach is able to localize the discriminative image locations across different network architectures, diverse vision tasks, and data modalities. | Zhou @cite_14 showed that class specific activation maps can be obtained by combining the feature maps before the GAP (Global Average Pooling) layer according to the weights connecting the GAP layer to the class activation in the classification layer. However, their method is architecture specific, restricted to networks with GAP layer. Selvaraju @cite_31 address this issue by making it a more generic approach utilizing gradient information. Despite this, @cite_31 still computes low resolution maps (e.g. @math ). Majority of these methods compute partial derivatives of the class scores with respect to image pixels or intermediate feature maps for localizing the image regions. | {
"cite_N": [
"@cite_31",
"@cite_14"
],
"mid": [
"2962858109",
"2295107390"
],
"abstract": [
"We propose a technique for producing ‘visual explanations’ for decisions from a large class of Convolutional Neural Network (CNN)-based models, making them more transparent. Our approach – Gradient-weighted Class Activation Mapping (Grad-CAM), uses the gradients of any target concept (say logits for ‘dog’ or even a caption), flowing into the final convolutional layer to produce a coarse localization map highlighting the important regions in the image for predicting the concept. Unlike previous approaches, Grad- CAM is applicable to a wide variety of CNN model-families: (1) CNNs with fully-connected layers (e.g. VGG), (2) CNNs used for structured outputs (e.g. captioning), (3) CNNs used in tasks with multi-modal inputs (e.g. visual question answering) or reinforcement learning, without architectural changes or re-training. We combine Grad-CAM with existing fine-grained visualizations to create a high-resolution class-discriminative visualization, Guided Grad-CAM, and apply it to image classification, image captioning, and visual question answering (VQA) models, including ResNet-based architectures. In the context of image classification models, our visualizations (a) lend insights into failure modes of these models (showing that seemingly unreasonable predictions have reasonable explanations), (b) outperform previous methods on the ILSVRC-15 weakly-supervised localization task, (c) are more faithful to the underlying model, and (d) help achieve model generalization by identifying dataset bias. For image captioning and VQA, our visualizations show even non-attention based models can localize inputs. Finally, we design and conduct human studies to measure if Grad-CAM explanations help users establish appropriate trust in predictions from deep networks and show that Grad-CAM helps untrained users successfully discern a ‘stronger’ deep network from a ‘weaker’ one even when both make identical predictions. Our code is available at https: github.com ramprs grad-cam along with a demo on CloudCV [2] and video at youtu.be COjUB9Izk6E.",
"In this work, we revisit the global average pooling layer proposed in [13], and shed light on how it explicitly enables the convolutional neural network (CNN) to have remarkable localization ability despite being trained on imagelevel labels. While this technique was previously proposed as a means for regularizing training, we find that it actually builds a generic localizable deep representation that exposes the implicit attention of CNNs on an image. Despite the apparent simplicity of global average pooling, we are able to achieve 37.1 top-5 error for object localization on ILSVRC 2014 without training on any bounding box annotation. We demonstrate in a variety of experiments that our network is able to localize the discriminative image regions despite just being trained for solving classification task1."
