aid stringlengths 9 15 | mid stringlengths 7 10 | abstract stringlengths 78 2.56k | related_work stringlengths 92 1.77k | ref_abstract dict |
|---|---|---|---|---|
1611.06824 | 2555927448 | We consider the problem of learning hierarchical policies for Reinforcement Learning able to discover options, an option corresponding to a sub-policy over a set of primitive actions. Different models have been proposed during the last decade that usually rely on a predefined set of options. We specifically address the problem of automatically discovering options in decision processes. We describe a new RL learning framework called Bi-POMDP, and a new learning model called Budgeted Option Neural Network (BONN) able to discover options based on a budgeted learning objective. Since Bi-POMDP are more general than POMDP, our model can also be used to discover options for classical RL tasks. The BONN model is evaluated on different classical RL problems, demonstrating both quantitative and qualitative interesting results. | The closest architecture to BONN is the Hierarchical Multiscale Recurrent Neural Network @cite_26 that discovers hierarchical structures in sequences. It uses a binary boundary detector learned with a straight-through estimator, similar to the acquisition model (see Section ) of BONN.i | {
"cite_N": [
"@cite_26"
],
"mid": [
"2510842514"
],
"abstract": [
"Learning both hierarchical and temporal representation has been among the long-standing challenges of recurrent neural networks. Multiscale recurrent neural networks have been considered as a promising approach to resolve this issue, yet there has been a lack of empirical evidence showing that this type of models can actually capture the temporal dependencies by discovering the latent hierarchical structure of the sequence. In this paper, we propose a novel multiscale approach, called the hierarchical multiscale recurrent neural networks, which can capture the latent hierarchical structure in the sequence by encoding the temporal dependencies with different timescales using a novel update mechanism. We show some evidence that our proposed multiscale architecture can discover underlying hierarchical structure in the sequences without using explicit boundary information. We evaluate our proposed model on character-level language modelling and handwriting sequence modelling."
]
} |
1611.06824 | 2555927448 | We consider the problem of learning hierarchical policies for Reinforcement Learning able to discover options, an option corresponding to a sub-policy over a set of primitive actions. Different models have been proposed during the last decade that usually rely on a predefined set of options. We specifically address the problem of automatically discovering options in decision processes. We describe a new RL learning framework called Bi-POMDP, and a new learning model called Budgeted Option Neural Network (BONN) able to discover options based on a budgeted learning objective. Since Bi-POMDP are more general than POMDP, our model can also be used to discover options for classical RL tasks. The BONN model is evaluated on different classical RL problems, demonstrating both quantitative and qualitative interesting results. | At last, some articles propose hierarchical policies based on . A first category of models is those that use open-loop policies i.e. do not use observation from the environment at every time step. @cite_19 propose a model that mixes open-loop and closed-loop control while considering that sensing incurs a cost. Some models focus on the problem of learning @cite_27 @cite_2 : in that case, a given state is mapped to a sequence of actions. Another category of models divides the state space into several components. For instance, the Abstract Hidden Markov Model @cite_25 is based on discrete options defined on each space region. @cite_11 use a low-level controller that has only access to the proprioceptive information, and a high-level controller has access to all observations. @cite_0 use a similar idea of factoring the state space into two components, and learn a stochastic neural network for the high-level controller. The of the BONN model, described in Section , is similar to (stochastic) macro-actions, but open-loop policies are rather limited in complex environments. The general BONN architecture is more comparable to works using two different observations, however, those models do not learn to use the high-level controller. | {
"cite_N": [
"@cite_0",
"@cite_19",
"@cite_27",
"@cite_2",
"@cite_25",
"@cite_11"
],
"mid": [
"2963286043",
"2162137442",
"1631187438",
"2442341664",
"1674411155",
"2534060593"
],
"abstract": [
"Deep reinforcement learning has achieved many impressive results in recent years. However, tasks with sparse rewards or long horizons continue to pose significant challenges. To tackle these important problems, we propose a general framework that first learns useful skills in a pre-training environment, and then leverages the acquired skills for learning faster in downstream tasks. Our approach brings together some of the strengths of intrinsic motivation and hierarchical methods: the learning of useful skill is guided by a single proxy reward, the design of which requires very minimal domain knowledge about the downstream tasks. Then a high-level policy is trained on top of these skills, providing a significant improvement of the exploration and allowing to tackle sparse rewards in the downstream tasks. To efficiently pre-train a large span of skills, we use Stochastic Neural Networks combined with an information-theoretic regularizer. Our experiments show that this combination is effective in learning a wide span of interpretable skills in a sample-efficient way, and can significantly boost the learning performance uniformly across a wide range of downstream tasks.",
"Closed-loop control relies on sensory feedback that is usually assumed to be free. But if sensing incurs a cost, it may be cost-effective to take sequences of actions in open-loop mode. We describe a reinforcement learning algorithm that learns to combine open-loop and closed-loop control when sensing incurs a cost. Although we assume reliable sensors, use of open-loop control means that actions must sometimes be taken when the current state of the controlled system is uncertain. This is a special case of the hidden-state problem in reinforcement learning, and to cope, our algorithm relies on short-term memory. The main result of the paper is a rule that significantly limits exploration of possible memory states by pruning memory states for which the estimated value of information is greater than its cost. We prove that this rule allows convergence to an optimal policy.",
"We investigate the use of temporally abstract actions, or macro-actions, in the solution of Markov decision processes. Unlike current models that combine both primitive actions and macro-actions and leave the state space unchanged, we propose a hierarchical model (using an abstract MDP) that works with macro-actions only, and that significantly reduces the size of the state space. This is achieved by treating macroactions as local policies that act in certain regions of state space, and by restricting states in the abstract MDP to those at the boundaries of regions. The abstract MDP approximates the original and can be solved more efficiently. We discuss several ways in which macro-actions can be generated to ensure good solution quality. Finally, we consider ways in which macro-actions can be reused to solve multiple, related MDPs; and we show that this can justify the computational overhead of macro-action generation.",
"We present a novel deep recurrent neural network architecture that learns to build implicit plans in an end-to-end manner by purely interacting with an environment in reinforcement learning setting. The network builds an internal plan, which is continuously updated upon observation of the next input from the environment. It can also partition this internal representation into contiguous sub- sequences by learning for how long the plan can be committed to - i.e. followed without re-planing. Combining these properties, the proposed model, dubbed STRategic Attentive Writer (STRAW) can learn high-level, temporally abstracted macro- actions of varying lengths that are solely learnt from data without any prior information. These macro-actions enable both structured exploration and economic computation. We experimentally demonstrate that STRAW delivers strong improvements on several ATARI games by employing temporally extended planning strategies (e.g. Ms. Pacman and Frostbite). It is at the same time a general algorithm that can be applied on any sequence data. To that end, we also show that when trained on text prediction task, STRAW naturally predicts frequent n-grams (instead of macro-actions), demonstrating the generality of the approach.",
"In this paper, we present a method for recognising an agent's behaviour in dynamic, noisy, uncertain domains, and across multiple levels of abstraction. We term this problem on-line plan recognition under uncertainty and view it generally as probabilistic inference on the stochastic process representing the execution of the agent's plan. Our contributions in this paper are twofold. In terms of probabilistic inference, we introduce the Abstract Hidden Markov Model (AHMM), a novel type of stochastic processes, provide its dynamic Bayesian network (DBN) structure and analyse the properties of this network. We then describe an application of the Rao-Blackwellised Particle Filter to the AHMM which allows us to construct an efficient, hybrid inference method for this model. In terms of plan recognition, we propose a novel plan recognition framework based on the AHMM as the plan execution model. The Rao-Blackwellised hybrid inference for AHMM can take advantage of the independence properties inherent in a model of plan execution, leading to an algorithm for online probabilistic plan recognition that scales well with the number of levels in the plan hierarchy. This illustrates that while stochastic models for plan execution can be complex, they exhibit special structures which, if exploited, can lead to efficient plan recognition algorithms. We demonstrate the usefulness of the AHMM framework via a behaviour recognition system in a complex spatial environment using distributed video surveillance data.",
"We study a novel architecture and training procedure for locomotion tasks. A high-frequency, low-level \"spinal\" network with access to proprioceptive sensors learns sensorimotor primitives by training on simple tasks. This pre-trained module is fixed and connected to a low-frequency, high-level \"cortical\" network, with access to all sensors, which drives behavior by modulating the inputs to the spinal network. Where a monolithic end-to-end architecture fails completely, learning with a pre-trained spinal module succeeds at multiple high-level tasks, and enables the effective exploration required to learn from sparse rewards. We test our proposed architecture on three simulated bodies: a 16-dimensional swimming snake, a 20-dimensional quadruped, and a 54-dimensional humanoid. Our results are illustrated in the accompanying video at this https URL"
]
} |
1611.06882 | 2550907189 | Many prediction problems can be phrased as inferences over local neighborhoods of graphs. The graph represents the interaction between entities, and the neighborhood of each entity contains information that allows the inferences or predictions. We present an approach for applying machine learning directly to such graph neighborhoods, yielding predicitons for graph nodes on the basis of the structure of their local neighborhood and the features of the nodes in it. Our approach allows predictions to be learned directly from examples, bypassing the step of creating and tuning an inference model or summarizing the neighborhoods via a fixed set of hand-crafted features. The approach is based on a multi-level architecture built from Long Short-Term Memory neural nets (LSTMs); the LSTMs learn how to summarize the neighborhood from data. We demonstrate the effectiveness of the proposed technique on a synthetic example and on real-world data related to crowdsourced grading, Bitcoin transactions, and Wikipedia edit reversions. | Several approaches have been proposed for summarizing graph structures in feature vectors. The algorithm node2vec @cite_4 enables the construction of feature vectors for graph nodes in such a way that the feature vector optimally represents the node's location in the graph. Specifically, the feature vector maximizes the a-posteriori probability of graph neighborhoods given the feature vector. The resulting feature vector thus summarizes a node's location in a graph, but it does not summarize the original features of the node, or of its neighbors. In contrast, the techniques we introduce allow us to feed to machine learning the node features of an entire graph neighborhood. | {
"cite_N": [
"@cite_4"
],
"mid": [
"2366141641"
],
"abstract": [
"Prediction tasks over nodes and edges in networks require careful effort in engineering features used by learning algorithms. Recent research in the broader field of representation learning has led to significant progress in automating prediction by learning the features themselves. However, present feature learning approaches are not expressive enough to capture the diversity of connectivity patterns observed in networks. Here we propose node2vec, an algorithmic framework for learning continuous feature representations for nodes in networks. In node2vec, we learn a mapping of nodes to a low-dimensional space of features that maximizes the likelihood of preserving network neighborhoods of nodes. We define a flexible notion of a node's network neighborhood and design a biased random walk procedure, which efficiently explores diverse neighborhoods. Our algorithm generalizes prior work which is based on rigid notions of network neighborhoods, and we argue that the added flexibility in exploring neighborhoods is the key to learning richer representations. We demonstrate the efficacy of node2vec over existing state-of-the-art techniques on multi-label classification and link prediction in several real-world networks from diverse domains. Taken together, our work represents a new way for efficiently learning state-of-the-art task-independent representations in complex networks."
]
} |
1611.06882 | 2550907189 | Many prediction problems can be phrased as inferences over local neighborhoods of graphs. The graph represents the interaction between entities, and the neighborhood of each entity contains information that allows the inferences or predictions. We present an approach for applying machine learning directly to such graph neighborhoods, yielding predicitons for graph nodes on the basis of the structure of their local neighborhood and the features of the nodes in it. Our approach allows predictions to be learned directly from examples, bypassing the step of creating and tuning an inference model or summarizing the neighborhoods via a fixed set of hand-crafted features. The approach is based on a multi-level architecture built from Long Short-Term Memory neural nets (LSTMs); the LSTMs learn how to summarize the neighborhood from data. We demonstrate the effectiveness of the proposed technique on a synthetic example and on real-world data related to crowdsourced grading, Bitcoin transactions, and Wikipedia edit reversions. | In DeepWalk @cite_9 , feature vectors for graph nodes are constructed by performing random walks from the nodes, and applying them various summarization techniques to the list of feature vectors of the visited nodes. This approach enables the consideration of variable-diameter neighborhoods, in contrast to our exploration, which proceeds strictly breath-first. In DeepWalk, the construction of the summarizing feature vector proceeds according to a chosen algorithm, and is not guided by backpropagation from the learning goal. In other words, the summarization is not learned from the overall ML task. In contrast, in our approach the summarization itself, carried out by the LSTMs, is learned via backpropagation from the goal. | {
"cite_N": [
"@cite_9"
],
"mid": [
"2154851992"
],
"abstract": [
"We present DeepWalk, a novel approach for learning latent representations of vertices in a network. These latent representations encode social relations in a continuous vector space, which is easily exploited by statistical models. DeepWalk generalizes recent advancements in language modeling and unsupervised feature learning (or deep learning) from sequences of words to graphs. DeepWalk uses local information obtained from truncated random walks to learn latent representations by treating walks as the equivalent of sentences. We demonstrate DeepWalk's latent representations on several multi-label network classification tasks for social networks such as BlogCatalog, Flickr, and YouTube. Our results show that DeepWalk outperforms challenging baselines which are allowed a global view of the network, especially in the presence of missing information. DeepWalk's representations can provide F1 scores up to 10 higher than competing methods when labeled data is sparse. In some experiments, DeepWalk's representations are able to outperform all baseline methods while using 60 less training data. DeepWalk is also scalable. It is an online learning algorithm which builds useful incremental results, and is trivially parallelizable. These qualities make it suitable for a broad class of real world applications such as network classification, and anomaly detection."
]
} |
1611.06882 | 2550907189 | Many prediction problems can be phrased as inferences over local neighborhoods of graphs. The graph represents the interaction between entities, and the neighborhood of each entity contains information that allows the inferences or predictions. We present an approach for applying machine learning directly to such graph neighborhoods, yielding predicitons for graph nodes on the basis of the structure of their local neighborhood and the features of the nodes in it. Our approach allows predictions to be learned directly from examples, bypassing the step of creating and tuning an inference model or summarizing the neighborhoods via a fixed set of hand-crafted features. The approach is based on a multi-level architecture built from Long Short-Term Memory neural nets (LSTMs); the LSTMs learn how to summarize the neighborhood from data. We demonstrate the effectiveness of the proposed technique on a synthetic example and on real-world data related to crowdsourced grading, Bitcoin transactions, and Wikipedia edit reversions. | Learning over graphs can be reduced to a standard machine-learning problem by summarizing the information available at each node in a fixed set of features. This has been done, for instance, with the goal of link prediction, consisting in predicting which users in a social network will collaborate or connect next @cite_26 . Graph summarization typically requires deep insight into the problem, in order to design the summary features. The multi-level LSTMs we propose here constitute a way of learning such graph summarization. | {
"cite_N": [
"@cite_26"
],
"mid": [
"2768375068"
],
"abstract": [
"Social network analysis has attracted much attention in recent years. Link prediction is a key research directions within this area. In this research, we study link prediction as a supervised learning task. Along the way, we identify a set of features that are key to the superior performance under the supervised learning setup. The identified features are very easy to compute, and at the same time surprisingly effective in solving the link prediction problem. We also explain the effectiveness of the features from their class density distribution. Then we compare different classes of supervised learning algorithms in terms of their prediction performance using various performance metrics, such as accuracy, precision-recall, F-values, squared error etc. with a 5-fold cross validation. Our results on two practical social network datasets shows that most of the well-known classification algorithms (decision tree, k-nn,multilayer perceptron, SVM, rbf network) can predict link with surpassing performances, but SVM defeats all of them with narrow margin in all different performance measures. Again, ranking of features with popular feature ranking algorithms shows that a small subset of features always plays a significant role in the link prediction job."