]
} |
1708.06670 | 2963781258 | Deep convolutional neural networks (CNNs) have revolutionized the computer vision research and have seen unprecedented adoption for multiple tasks, such as classification, detection, and caption generation. However, they offer little transparency into their inner workings and are often treated as black boxes that deliver excellent performance. In this paper, we aim at alleviating this opaqueness of CNNs by providing visual explanations for the network’s predictions. Our approach can analyze a variety of CNN-based models trained for computer vision applications, such as object recognition and caption generation. Unlike the existing methods, we achieve this via unraveling the forward pass operation. The proposed method exploits feature dependencies across the layer hierarchy and uncovers the discriminative image locations that guide the network’s predictions. We name these locations CNN fixations, loosely analogous to human eye fixations. Our approach is a generic method that requires no architectural changes, additional training, or gradient computation, and computes the important image locations (CNN fixations). We demonstrate through a variety of applications that our approach is able to localize the discriminative image locations across different network architectures, diverse vision tasks, and data modalities. | Another set of works (e.g. @cite_18 @cite_3 @cite_5 @cite_10 ) take a different approach and assign a relevance score for each feature with respect to a class. The underlying idea is to estimate how the prediction changes if a feature is absent. Large difference in prediction indicates that the feature is important for prediction and small changes do not affect the decision. @cite_10 , authors find out the probabilistic contribution of each image patch to the confidence of a classifier and then they incorporate the neighborhood information to improve their weakly supervised saliency prediction. Zhang @cite_39 compute top down attention maps at different layers in the neural networks via a probabilistic winner takes all framework. They compute marginal winning probabilities for neurons at each layer by exploring feature expectancies. At each layer, the attention map is computed as the sum of these probabilities across the feature maps. | {
"cite_N": [
"@cite_18",
"@cite_3",
"@cite_39",
"@cite_5",
"@cite_10"
],
"mid": [
"2113211233",
"1787224781",
"2503388974",
"2962680264",
"2442791716"
],
"abstract": [
"We present a method for explaining predictions for individual instances. The presented approach is general and can be used with all classification models that output probabilities. It is based on the decomposition of a model's predictions on individual contributions of each attribute. Our method works for the so-called black box models such as support vector machines, neural networks, and nearest neighbor algorithms, as well as for ensemble methods such as boosting and random forests. We demonstrate that the generated explanations closely follow the learned models and present a visualization technique that shows the utility of our approach and enables the comparison of different prediction methods.",
"Understanding and interpreting classification decisions of automated image classification systems is of high value in many applications, as it allows to verify the reasoning of the system and provides additional information to the human expert. Although machine learning methods are solving very successfully a plethora of tasks, they have in most cases the disadvantage of acting as a black box, not providing any information about what made them arrive at a particular decision. This work proposes a general solution to the problem of understanding classification decisions by pixel-wise decomposition of nonlinear classifiers. We introduce a methodology that allows to visualize the contributions of single pixels to predictions for kernel-based classifiers over Bag of Words features and for multilayered neural networks. These pixel contributions can be visualized as heatmaps and are provided to a human expert who can intuitively not only verify the validity of the classification decision, but also focus further analysis on regions of potential interest. We evaluate our method for classifiers trained on PASCAL VOC 2009 images, synthetic image data containing geometric shapes, the MNIST handwritten digits data set and for the pre-trained ImageNet model available as part of the Caffe open source package.",
"We aim to model the top-down attention of a convolutional neural network (CNN) classifier for generating task-specific attention maps. Inspired by a top-down human visual attention model, we propose a new backpropagation scheme, called Excitation Backprop, to pass along top-down signals downwards in the network hierarchy via a probabilistic Winner-Take-All process. Furthermore, we introduce the concept of contrastive attention to make the top-down attention maps more discriminative. We show a theoretic connection between the proposed contrastive attention formulation and the Class Activation Map computation. Efficient implementation of Excitation Backprop for common neural network layers is also presented. In experiments, we visualize the evidence of a model’s classification decision by computing the proposed top-down attention maps. For quantitative evaluation, we report the accuracy of our method in weakly supervised localization tasks on the MS COCO, PASCAL VOC07 and ImageNet datasets. The usefulness of our method is further validated in the text-to-region association task. On the Flickr30k Entities dataset, we achieve promising performance in phrase localization by leveraging the top-down attention of a CNN model that has been trained on weakly labeled web images. Finally, we demonstrate applications of our method in model interpretation and data annotation assistance for facial expression analysis and medical imaging tasks.",
"This article presents the prediction difference analysis method for visualizing the response of a deep neural network to a specific input. When classifying images, the method highlights areas in a given input image that provide evidence for or against a certain class. It overcomes several shortcoming of previous methods and provides great additional insight into the decision making process of classifiers. Making neural network decisions interpretable through visualization is important both to improve models and to accelerate the adoption of black-box classifiers in application areas such as medicine. We illustrate the method in experiments on natural images (ImageNet data), as well as medical images (MRI brain scans).",
"Top-down saliency models produce a probability map that peaks at target locations specified by a task goal such as object detection. They are usually trained in a supervised setting involving annotations of objects. We propose a weakly supervised top-down saliency framework using only binary labels that indicate the presence absence of an object in an image. First, the probabilistic contribution of each image patch to the confidence of an ScSPM-based classifier produces a Reverse-ScSPM (R-ScSPM) saliency map. Neighborhood information is then incorporated through a contextual saliency map which is estimated using logistic regression learnt on patches having high R-ScSPM saliency. Both the saliency maps are combined to obtain the final saliency map. We evaluate the performance of the proposed weakly supervised top-down saliency and achieves comparable performance with fully supervised approaches. Experiments are carried out on 5 challenging datasets across 3 different applications."