]
} |
1611.06882 | 2550907189 | Many prediction problems can be phrased as inferences over local neighborhoods of graphs. The graph represents the interaction between entities, and the neighborhood of each entity contains information that allows the inferences or predictions. We present an approach for applying machine learning directly to such graph neighborhoods, yielding predicitons for graph nodes on the basis of the structure of their local neighborhood and the features of the nodes in it. Our approach allows predictions to be learned directly from examples, bypassing the step of creating and tuning an inference model or summarizing the neighborhoods via a fixed set of hand-crafted features. The approach is based on a multi-level architecture built from Long Short-Term Memory neural nets (LSTMs); the LSTMs learn how to summarize the neighborhood from data. We demonstrate the effectiveness of the proposed technique on a synthetic example and on real-world data related to crowdsourced grading, Bitcoin transactions, and Wikipedia edit reversions. | Some recent work has looked at the problem of summarizing very large graphs into feature vectors @cite_8 . The goals (and methods) are thus different from those in the present paper, where the emphasis consists in considering nodes together with their immediate neighborhoods as input to machine learning. | {
"cite_N": [
"@cite_8"
],
"mid": [
"1888005072"
],
"abstract": [
"This paper studies the problem of embedding very large information networks into low-dimensional vector spaces, which is useful in many tasks such as visualization, node classification, and link prediction. Most existing graph embedding methods do not scale for real world information networks which usually contain millions of nodes. In this paper, we propose a novel network embedding method called the LINE,'' which is suitable for arbitrary types of information networks: undirected, directed, and or weighted. The method optimizes a carefully designed objective function that preserves both the local and global network structures. An edge-sampling algorithm is proposed that addresses the limitation of the classical stochastic gradient descent and improves both the effectiveness and the efficiency of the inference. Empirical experiments prove the effectiveness of the LINE on a variety of real-world information networks, including language networks, social networks, and citation networks. The algorithm is very efficient, which is able to learn the embedding of a network with millions of vertices and billions of edges in a few hours on a typical single machine. The source code of the LINE is available online https: github.com tangjianpku LINE ."
]
} |
1611.06678 | 2950554226 | The CNN-encoding of features from entire videos for the representation of human actions has rarely been addressed. Instead, CNN work has focused on approaches to fuse spatial and temporal networks, but these were typically limited to processing shorter sequences. We present a new video representation, called temporal linear encoding (TLE) and embedded inside of CNNs as a new layer, which captures the appearance and motion throughout entire videos. It encodes this aggregated information into a robust video feature representation, via end-to-end learning. Advantages of TLEs are: (a) they encode the entire video into a compact feature representation, learning the semantics and a discriminative feature space; (b) they are applicable to all kinds of networks like 2D and 3D CNNs for video classification; and (c) they model feature interactions in a more expressive way and without loss of information. We conduct experiments on two challenging human action datasets: HMDB51 and UCF101. The experiments show that TLE outperforms current state-of-the-art methods on both datasets. | Over the last two decades, several action recognition techniques in videos have been proposed by the vision community. Quite a few are concerned with effective representations using local spatio-temporal features, suc h as HOG3D @cite_19 , SIFT3D @cite_6 , HOF @cite_13 , ESURF @cite_9 , and MBH @cite_43 . Recently, IDT @cite_28 was proposed, which is currently the state-of-the-art among hand-crafted features. Despite this good performance, these features have several shortcomings: they are computationally expensive; they fail to capture semantic concepts; they lack discriminative capacity as well as scalability. To overcome such issues, several techniques have been proposed to model the temporal structure for action recognition, such as the actom sequence model @cite_26 which considers sequence of histograms; temporal action decomposition @cite_36 which exploits the temporal structure of human actions by temporally decomposing video frames; dynamic poselets @cite_4 which uses a relational model for action detection; and the temporal evolution of appearance representations @cite_37 which uses a ranking function capable of modeling the evolution of both appearance and motion over time. | {
"cite_N": [
"@cite_37",
"@cite_26",
"@cite_4",
"@cite_28",
"@cite_36",
"@cite_9",
"@cite_6",
"@cite_19",
"@cite_43",
"@cite_13"
],
"mid": [
"1926645898",
"2084341401",
"410625161",
"2105101328",
"1498368596",
"1534763723",
"2108333036",
"2024868105",
"",
"2142194269"
],
"abstract": [
"In this paper we present a method to capture video-wide temporal information for action recognition. We postulate that a function capable of ordering the frames of a video temporally (based on the appearance) captures well the evolution of the appearance within the video. We learn such ranking functions per video via a ranking machine and use the parameters of these as a new video representation. The proposed method is easy to interpret and implement, fast to compute and effective in recognizing a wide variety of actions. We perform a large number of evaluations on datasets for generic action recognition (Hollywood2 and HMDB51), fine-grained actions (MPII- cooking activities) and gestures (Chalearn). Results show that the proposed method brings an absolute improvement of 7–10 , while being compatible with and complementary to further improvements in appearance and local motion based methods.",
"We address the problem of localizing actions, such as opening a door, in hours of challenging video data. We propose a model based on a sequence of atomic action units, termed \"actoms,\" that are semantically meaningful and characteristic for the action. Our actom sequence model (ASM) represents an action as a sequence of histograms of actom-anchored visual features, which can be seen as a temporally structured extension of the bag-of-features. Training requires the annotation of actoms for action examples. At test time, actoms are localized automatically based on a nonparametric model of the distribution of actoms, which also acts as a prior on an action's temporal structure. We present experimental results on two recent benchmarks for action localization \"Coffee and Cigarettes\" and the \"DLSBP\" dataset. We also adapt our approach to a classification-by-localization set-up and demonstrate its applicability on the challenging \"Hollywood 2\" dataset. We show that our ASM method outperforms the current state of the art in temporal action localization, as well as baselines that localize actions with a sliding window method.",
"Action detection is of great importance in understanding human motion from video. Compared with action recognition, it not only recognizes action type, but also localizes its spatiotemporal extent. This paper presents a relational model for action detection, which first decomposes human action into temporal “key poses” and then further into spatial “action parts”. Specifically, we start by clustering cuboids around each human joint into dynamic-poselets using a new descriptor. The cuboids from the same cluster share consistent geometric and dynamic structure, and each cluster acts as a mixture of body parts. We then propose a sequential skeleton model to capture the relations among dynamic-poselets. This model unifies the tasks of learning the composites of mixture dynamic-poselets, the spatiotemporal structures of action parts, and the local model for each action part in a single framework. Our model not only allows to localize the action in a video stream, but also enables a detailed pose estimation of an actor. We formulate the model learning problem in a structured SVM framework and speed up model inference by dynamic programming. We conduct experiments on three challenging action detection datasets: the MSR-II dataset, the UCF Sports dataset, and the JHMDB dataset. The results show that our method achieves superior performance to the state-of-the-art methods on these datasets.",
"Recently dense trajectories were shown to be an efficient video representation for action recognition and achieved state-of-the-art results on a variety of datasets. This paper improves their performance by taking into account camera motion to correct them. To estimate camera motion, we match feature points between frames using SURF descriptors and dense optical flow, which are shown to be complementary. These matches are, then, used to robustly estimate a homography with RANSAC. Human motion is in general different from camera motion and generates inconsistent matches. To improve the estimation, a human detector is employed to remove these matches. Given the estimated camera motion, we remove trajectories consistent with it. We also use this estimation to cancel out camera motion from the optical flow. This significantly improves motion-based descriptors, such as HOF and MBH. Experimental results on four challenging action datasets (i.e., Hollywood2, HMDB51, Olympic Sports and UCF50) significantly outperform the current state of the art.",
"Much recent research in human activity recognition has focused on the problem of recognizing simple repetitive (walking, running, waving) and punctual actions (sitting up, opening a door, hugging). However, many interesting human activities are characterized by a complex temporal composition of simple actions. Automatic recognition of such complex actions can benefit from a good understanding of the temporal structures. We present in this paper a framework for modeling motion by exploiting the temporal structure of the human activities. In our framework, we represent activities as temporal compositions of motion segments. We train a discriminative model that encodes a temporal decomposition of video sequences, and appearance models for each motion segment. In recognition, a query video is matched to the model according to the learned appearances and motion segment decomposition. Classification is made based on the quality of matching between the motion segment classifiers and the temporal segments in the query sequence. To validate our approach, we introduce a new dataset of complex Olympic Sports activities. We show that our algorithm performs better than other state of the art methods.",
"Over the years, several spatio-temporal interest point detectors have been proposed. While some detectors can only extract a sparse set of scale-invariant features, others allow for the detection of a larger amount of features at user-defined scales. This paper presents for the first time spatio-temporal interest points that are at the same time scale-invariant (both spatially and temporally) and densely cover the video content. Moreover, as opposed to earlier work, the features can be computed efficiently. Applying scale-space theory, we show that this can be achieved by using the determinant of the Hessian as the saliency measure. Computations are speeded-up further through the use of approximative box-filter operations on an integral video structure. A quantitative evaluation and experimental results on action recognition show the strengths of the proposed detector in terms of repeatability, accuracy and speed, in comparison with previously proposed detectors.",
"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.",
"In this work, we present a novel local descriptor for video sequences. The proposed descriptor is based on histograms of oriented 3D spatio-temporal gradients. Our contribution is four-fold. (i) To compute 3D gradients for arbitrary scales, we develop a memory-efficient algorithm based on integral videos. (ii) We propose a generic 3D orientation quantization which is based on regular polyhedrons. (iii) We perform an in-depth evaluation of all descriptor parameters and optimize them for action recognition. (iv) We apply our descriptor to various action datasets (KTH, Weizmann, Hollywood) and show that we outperform the state-of-the-art.",
"",
"The aim of this paper is to address recognition of natural human actions in diverse and realistic video settings. This challenging but important subject has mostly been ignored in the past due to several problems one of which is the lack of realistic and annotated video datasets. Our first contribution is to address this limitation and to investigate the use of movie scripts for automatic annotation of human actions in videos. We evaluate alternative methods for action retrieval from scripts and show benefits of a text-based classifier. Using the retrieved action samples for visual learning, we next turn to the problem of action classification in video. We present a new method for video classification that builds upon and extends several recent ideas including local space-time features, space-time pyramids and multi-channel non-linear SVMs. The method is shown to improve state-of-the-art results on the standard KTH action dataset by achieving 91.8 accuracy. Given the inherent problem of noisy labels in automatic annotation, we particularly investigate and show high tolerance of our method to annotation errors in the training set. We finally apply the method to learning and classifying challenging action classes in movies and show promising results."
]
} |
1611.06678 | 2950554226 | The CNN-encoding of features from entire videos for the representation of human actions has rarely been addressed. Instead, CNN work has focused on approaches to fuse spatial and temporal networks, but these were typically limited to processing shorter sequences. We present a new video representation, called temporal linear encoding (TLE) and embedded inside of CNNs as a new layer, which captures the appearance and motion throughout entire videos. It encodes this aggregated information into a robust video feature representation, via end-to-end learning. Advantages of TLEs are: (a) they encode the entire video into a compact feature representation, learning the semantics and a discriminative feature space; (b) they are applicable to all kinds of networks like 2D and 3D CNNs for video classification; and (c) they model feature interactions in a more expressive way and without loss of information. We conduct experiments on two challenging human action datasets: HMDB51 and UCF101. The experiments show that TLE outperforms current state-of-the-art methods on both datasets. | As to prior encoding methods, there is a vast literature on BoW @cite_34 @cite_33 , Fisher vector encoding @cite_17 and sparse encoding @cite_12 . Such methods have performed very well in various vision tasks. FV encoding @cite_25 and VLAD @cite_41 @cite_42 have lately been integrated as a layer in ConvNet architectures, and CNN encoded features have produced superior results for several challenging tasks. Likewise, Bilinear models @cite_29 @cite_16 have been widely used and have achieved state-of-the-art results. Bilinear models are computationally expensive as they return matrix outer products, hence can lead to prohibitively high dimensions. To tackle this problem, compact bilinear pooling @cite_1 was proposed which uses the Tensor Sketch Algorithm @cite_21 , to project features from a high dimensional space to a lower dimensional one, while retaining state-of-the-art performances. Compact bilinear pooling has shown to perform better than FV encoding and fully-connected networks @cite_1 . Moreover, this type of feature representation is compact, non-redundant, avoids over-fitting, and reduces the number of parameters of CNNs significantly, as it replaces fully-connected layers. | {
"cite_N": [
"@cite_33",
"@cite_41",
"@cite_29",
"@cite_42",
"@cite_1",
"@cite_21",
"@cite_16",
"@cite_34",
"@cite_25",
"@cite_12",
"@cite_17"
],
"mid": [
"2131846894",
"2951019013",
"",
"1524680991",
"2261271299",
"2146897752",
"2170653751",
"1625255723",
"2500786414",
"2097018403",
"1606858007"
],
"abstract": [
"We describe an approach to object and scene retrieval which searches for and localizes all the occurrences of a user outlined object in a video. The object is represented by a set of viewpoint invariant region descriptors so that recognition can proceed successfully despite changes in viewpoint, illumination and partial occlusion. The temporal continuity of the video within a shot is used to track the regions in order to reject unstable regions and reduce the effects of noise in the descriptors. The analogy with text retrieval is in the implementation where matches on descriptors are pre-computed (using vector quantization), and inverted file systems and document rankings are used. The result is that retrieved is immediate, returning a ranked list of key frames shots in the manner of Google. The method is illustrated for matching in two full length feature films.",
"We tackle the problem of large scale visual place recognition, where the task is to quickly and accurately recognize the location of a given query photograph. We present the following three principal contributions. First, we develop a convolutional neural network (CNN) architecture that is trainable in an end-to-end manner directly for the place recognition task. The main component of this architecture, NetVLAD, is a new generalized VLAD layer, inspired by the \"Vector of Locally Aggregated Descriptors\" image representation commonly used in image retrieval. The layer is readily pluggable into any CNN architecture and amenable to training via backpropagation. Second, we develop a training procedure, based on a new weakly supervised ranking loss, to learn parameters of the architecture in an end-to-end manner from images depicting the same places over time downloaded from Google Street View Time Machine. Finally, we show that the proposed architecture significantly outperforms non-learnt image representations and off-the-shelf CNN descriptors on two challenging place recognition benchmarks, and improves over current state-of-the-art compact image representations on standard image retrieval benchmarks.",
"",
"Deep convolutional neural networks (CNN) have shown their promise as a universal representation for recognition. However, global CNN activations lack geometric invariance, which limits their robustness for classification and matching of highly variable scenes. To improve the invariance of CNN activations without degrading their discriminative power, this paper presents a simple but effective scheme called multi-scale orderless pooling (MOP-CNN). This scheme extracts CNN activations for local patches at multiple scale levels, performs orderless VLAD pooling of these activations at each level separately, and concatenates the result. The resulting MOP-CNN representation can be used as a generic feature for either supervised or unsupervised recognition tasks, from image classification to instance-level retrieval; it consistently outperforms global CNN activations without requiring any joint training of prediction layers for a particular target dataset. In absolute terms, it achieves state-of-the-art results on the challenging SUN397 and MIT Indoor Scenes classification datasets, and competitive results on ILSVRC2012 2013 classification and INRIA Holidays retrieval datasets.",
"Bilinear models has been shown to achieve impressive performance on a wide range of visual tasks, such as semantic segmentation, fine grained recognition and face recognition. However, bilinear features are high dimensional, typically on the order of hundreds of thousands to a few million, which makes them impractical for subsequent analysis. We propose two compact bilinear representations with the same discriminative power as the full bilinear representation but with only a few thousand dimensions. Our compact representations allow back-propagation of classification errors enabling an end-to-end optimization of the visual recognition system. The compact bilinear representations are derived through a novel kernelized analysis of bilinear pooling which provide insights into the discriminative power of bilinear pooling, and a platform for further research in compact pooling methods. Experimentation illustrate the utility of the proposed representations for image classification and few-shot learning across several datasets.",
"Approximation of non-linear kernels using random feature mapping has been successfully employed in large-scale data analysis applications, accelerating the training of kernel machines. While previous random feature mappings run in O(ndD) time for @math training samples in d-dimensional space and D random feature maps, we propose a novel randomized tensor product technique, called Tensor Sketching, for approximating any polynomial kernel in O(n(d+D D )) time. Also, we introduce both absolute and relative error bounds for our approximation to guarantee the reliability of our estimation algorithm. Empirically, Tensor Sketching achieves higher accuracy and often runs orders of magnitude faster than the state-of-the-art approach for large-scale real-world datasets.",
"Perceptual systems routinely separate “content” from “style,” classifying familiar words spoken in an unfamiliar accent, identifying a font or handwriting style across letters, or recognizing a familiar face or object seen under unfamiliar viewing conditions. Yet a general and tractable computational model of this ability to untangle the underlying factors of perceptual observations remains elusive (Hofstadter, 1985). Existing factor models (Mardia, Kent, & Bibby, 1979; Hinton & Zemel, 1994; Ghahramani, 1995; Bell & Sejnowski, 1995; Hinton, Dayan, Frey, & Neal, 1995; Dayan, Hinton, Neal, & Zemel, 1995; Hinton & Ghahramani, 1997) are either insufficiently rich to capture the complex interactions of perceptually meaningful factors such as phoneme and speaker accent or letter and font, or do not allow efficient learning algorithms. We present a general framework for learning to solve two-factor tasks using bilinear models, which provide sufficiently expressive representations of factor interactions but can nonetheless be fit to data using efficient algorithms based on the singular value decomposition and expectation-maximization. We report promising results on three different tasks in three different perceptual domains: spoken vowel classification with a benchmark multi-speaker database, extrapolation of fonts to unseen letters, and translation of faces to novel illuminants.",
"We present a novel method for generic visual categorization: the problem of identifying the object content of natural images while generalizing across variations inherent to the object class. This bag of keypoints method is based on vector quantization of affine invariant descriptors of image patches. We propose and compare two alternative implementations using different classifiers: Naive Bayes and SVM. The main advantages of the method are that it is simple, computationally efficient and intrinsically invariant. We present results for simultaneously classifying seven semantic visual categories. These results clearly demonstrate that the method is robust to background clutter and produces good categorization accuracy even without exploiting geometric information.",
"Despite the great success of convolutional neural networks (CNN) for the image classification task on datasets like Cifar and ImageNet, CNN's representation power is still somewhat limited in dealing with object images that have large variation in size and clutter, where Fisher Vector (FV) has shown to be an effective encoding strategy. FV encodes an image by aggregating local descriptors with a universal generative Gaussian Mixture Model (GMM). FV however has limited learning capability and its parameters are mostly fixed after constructing the codebook. To combine together the best of the two worlds, we propose in this paper a neural network structure with FV layer being part of an end-to-end trainable system that is differentiable; we name our network FisherNet that is learnable using backpropagation. Our proposed FisherNet combines convolutional neural network training and Fisher Vector encoding in a single end-to-end structure. We observe a clear advantage of FisherNet over plain CNN and standard FV in terms of both classification accuracy and computational efficiency on the challenging PASCAL VOC object classification task.",
"Recently SVMs using spatial pyramid matching (SPM) kernel have been highly successful in image classification. Despite its popularity, these nonlinear SVMs have a complexity O(n2 n3) in training and O(n) in testing, where n is the training size, implying that it is nontrivial to scaleup the algorithms to handle more than thousands of training images. In this paper we develop an extension of the SPM method, by generalizing vector quantization to sparse coding followed by multi-scale spatial max pooling, and propose a linear SPM kernel based on SIFT sparse codes. This new approach remarkably reduces the complexity of SVMs to O(n) in training and a constant in testing. In a number of image categorization experiments, we find that, in terms of classification accuracy, the suggested linear SPM based on sparse coding of SIFT descriptors always significantly outperforms the linear SPM kernel on histograms, and is even better than the nonlinear SPM kernels, leading to state-of-the-art performance on several benchmarks by using a single type of descriptors.",
"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."
]
} |
1611.06694 | 2952051353 | Deep neural networks with lots of parameters are typically used for large-scale computer vision tasks such as image classification. This is a result of using dense matrix multiplications and convolutions. However, sparse computations are known to be much more efficient. In this work, we train and build neural networks which implicitly use sparse computations. We introduce additional gate variables to perform parameter selection and show that this is equivalent to using a spike-and-slab prior. We experimentally validate our method on both small and large networks and achieve state-of-the-art compression results for sparse neural network models. | Our work is very similar to that of Architecture Learning @cite_10 , which uses a similar framework to minimize the total number of neurons in a neural network. On the other hand, we minimize the total number of weights. | {
"cite_N": [
"@cite_10"
],
"mid": [
"2507936800"
],
"abstract": [
"Deep neural networks with millions of parameters are at the heart of many state of the art machine learning models today. However, recent works have shown that models with much smaller number of parameters can also perform just as well. In this work, we introduce the problem of architecture-learning, i.e; learning the architecture of a neural network along with weights. We introduce a new trainable parameter called tri-state ReLU, which helps in eliminating unnecessary neurons. We also propose a smooth regularizer which encourages the total number of neurons after elimination to be small. The resulting objective is differentiable and simple to optimize. We experimentally validate our method on both small and large networks, and show that it can learn models with a considerably small number of parameters without affecting prediction accuracy."
]
} |
1611.06639 | 2950141408 | Recurrent Neural Network (RNN) is one of the most popular architectures used in Natural Language Processsing (NLP) tasks because its recurrent structure is very suitable to process variable-length text. RNN can utilize distributed representations of words by first converting the tokens comprising each text into vectors, which form a matrix. And this matrix includes two dimensions: the time-step dimension and the feature vector dimension. Then most existing models usually utilize one-dimensional (1D) max pooling operation or attention-based operation only on the time-step dimension to obtain a fixed-length vector. However, the features on the feature vector dimension are not mutually independent, and simply applying 1D pooling operation over the time-step dimension independently may destroy the structure of the feature representation. On the other hand, applying two-dimensional (2D) pooling operation over the two dimensions may sample more meaningful features for sequence modeling tasks. To integrate the features on both dimensions of the matrix, this paper explores applying 2D max pooling operation to obtain a fixed-length representation of the text. This paper also utilizes 2D convolution to sample more meaningful information of the matrix. Experiments are conducted on six text classification tasks, including sentiment analysis, question classification, subjectivity classification and newsgroup classification. Compared with the state-of-the-art models, the proposed models achieve excellent performance on 4 out of 6 tasks. Specifically, one of the proposed models achieves highest accuracy on Stanford Sentiment Treebank binary classification and fine-grained classification tasks. | The proposed model BLSTM-2DCNN is most relevant to DSCNN @cite_18 and RCNN @cite_8 . The difference is that the former two utilize LSTM, bidirectional RNN respectively, while this work applies BLSTM, to capture long-term sentence dependencies. After that the former two both apply 1D convolution and 1D max pooling operation, while this paper uses 2D convolution and 2D max pooling operation, to obtain the whole sentence representation. | {
"cite_N": [
"@cite_18",
"@cite_8"
],
"mid": [
"2462025561",
"2462509432"
],
"abstract": [
"The goal of sentence and document modeling is to accurately represent the meaning of sentences and documents for various Natural Language Processing tasks. In this work, we present Dependency Sensitive Convolutional Neural Networks (DSCNN) as a generalpurpose classification system for both sentences and documents. DSCNN hierarchically builds textual representations by processing pretrained word embeddings via Long ShortTerm Memory networks and subsequently extracting features with convolution operators. Compared with existing recursive neural models with tree structures, DSCNN does not rely on parsers and expensive phrase labeling, and thus is not restricted to sentencelevel tasks. Moreover, unlike other CNNbased models that analyze sentences locally by sliding windows, our system captures both the dependency information within each sentence and relationships across sentences in the same document. Experiment results demonstrate that our approach is achieving state-ofthe-art performance on several tasks, including sentiment analysis, question type classification, and subjectivity classification.",
"Recently, the rapid development of word embedding and neural networks has brought new inspiration to various NLP and IR tasks. In this paper, we describe a staged hybrid model combining Recurrent Convolutional Neural Networks (RCNN) with highway layers. The highway network module is incorporated in the middle takes the output of the bi-directional Recurrent Neural Network (Bi-RNN) module in the first stage and provides the Convolutional Neural Network (CNN) module in the last stage with the input. The experiment shows that our model outperforms common neural network models (CNN, RNN, Bi-RNN) on a sentiment analysis task. Besides, the analysis of how sequence length influences the RCNN with highway layers shows that our model could learn good representation for the long text."
]
} |
1611.06722 | 2554259713 | Transliterations play an important role in multilingual entity reference resolution, because proper names increasingly travel between languages in news and social media. Previous work associated with machine translation targets transliteration only single between language pairs, focuses on specific classes of entities (such as cities and celebrities) and relies on manual curation, which limits the expression power of transliteration in multilingual environment. By contrast, we present an unsupervised transliteration model covering 69 major languages that can generate good transliterations for arbitrary strings between any language pair. Our model yields top-(1, 20, 100) averages of (32.85 , 60.44 , 83.20 ) in matching gold standard transliteration compared to results from a recently-published system of (26.71 , 50.27 , 72.79 ). We also show the quality of our model in detecting true and false friends from Wikipedia high frequency lexicons. Our method indicates a strong signal of pronunciation similarity and boosts the probability of finding true friends in 68 out of 69 languages. | Transliteration research first associates with the field of orthographic similarity detections since sound similarities co-exist with orthographic similarities @cite_18 @cite_10 @cite_3 @cite_5 @cite_24 @cite_9 . This work shows reasonableness of character-based transliteration between close languages (i.e. languages sharing characters) but does not discuss on distant language pairs. | {
"cite_N": [
"@cite_18",
"@cite_9",
"@cite_3",
"@cite_24",
"@cite_5",
"@cite_10"
],
"mid": [
"1497358850",
"2108177527",
"2143263738",
"",
"2151338697",
"2091889711"
],
"abstract": [
"",
"In this article we want to investigate the semantic (figurative) structures that underlie false friends, especially semantic false friends, in various European languages (Spanish, French, German and English). Chance false friends share the same form but have different etymologies and different meanings in different languages. They can be compared to homonyms in a single natural language. Semantic false friends, by contrast, have the same etymological origin, their meanings differ in different language, but one can still detect semantic relations between them. They can be considered to be cross-linguistic equivalents to polysemous words in a single natural language. The links between their meanings in different languages can be based on metaphor, metonymy and euphemism, but also on specialisation and generalisation. Semantic false friends are the semantic relics of pragmatic language use over time and space. Studying false friends is, however, more than an exercise in diachronic pragmatics. It has important implications for translation and cross-linguistic communication, where an awareness of false friends is important together with knowledge of certain pragmatic strategies, which help to avoid misunderstandings or mistranslations. The study of the underlying figurative links between false friends also adds a new dimension to cognitive semantics.",
"Abstract In two experiments Dutch–English bilinguals were tested with English words varying in their degree of orthographic, phonological, and semantic overlap with Dutch words. Thus, an English word target could be spelled the same as a Dutch word and or could be a near-homophone of a Dutch word. Whether such form similarity was accompanied with semantic identity (translation equivalence) was also varied. In a progressive demasking task and a visual lexical decision task very similar results were obtained. Both tasks showed facilitatory effects of cross-linguistic orthographic and semantic similarity on response latencies to target words, but inhibitory effects of phonological overlap. A third control experiment involving English lexical decision with monolinguals indicated that these results were not due to specific characteristics of the stimulus material. The results are interpreted within an interactive activation model for monolingual and bilingual word recognition (the Bilingual Interactive Activation model) expanded with a phonological and a semantic component.",
"",
"A series of progressive demasking and lexical decision experiments investigated how the recognition of target words exclusively belonging to one language is affected by the existence of orthographic neighbors from the same or the other language of bilingual participants. Increasing the number of orthographic neighbors in Dutch systematically slowed response times to English target words in Dutch English bilinguals, while an increase in target language neighbors consistently produced inhibitory effects for Dutch and facilitatory effects for English target words. Monolingual English speakers also showed facilitation due to English neighbors, but no effect of Dutch neighbors. The experiments provide evidence for parallel activation of words in an integrated Dutch English lexicon. An implemented version of such a model making these assumptions, the Bilingual Interactive Activation (BIA) model, is shown to account for the overall pattern of results. © 1998 Academic Press One of the striking features of bilingual language performance is the apparent ease with which the bilingual manages to keep interference from the non-target language at a minimal level. The fact remains, however, that interference from one language to the other does occur and is observable with respect to both language structure and linguistic processing. For example, in language production, interference from the first language can be noticed both at the phonological level (foreign accents) and at the sentence level (borrowed syntax), as well as in intrusions of words from the other language (accidental lexical borrowings).",
"This paper presents a method for inducing translation lexicons based on transduction models of cognate pairs via bridge languages. Bilingual lexicons within languages families are induced using probabilistic string edit distance models. Translation lexicons for arbitrary distant language pairs are then generated by a combination of these intra-family translation models and one or more cross-family on-line dictionaries. Up to 95 exact match accuracy is achieved on the target vocabulary (30-68 of inter-family test pairs). Thus substantial portions of translation lexicons can be generated accurately for languages where no bilingual dictionary or parallel corpora may exist."