]
} |
1708.06720 | 2749425057 | Imagery texts are usually organized as a hierarchy of several visual elements, i.e. characters, words, text lines and text blocks. Among these elements, character is the most basic one for various languages such as Western, Chinese, Japanese, mathematical expression and etc. It is natural and convenient to construct a common text detection engine based on character detectors. However, training character detectors requires a vast of location annotated characters, which are expensive to obtain. Actually, the existing real text datasets are mostly annotated in word or line level. To remedy this dilemma, we propose a weakly supervised framework that can utilize word annotations, either in tight quadrangles or the more loose bounding boxes, for character detector training. When applied in scene text detection, we are thus able to train a robust character detector by exploiting word annotations in the rich large-scale real scene text datasets, e.g. ICDAR15 and COCO-text. The character detector acts as a key role in the pipeline of our text detection engine. It achieves the state-of-the-art performance on several challenging scene text detection benchmarks. We also demonstrate the flexibility of our pipeline by various scenarios, including deformed text detection and math expression recognition. | As mentioned earlier, character is a natural choice to build common detection engines. Nearly all existing character based methods rely on synthetic datasets for training @cite_22 @cite_41 @cite_10 @cite_36 @cite_47 @cite_37 , because of lacking character level annotated real data. However, synthetic data cannot have a full coverage of characters from various scenes, limiting the model's performance in representing challenging real scene texts. Actually, none of the current top methods for the popular ICDAR15 benchmark @cite_26 are based on character detection. Recently, some sophisticated synthetic technologies @cite_19 have been invented that the synthetic text images look more real''. Nevertheless, real text images are still indispensable in training more robust character models, as we will show in our experiments. | {
"cite_N": [
"@cite_37",
"@cite_26",
"@cite_22",
"@cite_41",
"@cite_36",
"@cite_19",
"@cite_47",
"@cite_10"
],
"mid": [
"2472159136",
"",
"",
"2217433794",
"",
"2952302849",
"70975097",
"1998042868"
],
"abstract": [
"We propose a system that finds text in natural scenes using a variety of cues. Our novel data-driven method incorporates coarse-to-fine detection of character pixels using convolutional features (Text-Conv), followed by extracting connected components (CCs) from characters using edge and color features, and finally performing a graph-based segmentation of CCs into words (Word-Graph). For Text-Conv, the initial detection is based on convolutional feature maps similar to those used in Convolutional Neural Networks (CNNs), but learned using Convolutional k-means. Convolution masks defined by local and neighboring patch features are used to improve detection accuracy. The Word-Graph algorithm uses contextual information to both improve word segmentation and prune false character word detections. Different definitions for foreground (text) regions are used to train the detection stages, some based on bounding box intersection, and others on bounding box and pixel intersection. Our system obtains pixel, character, and word detection f-measures of 93.14 , 90.26 , and 86.77 respectively for the ICDAR 2015 Robust Reading Focused Scene Text dataset, out-performing state-of-the-art systems. This approach may work for other detection targets with homogenous color in natural scenes.",
"",
"",
"Recent deep learning models have demonstrated strong capabilities for classifying text and non-text components in natural images. They extract a high-level feature globally computed from a whole image component (patch), where the cluttered background information may dominate true text features in the deep representation. This leads to less discriminative power and poorer robustness. In this paper, we present a new system for scene text detection by proposing a novel text-attentional convolutional neural network (Text-CNN) that particularly focuses on extracting text-related regions and features from the image components. We develop a new learning mechanism to train the Text-CNN with multi-level and