]
} |
1611.06722 | 2554259713 | Transliterations play an important role in multilingual entity reference resolution, because proper names increasingly travel between languages in news and social media. Previous work associated with machine translation targets transliteration only single between language pairs, focuses on specific classes of entities (such as cities and celebrities) and relies on manual curation, which limits the expression power of transliteration in multilingual environment. By contrast, we present an unsupervised transliteration model covering 69 major languages that can generate good transliterations for arbitrary strings between any language pair. Our model yields top-(1, 20, 100) averages of (32.85 , 60.44 , 83.20 ) in matching gold standard transliteration compared to results from a recently-published system of (26.71 , 50.27 , 72.79 ). We also show the quality of our model in detecting true and false friends from Wikipedia high frequency lexicons. Our method indicates a strong signal of pronunciation similarity and boosts the probability of finding true friends in 68 out of 69 languages. | Similarly, work on cognate identification also focus on close language pairs @cite_16 @cite_0 @cite_15 @cite_13 @cite_19 @cite_6 @cite_30 . However, we believe multilingual transliterations contribute to even distant languages (e.g. English and Japanese) when handling OOV words and resolving ambiguities. | {
"cite_N": [
"@cite_30",
"@cite_6",
"@cite_0",
"@cite_19",
"@cite_15",
"@cite_16",
"@cite_13"
],
"mid": [
"1659833910",
"",
"",
"1546602224",
"2064550430",
"235841388",
"2126373519"
],
"abstract": [
"This article presents a measure of semantic similarity in an IS-A taxonomy based on the notion of shared information content. Experimental evaluation against a benchmark set of human similarity judgments demonstrates that the measure performs better than the traditional edge-counting approach. The article presents algorithms that take advantage of taxonomic similarity in resolving syntactic and semantic ambiguity, along with experimental results demonstrating their effectiveness.",
"",
"",
"This paper 1 presents DISCO, a tool for retrieving the distributional similarity be- tween two given words, and for retrieving the distributionally most similar words for a given word. Pre-computed word spaces are freely available for a number of languages including English, German, French and Italian, so DISCO can be used off the shelf. The tool is imple- mented in Java, provides a Java API, and can also be called from the command line. The per- formance of DISCO is evaluated by measuring the correlation with WordNet-based semantic similarities and with human relatedness judgements. The evaluations show that DISCO has a higher correlation with semantic similarities derived from WordNet than latent semantic analysis (LSA) and the web-based PMI-IR.",
"The coinciding form and meaning similarity of cognates, e.g. ‘flamme’ (French), ‘Flamme’ (German), ‘vlam’ (Dutch), meaning ‘flame’ in English, facilitates learning of additional languages. The cross-language frequency and similarity distributions of cognates vary according to evolutionary change and language contact. We compare frequency and orthographic (O), phonetic (P), and semantic similarity of cognates, automatically identified in semi-complete lexicons of six widely spoken languages. Comparisons of P and O similarity reveal inconsistent mappings in language pairs with deep orthographies. The frequency distributions show that cognate frequency is reduced in less closely related language pairs as compared to more closely related languages (e.g., French-English vs. German-English). These frequency and similarity patterns may support a better understanding of cognate processing in natural and experimental settings. The automatically identified cognates are available in the supplementary materials, including the frequency and similarity measurements.",
"A pair of pressure sensing probe tubes are placed in the fan discharge of a turbofan engine, aft of the outlet guide vanes (OGVs) the tubes being circumferentially spaced by a fraction of the OGV pitch. A comparing means receives both pressures sensed by the double-barrelled probes and aerodynamically actuates to automatically select the higher of the two pressures. An accurate pressure indication representative of the mainstream pressure is therefore obtained despite the localized lower pressure turbulence area that may be caused by the presence of the upstream outlet guide vane(s).",
"When participants are asked to translate an ambiguous word, they are slower and less accurate than in the case of single-translation words (e.g., Laxen & Lavour, 2010; Tokowicz & Kroll, 2007). We report an experiment to further examine this multiple-translation effect by investigating the influence of variables shown to be relevant in bilingual processing. The experiment included cognates and non-cognates with one translation or with multiple translations. The latter were presented with their dominant or subordinate translations. Highly-proficient balanced bilinguals responded to a translation recognition task in the two language directions (Catalan–Spanish and Spanish–Catalan). The results showed a significant multiple-translation effect in both cognates and non-cognates. Moreover, this effect was obtained regardless of language dominance and translation direction. Participants were faster and more accurate when performing translation recognition for the dominant than for the subordinate translations. The findings are interpreted adopting the Distributed Representation Model (de Groot, 1992b)."
]
} |
1611.06722 | 2554259713 | Transliterations play an important role in multilingual entity reference resolution, because proper names increasingly travel between languages in news and social media. Previous work associated with machine translation targets transliteration only single between language pairs, focuses on specific classes of entities (such as cities and celebrities) and relies on manual curation, which limits the expression power of transliteration in multilingual environment. By contrast, we present an unsupervised transliteration model covering 69 major languages that can generate good transliterations for arbitrary strings between any language pair. Our model yields top-(1, 20, 100) averages of (32.85 , 60.44 , 83.20 ) in matching gold standard transliteration compared to results from a recently-published system of (26.71 , 50.27 , 72.79 ). We also show the quality of our model in detecting true and false friends from Wikipedia high frequency lexicons. Our method indicates a strong signal of pronunciation similarity and boosts the probability of finding true friends in 68 out of 69 languages. | Further transliteration researches divide into two branches. One tries to study delicate sound changing rules of specific languages @cite_7 @cite_11 @cite_31 @cite_12 @cite_4 @cite_28 @cite_22 . Especially, an excellent ideas of using Wikipedia external links is proposed in @cite_29 @cite_27 and achieve promising results in English-Hebrew transliteration using Moses @cite_25 . However, all these systems are supervised and require extra linguistic background knowledge during processing. Plus, only one among this work evaluates transliteration on up to 4 languages and it is hard to generalize for multiple languages. | {
"cite_N": [
"@cite_4",
"@cite_22",
"@cite_7",
"@cite_28",
"@cite_29",
"@cite_27",
"@cite_31",
"@cite_25",
"@cite_12",
"@cite_11"
],
"mid": [
"2055543848",
"2162070646",
"",
"205211590",
"2170096403",
"72691019",
"",
"2595715041",
"1563762850",
"2151858239"
],
"abstract": [
"We address the problem of transliterating English names using Chinese orthography in support of cross-lingual speech and text processing applications. We demonstrate the application of statistical machine translation techniques to \"translate\" the phonemic representation of an English name, obtained by using an automatic text-to-speech system, to a sequence of initials and finals, commonly used sub-word units of pronunciation for Chinese. We then use another statistical translation model to map the initial final sequence to Chinese characters. We also present an evaluation of this module in retrieval of Mandarin spoken documents from the TDT corpus using English text queries.",
"This paper presents a hybrid approach to English-Korean name transliteration. The base system is built on MOSES with enabled factored translation features. We expand the base system by combining with various transliteration methods including a Web-based n-best re-ranking, a dictionary-based method, and a rule-based method. Our standard run and best non-standard run achieve 45.1 and 78.5, respectively, in top-1 accuracy. Experimental results show that expanding training data size significantly contributes to the performance. Also we discover that the Web-based re-ranking method can be successfully applied to the English-Korean transliteration.",
"",
"In this paper, we address the problem of building a multilingual transliteration system using an interlingual representation. Our approach uses international phonetic alphabet (IPA) to learn the interlingual representation and thus allows us to use any word and its IPA representation as a training example. Thus, our approach requires only monolingual resources: a phoneme dictionary that lists words and their IPA representations. By adding a phoneme dictionary of a new language, we can readily build a transliteration system into any of the existing previous languages, without the expense of all-pairs data or computation. We also propose a regularization framework for learning the interlingual representation, which accounts for language specific phonemic variability, and thus it can find better mappings between languages. Experimental results on the name transliteration task in five diverse languages show a maximum improvement of 29 accuracy and an average improvement of 17 accuracy compared to a state-of-the-art baseline system.",
"We present a Hebrew to English transliteration method in the context of a machine translation system. Our method uses machine learning to determine which terms are to be transliterated rather than translated. The training corpus for this purpose includes only positive examples, acquired semi-automatically. Our classifier reduces more than 38 of the errors made by a baseline method. The identified terms are then transliterated. We present an SMT-based transliteration model trained with a parallel corpus extracted from Wikipedia using a fairly simple method which requires minimal knowledge. The correct result is produced in more than 76 of the cases, and in 92 of the instances it is one of the top-5 results. We also demonstrate a small improvement in the performance of a Hebrew-to-English MT system that uses our transliteration module.",
"Transliteration is the rendering in one language of terms from another language (and, possibly, another writing system), approximating spelling and or phonetic equivalents between the two languages. A transliteration dictionary is a crucial resource for a variety of natural language applications, most notably machine translation. We describe a general method for creating bilingual transliteration dictionaries from Wikipedia article titles. The method can be applied to any language pair present on Wikipedia, independently of the writing systems involved, and requires only a single simple resource that can be provided by any literate bilingual speaker. It was successfully applied to extract a Hebrew-English transliteration dictionary which was used to train a transliteration module that, when incorporated in a machine translation system, indeed improved its performance.",
"",
"",
"A proper noun dictionary is never complete rendering name translation from English to Chinese ineffective. One way to solve this problem is not to rely on a dictionary alone but to adopt automatic translation according to pronunciation similarities, i.e. to map phonemes comprising an English name to the phonetic representations of the corresponding Chinese name. This process is called transliteration. We present a statistical transliteration method. An efficient algorithm for aligning phoneme chunks is described. Unlike rule-based approaches, our method is data-driven. Compared to source-channel based statistical approaches, we adopt a direct transliteration model, i.e. the direction of probabilistic estimation conforms to the transliteration direction. We demonstrate comparable performance to source-channel based system.",
"Out of vocabulary (OOV) words are problematic for cross language information retrieval. One way to deal with OOV words when the two languages have different alphabets, is to transliterate the unknown words, that is, to render them in the orthography of the second language. In the present study, we present a simple statistical technique to train an English to Arabic transliteration model from pairs of names. We call this a selected n-gram model because a two-stage training procedure first learns which n-gram segments should be added to the unigram inventory for the source language, and then a second stage learns the translation model over this inventory. This technique requires no heuristics or linguistic knowledge of either language. We evaluate the statistically-trained model and a simpler hand-crafted model on a test set of named entities from the Arabic AFP corpus and demonstrate that they perform better than two online translation sources. We also explore the effectiveness of these systems on the TREC 2002 cross language IR task. We find that transliteration either of OOV named entities or of all OOV words is an effective approach for cross language IR."
]
} |
1611.06492 | 2551432385 | Deep Neural Network architectures with external memory components allow the model to perform inference and capture long term dependencies, by storing information explicitly. In this paper, we generalize Key-Value Memory Networks to a multimodal setting, introducing a novel key-addressing mechanism to deal with sequence-to-sequence models. The advantages of the framework are demonstrated on the task of video captioning, i.e generating natural language descriptions for videos. Conditioning on the previous time-step attention distributions for the key-value memory slots, we introduce a temporal structure in the memory addressing schema. The proposed model naturally decomposes the problem of video captioning into vision and language segments, dealing with them as key-value pairs. More specifically, we learn a semantic embedding (v) corresponding to each frame (k) in the video, thereby creating (k, v) memory slots. This allows us to exploit the temporal dependencies at multiple hierarchies (in the recurrent key-addressing; and in the language decoder). Exploiting this flexibility of the framework, we additionally capture spatial dependencies while mapping from the visual to semantic embedding. Extensive experiments on the Youtube2Text dataset demonstrate usefulness of recurrent key-addressing, while achieving competitive scores on BLEU@4, METEOR metrics against state-of-the-art models. | Following the success of end-to-end neural architectures and attention mechanisms, there is a growing body of literature for captioning tasks, in images and more recently videos. To deal with the multimodal nature of the problem, classical approaches relied on manually engineered templates @cite_0 @cite_52 . And while some recent approaches in this direction show promise @cite_35 , but the models lack generalization to deal with complex scenes, videos. | {
"cite_N": [
"@cite_0",
"@cite_35",
"@cite_52"
],
"mid": [
"1969616664",
"2949769367",
"1995820507"
],
"abstract": [
"We present a system to automatically generate natural language descriptions from images. This system consists of two parts. The first part, content planning, smooths the output of computer vision-based detection and recognition algorithms with statistics mined from large pools of visually descriptive text to determine the best content words to use to describe an image. The second step, surface realization, chooses words to construct natural language sentences based on the predicted content and general statistics from natural language. We present multiple approaches for the surface realization step and evaluate each using automatic measures of similarity to human generated reference descriptions. We also collect forced choice human evaluations between descriptions from the proposed generation system and descriptions from competing approaches. The proposed system is very effective at producing relevant sentences for images. It also generates descriptions that are notably more true to the specific image content than previous work.",
"This paper presents a novel approach for automatically generating image descriptions: visual detectors, language models, and multimodal similarity models learnt directly from a dataset of image captions. We use multiple instance learning to train visual detectors for words that commonly occur in captions, including many different parts of speech such as nouns, verbs, and adjectives. The word detector outputs serve as conditional inputs to a maximum-entropy language model. The language model learns from a set of over 400,000 image descriptions to capture the statistics of word usage. We capture global semantics by re-ranking caption candidates using sentence-level features and a deep multimodal similarity model. Our system is state-of-the-art on the official Microsoft COCO benchmark, producing a BLEU-4 score of 29.1 . When human judges compare the system captions to ones written by other people on our held-out test set, the system captions have equal or better quality 34 of the time.",
"The problem of describing images through natural language has gained importance in the computer vision community. Solutions to image description have either focused on a top-down approach of generating language through combinations of object detections and language models or bottom-up propagation of keyword tags from training images to test images through probabilistic or nearest neighbor techniques. In contrast, describing videos with natural language is a less studied problem. In this paper, we combine ideas from the bottom-up and top-down approaches to image description and propose a method for video description that captures the most relevant contents of a video in a natural language description. We propose a hybrid system consisting of a low level multimodal latent topic model for initial keyword annotation, a middle level of concept detectors and a high level module to produce final lingual descriptions. We compare the results of our system to human descriptions in both short and long forms on two datasets, and demonstrate that final system output has greater agreement with the human descriptions than any single level."
]
} |
1611.06492 | 2551432385 | Deep Neural Network architectures with external memory components allow the model to perform inference and capture long term dependencies, by storing information explicitly. In this paper, we generalize Key-Value Memory Networks to a multimodal setting, introducing a novel key-addressing mechanism to deal with sequence-to-sequence models. The advantages of the framework are demonstrated on the task of video captioning, i.e generating natural language descriptions for videos. Conditioning on the previous time-step attention distributions for the key-value memory slots, we introduce a temporal structure in the memory addressing schema. The proposed model naturally decomposes the problem of video captioning into vision and language segments, dealing with them as key-value pairs. More specifically, we learn a semantic embedding (v) corresponding to each frame (k) in the video, thereby creating (k, v) memory slots. This allows us to exploit the temporal dependencies at multiple hierarchies (in the recurrent key-addressing; and in the language decoder). Exploiting this flexibility of the framework, we additionally capture spatial dependencies while mapping from the visual to semantic embedding. Extensive experiments on the Youtube2Text dataset demonstrate usefulness of recurrent key-addressing, while achieving competitive scores on BLEU@4, METEOR metrics against state-of-the-art models. | As an alternative approach, @cite_4 @cite_36 suggest learning a joint visual-semantic embedding, effectively a mapping from the visual to language space. The motivation of our work is strongly aligned with @cite_17 , who generate semantic representations for images using CRF models, as context for the language decoder. However, our approach significantly differs in the essence that we capture spatio-temporal dynamics in videos while generating the text description. | {
"cite_N": [
"@cite_36",
"@cite_4",
"@cite_17"
],
"mid": [
"",
"2123024445",
"2110933980"
],
"abstract": [
"",
"Modern visual recognition systems are often limited in their ability to scale to large numbers of object categories. This limitation is in part due to the increasing difficulty of acquiring sufficient training data in the form of labeled images as the number of object categories grows. One remedy is to leverage data from other sources - such as text data - both to train visual models and to constrain their predictions. In this paper we present a new deep visual-semantic embedding model trained to identify visual objects using both labeled image data as well as semantic information gleaned from unannotated text. We demonstrate that this model matches state-of-the-art performance on the 1000-class ImageNet object recognition challenge while making more semantically reasonable errors, and also show that the semantic information can be exploited to make predictions about tens of thousands of image labels not observed during training. Semantic knowledge improves such zero-shot predictions achieving hit rates of up to 18 across thousands of novel labels never seen by the visual model.",
"Humans use rich natural language to describe and communicate visual perceptions. In order to provide natural language descriptions for visual content, this paper combines two important ingredients. First, we generate a rich semantic representation of the visual content including e.g. object and activity labels. To predict the semantic representation we learn a CRF to model the relationships between different components of the visual input. And second, we propose to formulate the generation of natural language as a machine translation problem using the semantic representation as source language and the generated sentences as target language. For this we exploit the power of a parallel corpus of videos and textual descriptions and adapt statistical machine translation to translate between our two languages. We evaluate our video descriptions on the TACoS dataset, which contains video snippets aligned with sentence descriptions. Using automatic evaluation and human judgments we show significant improvements over several baseline approaches, motivated by prior work. Our translation approach also shows improvements over related work on an image description task."
]
} |
1611.06492 | 2551432385 | Deep Neural Network architectures with external memory components allow the model to perform inference and capture long term dependencies, by storing information explicitly. In this paper, we generalize Key-Value Memory Networks to a multimodal setting, introducing a novel key-addressing mechanism to deal with sequence-to-sequence models. The advantages of the framework are demonstrated on the task of video captioning, i.e generating natural language descriptions for videos. Conditioning on the previous time-step attention distributions for the key-value memory slots, we introduce a temporal structure in the memory addressing schema. The proposed model naturally decomposes the problem of video captioning into vision and language segments, dealing with them as key-value pairs. More specifically, we learn a semantic embedding (v) corresponding to each frame (k) in the video, thereby creating (k, v) memory slots. This allows us to exploit the temporal dependencies at multiple hierarchies (in the recurrent key-addressing; and in the language decoder). Exploiting this flexibility of the framework, we additionally capture spatial dependencies while mapping from the visual to semantic embedding. Extensive experiments on the Youtube2Text dataset demonstrate usefulness of recurrent key-addressing, while achieving competitive scores on BLEU@4, METEOR metrics against state-of-the-art models. | A major drawback of the above approach is induced by mean pooling, where context features across image frames are collapsed. For one, this looses the temporal structure across frames by treating them as "bag-of-images" model. Addressing this, @cite_18 propose Sequence-to-Sequence models for accounting for the temporal structure, and @cite_1 extend it to a video-captioning setting. However, passing a fixed vector as context at each time step, creates a bottleneck for the flow of gradients using Backpropagation Through Time (BPTT) @cite_10 at the encoder. | {
"cite_N": [
"@cite_10",
"@cite_18",
"@cite_1"
],
"mid": [
"2150355110",
"2949888546",
"2950019618"
],
"abstract": [
"Basic backpropagation, which is a simple method now being widely used in areas like pattern recognition and fault diagnosis, is reviewed. The basic equations for backpropagation through time, and applications to areas like pattern recognition involving dynamic systems, systems identification, and control are discussed. Further extensions of this method, to deal with systems other than neural networks, systems involving simultaneous equations, or true recurrent networks, and other practical issues arising with the method are described. Pseudocode is provided to clarify the algorithms. The chain rule for ordered derivatives-the theorem which underlies backpropagation-is briefly discussed. The focus is on designing a simpler version of backpropagation which can be translated into computer code and applied directly by neutral network users. >",
"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.",
"Real-world videos often have complex dynamics; and methods for generating open-domain video descriptions should be sensitive to temporal structure and allow both input (sequence of frames) and output (sequence of words) of variable length. To approach this problem, we propose a novel end-to-end sequence-to-sequence model to generate captions for videos. For this we exploit recurrent neural networks, specifically LSTMs, which have demonstrated state-of-the-art performance in image caption generation. Our LSTM model is trained on video-sentence pairs and learns to associate a sequence of video frames to a sequence of words in order to generate a description of the event in the video clip. Our model naturally is able to learn the temporal structure of the sequence of frames as well as the sequence model of the generated sentences, i.e. a language model. We evaluate several variants of our model that exploit different visual features on a standard set of YouTube videos and two movie description datasets (M-VAD and MPII-MD)."
]
} |
1611.06492 | 2551432385 | Deep Neural Network architectures with external memory components allow the model to perform inference and capture long term dependencies, by storing information explicitly. In this paper, we generalize Key-Value Memory Networks to a multimodal setting, introducing a novel key-addressing mechanism to deal with sequence-to-sequence models. The advantages of the framework are demonstrated on the task of video captioning, i.e generating natural language descriptions for videos. Conditioning on the previous time-step attention distributions for the key-value memory slots, we introduce a temporal structure in the memory addressing schema. The proposed model naturally decomposes the problem of video captioning into vision and language segments, dealing with them as key-value pairs. More specifically, we learn a semantic embedding (v) corresponding to each frame (k) in the video, thereby creating (k, v) memory slots. This allows us to exploit the temporal dependencies at multiple hierarchies (in the recurrent key-addressing; and in the language decoder). Exploiting this flexibility of the framework, we additionally capture spatial dependencies while mapping from the visual to semantic embedding. Extensive experiments on the Youtube2Text dataset demonstrate usefulness of recurrent key-addressing, while achieving competitive scores on BLEU@4, METEOR metrics against state-of-the-art models. | The notion of has a rich literature in Psychology and Neuroscience, and has recently found application in computer vision @cite_13 and machine translation @cite_6 . Allowing the network to selectively focus on the patches of images or segments of the input sequences, representative works @cite_28 @cite_26 @cite_45 @cite_14 @cite_37 @cite_39 @cite_54 have significantly pushed the state-of-the-art in their domain. The issues of fixed length representation and gradient bottleneck are largely addressed by selectively conditioning the decoder outputs on encoder states: where, @math is the readout, @math is the context from the encoder and @math , is the hidden state of decoder RNN (See @cite_6 for details). | {
"cite_N": [
"@cite_37",
"@cite_26",
"@cite_14",
"@cite_28",
"@cite_54",
"@cite_6",
"@cite_39",
"@cite_45",
"@cite_13"
],
"mid": [
"1957740064",
"",
"",
"2950178297",
"2951837690",
"2133564696",
"2273041409",
"2963758027",
"2951527505"
],
"abstract": [
"We present an approach that exploits hierarchical Recurrent Neural Networks (RNNs) to tackle the video captioning problem, i.e., generating one or multiple sentences to describe a realistic video. Our hierarchical framework contains a sentence generator and a paragraph generator. The sentence generator produces one simple short sentence that describes a specific short video interval. It exploits both temporal- and spatial-attention mechanisms to selectively focus on visual elements during generation. The paragraph generator captures the inter-sentence dependency by taking as input the sentential embedding produced by the sentence generator, combining it with the paragraph history, and outputting the new initial state for the sentence generator. We evaluate our approach on two large-scale benchmark datasets: YouTubeClips and TACoS-MultiLevel. The experiments demonstrate that our approach significantly outperforms the current state-of-the-art methods with BLEU@4 scores 0.499 and 0.305 respectively.",
"",
"",
"Inspired by recent work in machine translation and object detection, we introduce an attention based model that automatically learns to describe the content of images. We describe how we can train this model in a deterministic manner using standard backpropagation techniques and stochastically by maximizing a variational lower bound. We also show through visualization how the model is able to automatically learn to fix its gaze on salient objects while generating the corresponding words in the output sequence. We validate the use of attention with state-of-the-art performance on three benchmark datasets: Flickr8k, Flickr30k and MS COCO.",
"Automatically describing video content with natural language is a fundamental challenge of multimedia. Recurrent Neural Networks (RNN), which models sequence dynamics, has attracted increasing attention on visual interpretation. However, most existing approaches generate a word locally with given previous words and the visual content, while the relationship between sentence semantics and visual content is not holistically exploited. As a result, the generated sentences may be contextually correct but the semantics (e.g., subjects, verbs or objects) are not true. This paper presents a novel unified framework, named Long Short-Term Memory with visual-semantic Embedding (LSTM-E), which can simultaneously explore the learning of LSTM and visual-semantic embedding. The former aims to locally maximize the probability of generating the next word given previous words and visual content, while the latter is to create a visual-semantic embedding space for enforcing the relationship between the semantics of the entire sentence and visual content. Our proposed LSTM-E consists of three components: a 2-D and or 3-D deep convolutional neural networks for learning powerful video representation, a deep RNN for generating sentences, and a joint embedding model for exploring the relationships between visual content and sentence semantics. The experiments on YouTube2Text dataset show that our proposed LSTM-E achieves to-date the best reported performance in generating natural sentences: 45.3 and 31.0 in terms of BLEU@4 and METEOR, respectively. We also demonstrate that LSTM-E is superior in predicting Subject-Verb-Object (SVO) triplets to several state-of-the-art techniques.",
"Neural machine translation is a recently proposed approach to machine translation. Unlike the traditional statistical machine translation, the neural machine translation aims at building a single neural network that can be jointly tuned to maximize the translation performance. The models proposed recently for neural machine translation often belong to a family of encoder-decoders and consists of an encoder that encodes a source sentence into a fixed-length vector from which a decoder generates a translation. In this paper, we conjecture that the use of a fixed-length vector is a bottleneck in improving the performance of this basic encoder-decoder architecture, and propose to extend this by allowing a model to automatically (soft-)search for parts of a source sentence that are relevant to predicting a target word, without having to form these parts as a hard segment explicitly. With this new approach, we achieve a translation performance comparable to the existing state-of-the-art phrase-based system on the task of English-to-French translation. Furthermore, qualitative analysis reveals that the (soft-)alignments found by the model agree well with our intuition.",
"Recently, deep learning approach, especially deep Convolutional Neural Networks (ConvNets), have achieved overwhelming accuracy with fast processing speed for image classification. Incorporating temporal structure with deep ConvNets for video representation becomes a fundamental problem for video content analysis. In this paper, we propose a new approach, namely Hierarchical Recurrent Neural Encoder (HRNE), to exploit temporal information of videos. Compared to recent video representation inference approaches, this paper makes the following three contributions. First, our HRNE is able to efficiently exploit video temporal structure in a longer range by reducing the length of input information flow, and compositing multiple consecutive inputs at a higher level. Second, computation operations are significantly lessened while attaining more non-linearity. Third, HRNE is able to uncover temporal transitions between frame chunks with different granularities, i.e., it can model the temporal transitions between frames as well as the transitions between segments. We apply the new method to video captioning where temporal information plays a crucial role. Experiments demonstrate that our method outperforms the state-of-the-art on video captioning benchmarks. Notably, even using a single network with only RGB stream as input, HRNE beats all the recent systems which combine multiple inputs, such as RGB ConvNet plus 3D ConvNet.",
"We introduce the dense captioning task, which requires a computer vision system to both localize and describe salient regions in images in natural language. The dense captioning task generalizes object detection when the descriptions consist of a single word, and Image Captioning when one predicted region covers the full image. To address the localization and description task jointly we propose a Fully Convolutional Localization Network (FCLN) architecture that processes an image with a single, efficient forward pass, requires no external regions proposals, and can be trained end-to-end with a single round of optimization. The architecture is composed of a Convolutional Network, a novel dense localization layer, and Recurrent Neural Network language model that generates the label sequences. We evaluate our network on the Visual Genome dataset, which comprises 94,000 images and 4,100,000 region-grounded captions. We observe both speed and accuracy improvements over baselines based on current state of the art approaches in both generation and retrieval settings.",
"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."
]
} |
1611.06492 | 2551432385 | Deep Neural Network architectures with external memory components allow the model to perform inference and capture long term dependencies, by storing information explicitly. In this paper, we generalize Key-Value Memory Networks to a multimodal setting, introducing a novel key-addressing mechanism to deal with sequence-to-sequence models. The advantages of the framework are demonstrated on the task of video captioning, i.e generating natural language descriptions for videos. Conditioning on the previous time-step attention distributions for the key-value memory slots, we introduce a temporal structure in the memory addressing schema. The proposed model naturally decomposes the problem of video captioning into vision and language segments, dealing with them as key-value pairs. More specifically, we learn a semantic embedding (v) corresponding to each frame (k) in the video, thereby creating (k, v) memory slots. This allows us to exploit the temporal dependencies at multiple hierarchies (in the recurrent key-addressing; and in the language decoder). Exploiting this flexibility of the framework, we additionally capture spatial dependencies while mapping from the visual to semantic embedding. Extensive experiments on the Youtube2Text dataset demonstrate usefulness of recurrent key-addressing, while achieving competitive scores on BLEU@4, METEOR metrics against state-of-the-art models. | Key-Value MemNNs @cite_55 were originally proposed for QA task in the language domain, providing the last time-step hidden state, as input to the classifier. In this work, we address a more complex problem of video captioning by proposing a novel key-addressing scheme (details in Section ) and setup for exploiting the spatio-temporal structures. The model tracks the attention distribution at previous time steps, thereby providing a strong context on where to attend on the complete video sequence. This implicitly provides a global temporal structure at each readout. While similar in motivation to @cite_56 @cite_34 , the model architecture and domain of application, especially on capturing global temporal dynamics in videos as opposed to images or entailment, is significantly different. | {
"cite_N": [
"@cite_55",
"@cite_34",
"@cite_56"
],
"mid": [
"2409591106",
"",
"2953022248"
],
"abstract": [
"Directly reading documents and being able to answer questions from them is an unsolved challenge. To avoid its inherent difficulty, question answering (QA) has been directed towards using Knowledge Bases (KBs) instead, which has proven effective. Unfortunately KBs often suffer from being too restrictive, as the schema cannot support certain types of answers, and too sparse, e.g. Wikipedia contains much more information than Freebase. In this work we introduce a new method, Key-Value Memory Networks, that makes reading documents more viable by utilizing different encodings in the addressing and output stages of the memory read operation. To compare using KBs, information extraction or Wikipedia documents directly in a single framework we construct an analysis tool, WikiMovies, a QA dataset that contains raw text alongside a preprocessed KB, in the domain of movies. Our method reduces the gap between all three settings. It also achieves state-of-the-art results on the existing WikiQA benchmark.",
"",
"Automatically generating a natural language description of an image has attracted interests recently both because of its importance in practical applications and because it connects two major artificial intelligence fields: computer vision and natural language processing. Existing approaches are either top-down, which start from a gist of an image and convert it into words, or bottom-up, which come up with words describing various aspects of an image and then combine them. In this paper, we propose a new algorithm that combines both approaches through a model of semantic attention. Our algorithm learns to selectively attend to semantic concept proposals and fuse them into hidden states and outputs of recurrent neural networks. The selection and fusion form a feedback connecting the top-down and bottom-up computation. We evaluate our algorithm on two public benchmarks: Microsoft COCO and Flickr30K. Experimental results show that our algorithm significantly outperforms the state-of-the-art approaches consistently across different evaluation metrics."