rich supervised information, including text region mask, character label, and binary text non-text information. The rich supervision information enables the Text-CNN with a strong capability for discriminating ambiguous texts, and also increases its robustness against complicated background components. The training process is formulated as a multi-task learning problem, where low-level supervised information greatly facilitates the main task of text non-text classification. In addition, a powerful low-level detector called contrast-enhancement maximally stable extremal regions (MSERs) is developed, which extends the widely used MSERs by enhancing intensity contrast between text patterns and background. This allows it to detect highly challenging text patterns, resulting in a higher recall. Our approach achieved promising results on the ICDAR 2013 data set, with an F-measure of 0.82, substantially improving the state-of-the-art results.",
"",
"In this paper we introduce a new method for text detection in natural images. The method comprises two contributions: First, a fast and scalable engine to generate synthetic images of text in clutter. This engine overlays synthetic text to existing background images in a natural way, accounting for the local 3D scene geometry. Second, we use the synthetic images to train a Fully-Convolutional Regression Network (FCRN) which efficiently performs text detection and bounding-box regression at all locations and multiple scales in an image. We discuss the relation of FCRN to the recently-introduced YOLO detector, as well as other end-to-end object detection systems based on deep learning. The resulting detection network significantly out performs current methods for text detection in natural images, achieving an F-measure of 84.2 on the standard ICDAR 2013 benchmark. Furthermore, it can process 15 images per second on a GPU.",
"The goal of this work is text spotting in natural images. This is divided into two sequential tasks: detecting words regions in the image, and recognizing the words within these regions. We make the following contributions: first, we develop a Convolutional Neural Network (CNN) classifier that can be used for both tasks. The CNN has a novel architecture that enables efficient feature sharing (by using a number of layers in common) for text detection, character case-sensitive and insensitive classification, and bigram classification. It exceeds the state-of-the-art performance for all of these. Second, we make a number of technical changes over the traditional CNN architectures, including no downsampling for a per-pixel sliding window, and multi-mode learning with a mixture of linear models (maxout). Third, we have a method of automated data mining of Flickr, that generates word and character level annotations. Finally, these components are used together to form an end-to-end, state-of-the-art text spotting system. We evaluate the text-spotting system on two standard benchmarks, the ICDAR Robust Reading data set and the Street View Text data set, and demonstrate improvements over the state-of-the-art on multiple measures.",
"This paper focuses on the problem of word detection and recognition in natural images. The problem is significantly more challenging than reading text in scanned documents, and has only recently gained attention from the computer vision community. Sub-components of the problem, such as text detection and cropped image word recognition, have been studied in isolation [7, 4, 20]. However, what is unclear is how these recent approaches contribute to solving the end-to-end problem of word recognition. We fill this gap by constructing and evaluating two systems. The first, representing the de facto state-of-the-art, is a two stage pipeline consisting of text detection followed by a leading OCR engine. The second is a system rooted in generic object recognition, an extension of our previous work in [20]. We show that the latter approach achieves superior performance. While scene text recognition has generally been treated with highly domain-specific methods, our results demonstrate the suitability of applying generic computer vision methods. Adopting this approach opens the door for real world scene text recognition to benefit from the rapid advances that have been taking place in object recognition."
]
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.