]
} |
1611.06596 | 2949113568 | While recent deep neural networks have achieved a promising performance on object recognition, they rely implicitly on the visual contents of the whole image. In this paper, we train deep neural net- works on the foreground (object) and background (context) regions of images respectively. Consider- ing human recognition in the same situations, net- works trained on the pure background without ob- jects achieves highly reasonable recognition performance that beats humans by a large margin if only given context. However, humans still outperform networks with pure object available, which indicates networks and human beings have different mechanisms in understanding an image. Furthermore, we straightforwardly combine multiple trained networks to explore different visual cues learned by different networks. Experiments show that useful visual hints can be explicitly learned separately and then combined to achieve higher performance, which verifies the advantages of the proposed framework. | The milestone Convolutional Neural Network (CNN) is treated as a hierarchical model for large-scale visual recognition. In past years, neural networks have already been proved to be effective for simple recognition tasks @cite_2 . More recently, the availability of large-scale training data ( e.g. , ImageNet @cite_27 ) and powerful computation source like GPUs make it practical to train deep neural networks @cite_29 @cite_4 which significantly outperform the conventional models. Even deep features have been proved to be very successful on vision tasks like object discovery @cite_23 , object recognition @cite_6 , etc. A CNN is composed of numerous stacked layers, in which responses from the previous layer are then convoluted and activated by a differentiable function, followed by a non-linear transformation @cite_0 to avoid over-fitting. Recently, several efficient methods were proposed to help CNNs converge faster and prevent over-fitting @cite_29 . It is believed that deeper networks produce better recognition results @cite_28 @cite_10 , but also requires engineering tricks to be trained very well @cite_5 @cite_19 . | {
"cite_N": [
"@cite_4",
"@cite_28",
"@cite_29",
"@cite_6",
"@cite_0",
"@cite_19",
"@cite_27",
"@cite_23",
"@cite_2",
"@cite_5",
"@cite_10"
],
"mid": [
"2001353114",
"2950179405",
"",
"2558914813",
"",
"2949650786",
"2108598243",
"2220111505",
"2154579312",
"2949117887",
"1686810756"
],
"abstract": [
"We study the problem of how to build a deep learning representation for 3D shape. Deep learning has shown to be very effective in variety of visual applications, such as image classification and object detection. However, it has not been successfully applied to 3D shape recognition. This is because 3D shape has complex structure in 3D space and there are limited number of 3D shapes for feature learning. To address these problems, we project 3D shapes into 2D space and use autoencoder for feature learning on the 2D images. High accuracy 3D shape retrieval performance is obtained by aggregating the features learned on 2D images. In addition, we show the proposed deep learning feature is complementary to conventional local image descriptors. By combing the global deep learning representation and the local descriptor representation, our method can obtain the state-of-the-art performance on 3D shape retrieval benchmarks.",
"We propose a deep convolutional neural network architecture codenamed \"Inception\", which was responsible for setting the new state of the art for classification and detection in the ImageNet Large-Scale Visual Recognition Challenge 2014 (ILSVRC 2014). The main hallmark of this architecture is the improved utilization of the computing resources inside the network. This was achieved by a carefully crafted design that allows for increasing the depth and width of the network while keeping the computational budget constant. To optimize quality, the architectural decisions were based on the Hebbian principle and the intuition of multi-scale processing. One particular incarnation used in our submission for ILSVRC 2014 is called GoogLeNet, a 22 layers deep network, the quality of which is assessed in the context of classification and detection.",
"",
"State-of-the-art image classification approaches are mainly based on robust image representation, such as the bag-of-features (BoF) model or the convolutional neural network (CNN) architecture. In real applications, the orientation (left right) of an image or an object might vary from sample to sample, whereas some handcrafted descriptors (e.g., SIFT) and network operations (e.g., convolution) are not reversal-invariant, leading to the unsatisfied stability of image features extracted from these models. To deal with, a popular solution is to augment the dataset by adding a left-right reversed copy for each image. This strategy improves the recognition accuracy to some extent, but also brings the price of almost doubled time and memory consumptions on both the training and testing stages. In this paper, we present an alternative solution based on designing reversal-invariant representation of local patterns, so that we can obtain the identical representation for an image and its left-right reversed copy. For the BoF model, we design a reversal-invariant version of SIFT descriptor named Max-SIFT, a generalized RIDE algorithm which can be applied to a large family of local descriptors. For the CNN architecture, we present a simple idea of generating reversal-invariant deep features (RI-Deep), and, inspired by which, design reversal-invariant convolution (RI-Conv) layers to increase the CNN capacity without increasing the model complexity. Experiments reveal consistent accuracy gain on various image classification tasks, including scene understanding, fine-grained object recognition, and large-scale visual recognition.",
"",
"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.",
"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.",
"Multiple-instance learning (MIL) has served as an important tool for a wide range of vision applications, for instance, image classification, object detection, and visual tracking. In this paper, we propose a novel method to solve the classical MIL problem, named relaxed multiple-instance SVM (RMI-SVM). We treat the positiveness of instance as a continuous variable, use Noisy-OR model to enforce the MIL constraints, and optimize them jointly in a unified framework. The optimization problem can be efficiently solved using stochastic gradient decent. The extensive experiments demonstrate that RMI-SVM consistently achieves superior performance on various benchmarks for MIL. Moreover, we simply applied RMI-SVM to a challenging vision task, common object discovery. The state-of-the arts results of object discovery on PASCAL VOC datasets further confirm the advantages of the proposed method.",
"We present an application of back-propagation networks to handwritten digit recognition. Minimal preprocessing of the data was required, but architecture of the network was highly constrained and specifically designed for the task. The input of the network consists of normalized images of isolated digits. The method has 1 error rate and about a 9 reject rate on zipcode digits provided by the U.S. Postal Service.",
"Training Deep Neural Networks is complicated by the fact that the distribution of each layer's inputs changes during training, as the parameters of the previous layers change. This slows down the training by requiring lower learning rates and careful parameter initialization, and makes it notoriously hard to train models with saturating nonlinearities. We refer to this phenomenon as internal covariate shift, and address the problem by normalizing layer inputs. Our method draws its strength from making normalization a part of the model architecture and performing the normalization for each training mini-batch. Batch Normalization allows us to use much higher learning rates and be less careful about initialization. It also acts as a regularizer, in some cases eliminating the need for Dropout. Applied to a state-of-the-art image classification model, Batch Normalization achieves the same accuracy with 14 times fewer training steps, and beats the original model by a significant margin. Using an ensemble of batch-normalized networks, we improve upon the best published result on ImageNet classification: reaching 4.9 top-5 validation error (and 4.8 test error), exceeding the accuracy of human raters.",
"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."
]
} |
1611.06596 | 2949113568 | While recent deep neural networks have achieved a promising performance on object recognition, they rely implicitly on the visual contents of the whole image. In this paper, we train deep neural net- works on the foreground (object) and background (context) regions of images respectively. Consider- ing human recognition in the same situations, net- works trained on the pure background without ob- jects achieves highly reasonable recognition performance that beats humans by a large margin if only given context. However, humans still outperform networks with pure object available, which indicates networks and human beings have different mechanisms in understanding an image. Furthermore, we straightforwardly combine multiple trained networks to explore different visual cues learned by different networks. Experiments show that useful visual hints can be explicitly learned separately and then combined to achieve higher performance, which verifies the advantages of the proposed framework. | Very few techniques on background modeling @cite_3 have been developed for object recognition, despite the huge success of deep learning methods on various vision tasks. @cite_11 proposed the fully convolutional networks (FCN) for semantic segmentation, which are further trained on foreground and background defined by shape masks. They find it is not vital to learn a specifically designed background model. For face matching, @cite_15 developed methods only on the cropped out faces to alleviate the possible correlations between faces and their backgrounds. @cite_35 modeled the background in order to detect the salient objects from the background. @cite_13 showed using the object patch to predict its context as supervisory information can help discover object clusters, which is consistent with our motivation to utilize the pure context for visual recognition. To our best knowledge, we are the first to explicitly learn both the foreground and background models and then combine them together to be beneficial for the object recognition. | {
"cite_N": [
"@cite_35",
"@cite_3",
"@cite_15",
"@cite_13",
"@cite_11"
],
"mid": [
"2080142539",
"2507193517",
"1482998036",
"1032684693",
""
],
"abstract": [
"Detection of salient objects from images is gaining increasing research interest in recent years as it can substantially facilitate a wide range of content-based multimedia applications. Based on the assumption that foreground salient regions are distinctive within a certain context, most conventional approaches rely on a number of hand-designed features and their distinctiveness is measured using local or global contrast. Although these approaches have been shown to be effective in dealing with simple images, their limited capability may cause difficulties when dealing with more complicated images. This paper proposes a novel framework for saliency detection by first modeling the background and then separating salient objects from the background. We develop stacked denoising autoencoders with deep learning architectures to model the background where latent patterns are explored and more powerful representations of data are learned in an unsupervised and bottom-up manner. Afterward, we formulate the separation of salient objects from the background as a problem of measuring reconstruction residuals of deep autoencoders. Comprehensive evaluations of three benchmark datasets and comparisons with nine state-of-the-art algorithms demonstrate the superiority of this paper.",
"This paper addresses the problem of detecting people and vehicles on a surface mine by presenting an architecture that combines the complementary strengths of deep convolutional networks DCN with cluster-based analysis. We highlight that using a DCN in a naive black box approach results in a significantly high rate of errors due to the lack of mining-specific training data and the unique landscape in a mine site. In this work, we propose a background model that exploits the abundance of background-only images to discover the natural clusters in visual appearance using features extracted from the DCN. Both a simple nearest cluster-based background model and an extended model with cosine features are investigated for their ability to identify and suppress potential false positives made by the DCN. Furthermore, localization of objects of interest is enabled through region proposals, which have been tuned to increase recall within the constraints of a computational budget. Finally, a soft fusion framework is presented to combine the estimates of both the DCN and background model to improve the accuracy of the detection. Our system is tested on over 11 km of real mine site data in both day and night conditions where we were able to detect both light and heavy vehicles along with mining personnel. We show that the introduction of our background model improves the detection performance. In particular, soft fusion of the background model and the DCN output produces a relative improvement in the F1 score of 46 and 28 compared to a baseline pretrained DCN and a DCN retrained with mining images, respectively.",
"We propose a scalable face matching algorithm capable of dealing with faces subject to several concurrent and uncontrolled factors, such as variations in pose, expression, illumination, resolution, as well as scale and misalignment problems. Each face is described in terms of multi-region probabilistic histograms of visual words, followed by a normalised distance calculation between the histograms of two faces. We also propose a fast histogram approximation method which dramatically reduces the computational burden with minimal impact on discrimination performance. Experiments on the \"Labeled Faces in the Wild\" dataset (unconstrained environments) as well as FERET (controlled variations) show that the proposed algorithm obtains performance on par with a more complex method and displays a clear advantage over predecessor systems. Furthermore, the use of multiple regions (as opposed to a single overall region) improves accuracy in most cases, especially when dealing with illumination changes and very low resolution images. The experiments also show that normalised distances can noticeably improve robustness by partially counteracting the effects of image variations.",
"This paper addresses the well-established problem of unsupervised object discovery with a novel method inspired by weakly-supervised approaches. In particular, the ability of an object patch to predict the rest of the object (its context) is used as supervisory signal to help discover visually consistent object clusters. The main contributions of this work are: 1) framing unsupervised clustering as a leave-one-out context prediction task; 2) evaluating the quality of context prediction by statistical hypothesis testing between thing and stuff appearance models; and 3) an iterative region prediction and context alignment approach that gradually discovers a visual object cluster together with a segmentation mask and fine-grained correspondences. The proposed method outperforms previous unsupervised as well as weakly-supervised object discovery approaches, and is shown to provide correspondences detailed enough to transfer keypoint annotations.",
""
]
} |
1611.06596 | 2949113568 | While recent deep neural networks have achieved a promising performance on object recognition, they rely implicitly on the visual contents of the whole image. In this paper, we train deep neural net- works on the foreground (object) and background (context) regions of images respectively. Consider- ing human recognition in the same situations, net- works trained on the pure background without ob- jects achieves highly reasonable recognition performance that beats humans by a large margin if only given context. However, humans still outperform networks with pure object available, which indicates networks and human beings have different mechanisms in understanding an image. Furthermore, we straightforwardly combine multiple trained networks to explore different visual cues learned by different networks. Experiments show that useful visual hints can be explicitly learned separately and then combined to achieve higher performance, which verifies the advantages of the proposed framework. | Last but not the least, visualization of CNN activations is an effective method to understand the mechanism of CNNs. @cite_30 , a de-convolutional operation was proposed to capture visual patterns on different layers of a trained network. @cite_10 and @cite_25 show that different sets of neurons are activated when a network is used for detecting different visual patterns. In this work, we will use a much simpler way of visualization which is inspired by @cite_24 . | {
"cite_N": [
"@cite_30",
"@cite_10",
"@cite_25",
"@cite_24"
],
"mid": [
"2952186574",
"1686810756",
"2221625691",
""
],
"abstract": [
"Large Convolutional Network models have recently demonstrated impressive classification performance on the ImageNet benchmark. However there is no clear understanding of why they perform so well, or how they might be improved. In this paper we address both issues. We introduce a novel visualization technique that gives insight into the function of intermediate feature layers and the operation of the classifier. We also perform an ablation study to discover the performance contribution from different model layers. This enables us to find model architectures that outperform Krizhevsky al on the ImageNet classification benchmark. 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.",
"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.",
"While feedforward deep convolutional neural networks (CNNs) have been a great success in computer vision, it is important to note that the human visual cortex generally contains more feedback than feedforward connections. In this paper, we will briefly introduce the background of feedbacks in the human visual cortex, which motivates us to develop a computational feedback mechanism in deep neural networks. In addition to the feedforward inference in traditional neural networks, a feedback loop is introduced to infer the activation status of hidden layer neurons according to the \"goal\" of the network, e.g., high-level semantic labels. We analogize this mechanism as \"Look and Think Twice.\" The feedback networks help better visualize and understand how deep neural networks work, and capture visual attention on expected objects, even in images with cluttered background and multiple objects. Experiments on ImageNet dataset demonstrate its effectiveness in solving tasks such as image classification and object localization.",
""
]
} |
1611.06194 | 2950510923 | In this paper we introduce a model of lifelong learning, based on a Network of Experts. New tasks experts are learned and added to the model sequentially, building on what was learned before. To ensure scalability of this process, data from previous tasks cannot be stored and hence is not available when learning a new task. A critical issue in such context, not addressed in the literature so far, relates to the decision of which expert to deploy at test time. We introduce a gating autoencoder that learns a representation for the task at hand, and is used at test time to automatically forward the test sample to the relevant expert. This has the added advantage of being memory efficient as only one expert network has to be loaded into memory at any given time. Further, the autoencoders inherently capture the relatedness of one task to another, based on which the most relevant prior model to be used for training a new expert with fine-tuning or learning-without-forgetting can be selected. We evaluate our system on image classification and video prediction problems. | To determine which related tasks to utilize, @cite_37 cluster the tasks based on the mutual information gain when using the information from one task while learning another. This is an exhaustive process. As an alternative, @cite_6 @cite_16 @cite_2 assume that the parameters of related task models lie close by in the original space or in a lower dimensional subspace and thus cluster the tasks' parameters. They first learn task models independently, then use the tasks within the same cluster to help improving or relearning their models. This requires learning individual task models first. Alternatively, we use our tasks autoencoders, that are fast to train, to identify related tasks. | {
"cite_N": [
"@cite_37",
"@cite_16",
"@cite_6",
"@cite_2"
],
"mid": [
"1566538838",
"",
"2949664970",
"2949201716"
],
"abstract": [
"Recently, there has been an increased interest in machine learning methods that transfer knowledge across multiple learning tasks and “learn to learn.” Such methods have repeatedly been found to outperform conventional, single-task learning algorithms when the learning tasks are appropriately related. To increase robustness of such approaches, methods are desirable that can reason about the relatedness of individual learning tasks, in order to avoid the danger arising from tasks that are unrelated and thus potentially misleading.",
"",
"In multi-task learning several related tasks are considered simultaneously, with the hope that by an appropriate sharing of information across tasks, each task may benefit from the others. In the context of learning linear functions for supervised classification or regression, this can be achieved by including a priori information about the weight vectors associated with the tasks, and how they are expected to be related to each other. In this paper, we assume that tasks are clustered into groups, which are unknown beforehand, and that tasks within a group have similar weight vectors. We design a new spectral norm that encodes this a priori assumption, without the prior knowledge of the partition of tasks into groups, resulting in a new convex optimization formulation for multi-task learning. We show in simulations on synthetic examples and on the IEDB MHC-I binding dataset, that our approach outperforms well-known convex methods for multi-task learning, as well as related non convex methods dedicated to the same problem.",
"In the paradigm of multi-task learning, mul- tiple related prediction tasks are learned jointly, sharing information across the tasks. We propose a framework for multi-task learn- ing that enables one to selectively share the information across the tasks. We assume that each task parameter vector is a linear combi- nation of a finite number of underlying basis tasks. The coefficients of the linear combina- tion are sparse in nature and the overlap in the sparsity patterns of two tasks controls the amount of sharing across these. Our model is based on on the assumption that task pa- rameters within a group lie in a low dimen- sional subspace but allows the tasks in differ- ent groups to overlap with each other in one or more bases. Experimental results on four datasets show that our approach outperforms competing methods."
]
} |
1611.06194 | 2950510923 | In this paper we introduce a model of lifelong learning, based on a Network of Experts. New tasks experts are learned and added to the model sequentially, building on what was learned before. To ensure scalability of this process, data from previous tasks cannot be stored and hence is not available when learning a new task. A critical issue in such context, not addressed in the literature so far, relates to the decision of which expert to deploy at test time. We introduce a gating autoencoder that learns a representation for the task at hand, and is used at test time to automatically forward the test sample to the relevant expert. This has the added advantage of being memory efficient as only one expert network has to be loaded into memory at any given time. Further, the autoencoders inherently capture the relatedness of one task to another, based on which the most relevant prior model to be used for training a new expert with fine-tuning or learning-without-forgetting can be selected. We evaluate our system on image classification and video prediction problems. | One of the first examples of using multiple models, each one handling a subset of tasks, was by @cite_36 . They trained an adaptive mixture of experts (each a neural network) for multi-speaker vowel recognition and used a separate gating network to determine which network to use for each sample. They showed that this setup outperformed a single shared model. A downside, however, was that each training sample needed to pass through each expert, for the gating function to be learned. To avoid this issue, a mixture of one generalist model and many specialist models has been proposed @cite_17 @cite_1 . At test time, the generalist model acts as a gate, forwarding the sample to the correct network. However, unlike our model, these approaches require all the data to be available for learning the generalist model, which needs to be retrained each time a new task arrives. | {
"cite_N": [
"@cite_36",
"@cite_1",
"@cite_17"
],
"mid": [
"2150884987",
"1821462560",
"2953280703"
],
"abstract": [
"We present a new supervised learning procedure for systems composed of many separate networks, each of which learns to handle a subset of the complete set of training cases. The new procedure can be viewed either as a modular version of a multilayer supervised network, or as an associative version of competitive learning. It therefore provides a new link between these two apparently different approaches. We demonstrate that the learning procedure divides up a vowel discrimination task into appropriate subtasks, each of which can be solved by a very simple expert network.",
"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.",
"We present a tree-structured network architecture for large scale image classification. The trunk of the network contains convolutional layers optimized over all classes. At a given depth, the trunk splits into separate branches, each dedicated to discriminate a different subset of classes. Each branch acts as an expert classifying a set of categories that are difficult to tell apart, while the trunk provides common knowledge to all experts in the form of shared features. The training of our \"network of experts\" is completely end-to-end: the partition of categories into disjoint subsets is learned simultaneously with the parameters of the network trunk and the experts are trained jointly by minimizing a single learning objective over all classes. The proposed structure can be built from any existing convolutional neural network (CNN). We demonstrate its generality by adapting 4 popular CNNs for image categorization into the form of networks of experts. Our experiments on CIFAR100 and ImageNet show that in every case our method yields a substantial improvement in accuracy over the base CNN, and gives the best result achieved so far on CIFAR100. Finally, the improvement in accuracy comes at little additional cost: compared to the base network, the training time is only moderately increased and the number of parameters is comparable or in some cases even lower."
]
} |
1611.06345 | 2952209150 | The latest deep learning approaches perform better than the state-of-the-art signal processing approaches in various image restoration tasks. However, if an image contains many patterns and structures, the performance of these CNNs is still inferior. To address this issue, here we propose a novel feature space deep residual learning algorithm that outperforms the existing residual learning. The main idea is originated from the observation that the performance of a learning algorithm can be improved if the input and or label manifolds can be made topologically simpler by an analytic mapping to a feature space. Our extensive numerical studies using denoising experiments and NTIRE single-image super-resolution (SISR) competition demonstrate that the proposed feature space residual learning outperforms the existing state-of-the-art approaches. Moreover, our algorithm was ranked third in NTIRE competition with 5-10 times faster computational time compared to the top ranked teams. The source code is available on page : this https URL | One of the classical approaches for image denoising is a wavelet shrinkage approach @cite_7 , which decomposes an image into low and high frequency subbands and applies thresholding in the high frequency coefficients @cite_24 . Advanced algorithms in this field are to exploit the intra- and inter- correlations of the wavelet coefficients @cite_12 . | {
"cite_N": [
"@cite_24",
"@cite_12",
"@cite_7"
],
"mid": [
"2113945798",
"2134929491",
""
],
"abstract": [
"We describe a method for removing noise from digital images, based on a statistical model of the coefficients of an overcomplete multiscale oriented basis. Neighborhoods of coefficients at adjacent positions and scales are modeled as the product of two independent random variables: a Gaussian vector and a hidden positive scalar multiplier. The latter modulates the local variance of the coefficients in the neighborhood, and is thus able to account for the empirically observed correlation between the coefficient amplitudes. Under this model, the Bayesian least squares estimate of each coefficient reduces to a weighted average of the local linear estimates over all possible values of the hidden multiplier variable. We demonstrate through simulations with images contaminated by additive white Gaussian noise that the performance of this method substantially surpasses that of previously published methods, both visually and in terms of mean squared error.",
"Wavelet-based statistical signal processing techniques such as denoising and detection typically model the wavelet coefficients as independent or jointly Gaussian. These models are unrealistic for many real-world signals. We develop a new framework for statistical signal processing based on wavelet-domain hidden Markov models (HMMs) that concisely models the statistical dependencies and non-Gaussian statistics encountered in real-world signals. Wavelet-domain HMMs are designed with the intrinsic properties of the wavelet transform in mind and provide powerful, yet tractable, probabilistic signal models. Efficient expectation maximization algorithms are developed for fitting the HMMs to observational signal data. The new framework is suitable for a wide range of applications, including signal estimation, detection, classification, prediction, and even synthesis. To demonstrate the utility of wavelet-domain HMMs, we develop novel algorithms for signal denoising, classification, and detection.",
""
]
} |
1611.06345 | 2952209150 | The latest deep learning approaches perform better than the state-of-the-art signal processing approaches in various image restoration tasks. However, if an image contains many patterns and structures, the performance of these CNNs is still inferior. To address this issue, here we propose a novel feature space deep residual learning algorithm that outperforms the existing residual learning. The main idea is originated from the observation that the performance of a learning algorithm can be improved if the input and or label manifolds can be made topologically simpler by an analytic mapping to a feature space. Our extensive numerical studies using denoising experiments and NTIRE single-image super-resolution (SISR) competition demonstrate that the proposed feature space residual learning outperforms the existing state-of-the-art approaches. Moreover, our algorithm was ranked third in NTIRE competition with 5-10 times faster computational time compared to the top ranked teams. The source code is available on page : this https URL | In neural network literature, the work by @cite_11 was the first which demonstrated similar denoising performance to BM3D using multi-layer perceptron (MLP). Chen al @cite_6 @cite_26 proposed a deep learning approach called trainable nonlinear reaction diffusion (TNRD) that can train filters and influence functions by unfolding a variational optimization approach. Recently, based on skipped connection and encoder-decoder architecture, a very deep residual encoder-decoder networks (RED-Net) was proposed for image restoration problems @cite_23 . | {
"cite_N": [
"@cite_26",
"@cite_23",
"@cite_6",
"@cite_11"
],
"mid": [
"2953319141",
"2520164769",
"1906770428",
"2037642501"
],
"abstract": [
"For several decades, image restoration remains an active research topic in low-level computer vision and hence new approaches are constantly emerging. However, many recently proposed algorithms achieve state-of-the-art performance only at the expense of very high computation time, which clearly limits their practical relevance. In this work, we propose a simple but effective approach with both high computational efficiency and high restoration quality. We extend conventional nonlinear reaction diffusion models by several parametrized linear filters as well as several parametrized influence functions. We propose to train the parameters of the filters and the influence functions through a loss based approach. Experiments show that our trained nonlinear reaction diffusion models largely benefit from the training of the parameters and finally lead to the best reported performance on common test datasets for image restoration. Due to their structural simplicity, our trained models are highly efficient and are also well-suited for parallel computation on GPUs.",
"In this paper, we propose a very deep fully convolutional encoding-decoding framework for image restoration such as denoising and super-resolution. The network is composed of multiple layers of convolution and de-convolution operators, learning end-to-end mappings from corrupted images to the original ones. The convolutional layers act as the feature extractor, which capture the abstraction of image contents while eliminating noises corruptions. De-convolutional layers are then used to recover the image details. We propose to symmetrically link convolutional and de-convolutional layers with skip-layer connections, with which the training converges much faster and attains a higher-quality local optimum. First, The skip connections allow the signal to be back-propagated to bottom layers directly, and thus tackles the problem of gradient vanishing, making training deep networks easier and achieving restoration performance gains consequently. Second, these skip connections pass image details from convolutional layers to de-convolutional layers, which is beneficial in recovering the original image. Significantly, with the large capacity, we can handle different levels of noises using a single model. Experimental results show that our network achieves better performance than all previously reported state-of-the-art methods.",
"Image restoration is a long-standing problem in low-level computer vision with many interesting applications. We describe a flexible learning framework based on the concept of nonlinear reaction diffusion models for various image restoration problems. By embodying recent improvements in nonlinear diffusion models, we propose a dynamic nonlinear reaction diffusion model with time-dependent parameters ( i.e. , linear filters and influence functions). In contrast to previous nonlinear diffusion models, all the parameters, including the filters and the influence functions, are simultaneously learned from training data through a loss based approach. We call this approach TNRD— Trainable Nonlinear Reaction Diffusion . The TNRD approach is applicable for a variety of image restoration tasks by incorporating appropriate reaction force. We demonstrate its capabilities with three representative applications, Gaussian image denoising, single image super resolution and JPEG deblocking. Experiments show that our trained nonlinear diffusion models largely benefit from the training of the parameters and finally lead to the best reported performance on common test datasets for the tested applications. Our trained models preserve the structural simplicity of diffusion models and take only a small number of diffusion steps, thus are highly efficient. Moreover, they are also well-suited for parallel computation on GPUs, which makes the inference procedure extremely fast.",
"Image denoising can be described as the problem of mapping from a noisy image to a noise-free image. The best currently available denoising methods approximate this mapping with cleverly engineered algorithms. In this work we attempt to learn this mapping directly with a plain multi layer perceptron (MLP) applied to image patches. While this has been done before, we will show that by training on large image databases we are able to compete with the current state-of-the-art image denoising methods. Furthermore, our approach is easily adapted to less extensively studied types of noise (by merely exchanging the training data), for which we achieve excellent results as well."
]
} |
1611.06345 | 2952209150 | The latest deep learning approaches perform better than the state-of-the-art signal processing approaches in various image restoration tasks. However, if an image contains many patterns and structures, the performance of these CNNs is still inferior. To address this issue, here we propose a novel feature space deep residual learning algorithm that outperforms the existing residual learning. The main idea is originated from the observation that the performance of a learning algorithm can be improved if the input and or label manifolds can be made topologically simpler by an analytic mapping to a feature space. Our extensive numerical studies using denoising experiments and NTIRE single-image super-resolution (SISR) competition demonstrate that the proposed feature space residual learning outperforms the existing state-of-the-art approaches. Moreover, our algorithm was ranked third in NTIRE competition with 5-10 times faster computational time compared to the top ranked teams. The source code is available on page : this https URL | Residual learning has multiple realizations. The first approach is using a skipped connection that bypasses input data of a certain layer to another layer during forward and backward propagations. This type of residual learning concept was first introduced by He al @cite_22 for image recognition. In low-level computer vision problems, Kim al @cite_10 employed a residual learning for a super-resolution method. In these approaches, the residual learning was implemented by a skipped connection corresponding to an identity mapping. In another implementation, the label data is transformed into the difference between the input data and clean data. For example, Zhang al @cite_3 proposed a denoising convolutional neural networks (DnCNNs) @cite_3 , which has inspired our method. | {
"cite_N": [
"@cite_10",
"@cite_22",
"@cite_3"
],
"mid": [
"2951997238",
"2949650786",
"2508457857"
],
"abstract": [
"We present a highly accurate single-image super-resolution (SR) method. Our method uses a very deep convolutional network inspired by VGG-net used for ImageNet classification simonyan2015very . We find increasing our network depth shows a significant improvement in accuracy. Our final model uses 20 weight layers. By cascading small filters many times in a deep network structure, contextual information over large image regions is exploited in an efficient way. With very deep networks, however, convergence speed becomes a critical issue during training. We propose a simple yet effective training procedure. We learn residuals only and use extremely high learning rates ( @math times higher than SRCNN dong2015image ) enabled by adjustable gradient clipping. Our proposed method performs better than existing methods in accuracy and visual improvements in our results are easily noticeable.",
"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.",
"The discriminative model learning for image denoising has been recently attracting considerable attentions due to its favorable denoising performance. In this paper, we take one step forward by investigating the construction of feed-forward denoising convolutional neural networks (DnCNNs) to embrace the progress in very deep architecture, learning algorithm, and regularization method into image denoising. Specifically, residual learning and batch normalization are utilized to speed up the training process as well as boost the denoising performance. Different from the existing discriminative denoising models which usually train a specific model for additive white Gaussian noise at a certain noise level, our DnCNN model is able to handle Gaussian denoising with unknown noise level (i.e., blind Gaussian denoising). With the residual learning strategy, DnCNN implicitly removes the latent clean image in the hidden layers. This property motivates us to train a single DnCNN model to tackle with several general image denoising tasks, such as Gaussian denoising, single image super-resolution, and JPEG image deblocking. Our extensive experiments demonstrate that our DnCNN model can not only exhibit high effectiveness in several general image denoising tasks, but also be efficiently implemented by benefiting from GPU computing."
]
} |
1611.06173 | 2550947904 | A dynamical model consists of a continuous self-map @math of a compact state space @math and a continuous observation function @math . This paper considers the fitting of a parametrized family of dynamical models to an observed real-valued stochastic process using empirical risk minimization. The limiting behavior of the minimum risk parameters is studied in a general setting. We establish a general convergence theorem for minimum risk estimators and ergodic observations. We then study conditions under which empirical risk minimization can effectively separate the signal from the noise in an additive observational noise model. The key, necessary condition in the latter results is that the family of dynamical models has limited complexity, which is quantified through a notion of entropy for families of infinite sequences. Close connections between entropy and limiting average mean widths for stationary processes are established. | Modha and Masry @cite_5 , Meir @cite_11 , and Alquier and Wintenberger @cite_15 establish oracle inequalities and finite sample bounds for predicting the next value of a stationary process. Agarwal and Duchi @cite_38 , Kuznetsov and Mohri @cite_7 @cite_41 @cite_6 , and Zimin and Lampert @cite_17 establish finite sample performance bounds on the conditional risk of online learning algorithms for predicting dependent time series. Each of the papers cited above imposes mixing conditions on the observations as well as regularity conditions on the loss function and model family of interest. Shalizi and Kontorovich @cite_30 consider learning mixtures of stationary processes, while Kontorovich @cite_40 studies statistical estimation using finite automata with bounded memory. Hang and Steinwart @cite_4 obtain rates of convergence for empirical risk minimization from @math -mixing observations, while Wong @cite_2 establish finite sample bounds for Lasso-based inference under @math -mixing conditions. In another direction, Rakhlin @cite_3 and Rakhlin and Sridharan @cite_23 have established exponential inequalities for suprema of martingale difference sequences by using and extending ideas from machine learning, including Rademacher complexity and deterministic regret inequalities. | {
"cite_N": [
"@cite_30",
"@cite_38",
"@cite_4",
"@cite_7",
"@cite_41",
"@cite_17",
"@cite_6",
"@cite_3",
"@cite_40",
"@cite_23",
"@cite_2",
"@cite_5",
"@cite_15",
"@cite_11"
],
"mid": [
"",
"2127684734",
"2624324495",
"",
"",
"2641114818",
"",
"",
"",
"2963868636",
"2404827031",
"2137633226",
"2083047081",
"2078894510"
],
"abstract": [
"",
"We study the generalization performance of online learning algorithms trained on samples coming from a dependent source of data. We show that the generalization error of any stable online algorithm concentrates around its regret-an easily computable statistic of the online performance of the algorithm-when the underlying ergodic process is β- or φ -mixing. We show high-probability error bounds assuming the loss function is convex, and we also establish sharp convergence rates and deviation bounds for strongly convex losses and several linear prediction problems such as linear and logistic regression, least-squares SVM, and boosting on dependent data. In addition, our results have straightforward applications to stochastic optimization with dependent data, and our analysis requires only martingale convergence arguments; we need not rely on more powerful statistical tools such as empirical process theory.",
"We introduce and study a variational framework for the analysis of empirical risk based inference for dynamical systems and ergodic processes. The analysis applies to a two-stage estimation procedure in which (i) the trajectory of an observed (but unknown) system is fit to a trajectory from a known reference system by minimizing cumulative per-state loss, and (ii) a parameter estimate is obtained from the initial state of the best fit reference trajectory. We show that the empirical risk of the best fit trajectory converges almost surely to a constant that can be expressed in a variational form as the minimal expected loss over dynamically invariant couplings (joinings) of the observed and reference systems. Moreover, we establish that the family of joinings minimizing the expected loss is convex and compact, and that it fully characterizes the asymptotic behavior of the estimated parameters, addressing both identifiability and misspecification. The two-stage estimation framework and associated variational analysis apply to a broad family of empirical risk miminization procedures for dependent observations. To illustrate this, we apply variational analysis to the well studied problems of maximum likelihood and non-linear regression, and then undertake an extended analysis of system identification from quantized trajectories subject to noise, a problem of interest in dynamics, where the models themselves exhibit dynamical behavior across time.",
"",
"",
"",
"",
"",
"",
"We study an equivalence of (i) deterministic pathwise statements appearing in the online learning literature (termed regret bounds), (ii) high-probability tail bounds for the supremum of a collection of martingales (of a specific form arising from uniform laws of large numbers for martingales), and (iii) in-expectation bounds for the supremum. By virtue of the equivalence, we prove exponential tail bounds for norms of Banach space valued martingales via deterministic regret bounds for the online mirror descent algorithm with an adaptive step size. We extend these results beyond the linear structure of the Banach space: we define a notion of martingale type for general classes of real-valued functions and show its equivalence (up to a logarithmic factor) to various sequential complexities of the class (in particular, the sequential Rademacher complexity and its offset version). For classes with the general martingale type 2, we exhibit a finer notion of variation that allows partial adaptation to the function indexing the martingale. Our proof technique rests on sequential symmetrization and on certifying the existence of regret minimization strategies for certain online prediction problems.",
"The Lasso is one of the most popular methods in high dimensional statistical learning. Most existing theoretical results for the Lasso, however, require the samples to be iid. Recent work has provided guarantees for the Lasso assuming that the time series is generated by a sparse Vector Auto-Regressive (VAR) model with Gaussian innovations. Proofs of these results rely critically on the fact that the true data generating mechanism (DGM) is a finite-order Gaussian VAR. This assumption is quite brittle: linear transformations, including selecting a subset of variables, can lead to the violation of this assumption. In order to break free from such assumptions, we derive nonasymptotic inequalities for estimation error and prediction error of the Lasso estimate of the best linear predictor without assuming any special parametric form of the DGM. Instead, we rely only on (strict) stationarity and mixing conditions to establish consistency of the Lasso in the following two scenarios: (a) alpha-mixing Gaussian processes, and (b) beta-mixing sub-Gaussian random vectors. Our work provides an alternative proof of the consistency of the Lasso for sparse Gaussian VAR models. But the applicability of our results extends to non-Gaussian and non-linear times series models as the examples we provide demonstrate. In order to prove our results, we derive a novel Hanson-Wright type concentration inequality for beta-mixing sub-Gaussian random vectors that may be of independent interest.",
"We consider the problem of one-step-ahead prediction of a real-valued, stationary, strongly mixing random process (Xi) sub i=- spl infin sup spl infin . The best mean-square predictor of X sub 0 is its conditional mean given the entire infinite past (X sub i ) sub i=- spl infin sup -1 . Given a sequence of observations X sub 1 , X sub 2 , X sub N , we propose estimators for the conditional mean based on sequences of parametric models of increasing memory and of increasing dimension, for example, neural networks and Legendre polynomials. The proposed estimators select both the model memory and the model dimension, in a data-driven fashion, by minimizing certain complexity regularized least squares criteria. When the underlying predictor function has a finite memory, we establish that the proposed estimators are memory-universal: the proposed estimators, which do not know the true memory, deliver the same statistical performance (rates of integrated mean-squared error) as that delivered by estimators that know the true memory. Furthermore, when the underlying predictor function does not have a finite memory, we establish that the estimator based on Legendre polynomials is consistent.",
"Observing a stationary time series, we propose a two-step procedure for the predictionof the next value of the time series. The first step follows machine learning theory paradigmand consists in determining a set of possible predictors as randomized estimators in (possiblynumerous) different predictive models. The second step follows the model selection paradigmand consists in choosing one predictor with good properties among all the predictors of the firststeps. We study our procedure for two different types of observations: causal Bernoulli shifts andbounded weakly dependent processes. In both cases, we give oracle inequalities: the risk of thechosen predictor is close to the best prediction risk in all predictive models that we consider. Weapply our procedure for predictive models such as linear predictors, neural networks predictorsand non-parametric autoregressive predictors.",
"We consider the problem of one-step ahead prediction for time series generated by an underlying stationary stochastic process obeying the condition of absolute regularity, describing the mixing nature of process. We make use of recent results from the theory of empirical processes, and adapt the uniform convergence framework of Vapnik and Chervonenkis to the problem of time series prediction, obtaining finite sample bounds. Furthermore, by allowing both the model complexity and memory size to be adaptively determined by the data, we derive nonparametric rates of convergence through an extension of the method of structural risk minimization suggested by Vapnik. All our results are derived for general L error measures, and apply to both exponentially and algebraically mixing processes."
]
} |
1611.06173 | 2550947904 | A dynamical model consists of a continuous self-map @math of a compact state space @math and a continuous observation function @math . This paper considers the fitting of a parametrized family of dynamical models to an observed real-valued stochastic process using empirical risk minimization. The limiting behavior of the minimum risk parameters is studied in a general setting. We establish a general convergence theorem for minimum risk estimators and ergodic observations. We then study conditions under which empirical risk minimization can effectively separate the signal from the noise in an additive observational noise model. The key, necessary condition in the latter results is that the family of dynamical models has limited complexity, which is quantified through a notion of entropy for families of infinite sequences. Close connections between entropy and limiting average mean widths for stationary processes are established. | Furstenberg's original work on joinings @cite_52 includes an application of joinings to a nonlinear filtering problem. Beyond this application, we are not aware of other uses of joinings in the literature on statistical inference. Ornstein and Weiss @cite_57 studied the estimation of a stochastic process from its samples. They proposed an inference procedure, based on matching @math -block frequencies, and characterize when it produces consistent estimates of the observed stochastic process in the @math -bar metric. | {
"cite_N": [
"@cite_57",
"@cite_52"
],
"mid": [
"2069641973",
"2059735611"
],
"abstract": [
"A series of observations ξ 1 , ξ 2 , ξ 3 ,... is presented to us and at each time n, when we have observed the first n of them, we are called upon to give our guess for what stochastic process produced the data. A universal scheme is given which, for any Bernoulli process (not necessarily independent), gives a sequence of processes that converges in a strong sense (the d 36-metric) to the real process. It is shown that in a certain sense the class of Bernoulli processes is the largest one for which such a universal scheme is possible",
"0. Summary. The objects of ergodic theory -measure spaces with measure-preserving transformation groups-wil l be called processes, those of topological dynamics-compact metric spaces with groups of homeomorphisms-will be called flows. We shall be concerned with what may be termed the \"arithmetic\" of these classes of objects. One may form products of processes and of flows, and one may also speak of factor processes and factor flows. By analogy with the integers, we may say that two processes are relatively prime if they have no non-trivial factors in common. An alternative condition is that whenever the two processes appear as factors of a third process, then their product too appears as a factor. In our theories it is unknown whether these two conditions are equivalent. We choose the second of these conditions as the more useful and refer to it as disjointness. Our first applications of the concept of disjointness are to the classification of processes and flows. It will appear that certain classes of processes (flows) may be characterized by the property of being disjoint from the members of other classes of processes (flows). For example the processes with entropy 0 are just those which are disjoint from all Bernoulli flows. Another application of disjointness of processes is to the following filtering problem. If xn and Yn represent two stationary stochastic processes, when can xn be filtered perfectly from Xn + Yn ? We will find (Part I, §9) that a sufficient condition is the disjointness of the processes in question. For flows the principal application of disjointness is to the tudy of properties of minimal sets (Part III). Consider the flow on the unit circle K = z: [zl = 1 that arises from the transformation z -- z 2. What can be said about the \"size\" of the minimal sets for this flow, that is, closed subsets of K invariant under z z , but not containing proper subsets with these properties. Uncountably many such minimal sets exist in K. Writing z = exp (2 ri Ean 2n), an = 0, 1, we see that this amounts to studying the mini-"
]
} |
1611.06173 | 2550947904 | A dynamical model consists of a continuous self-map @math of a compact state space @math and a continuous observation function @math . This paper considers the fitting of a parametrized family of dynamical models to an observed real-valued stochastic process using empirical risk minimization. The limiting behavior of the minimum risk parameters is studied in a general setting. We establish a general convergence theorem for minimum risk estimators and ergodic observations. We then study conditions under which empirical risk minimization can effectively separate the signal from the noise in an additive observational noise model. The key, necessary condition in the latter results is that the family of dynamical models has limited complexity, which is quantified through a notion of entropy for families of infinite sequences. Close connections between entropy and limiting average mean widths for stationary processes are established. | Some of Furstenberg's original results are extended in recent work of Lev, Peled, and Peres @cite_10 . Given an infinite sequence equal to a target signal plus noise, they consider the problem of detecting whether the signal is non-zero, and the problem of recovering the signal from the given sequence. Target sequences are assumed to belong to a known family (as in @cite_53 ), and their analysis places no restrictions (beyond measurability) on the detection and filtering procedures, which can be functions of the entire sequence of observations. Finally, we mention that statistical inference in the context of dynamical systems has been considered in a variety of subject areas; see the survey @cite_49 for a broad overview and references. Dynamical systems in the observational noise setting have been studied in @cite_24 @cite_22 @cite_31 , and statistical prediction in the context of dynamical systems has been considered in @cite_20 @cite_14 @cite_12 @cite_16 . | {
"cite_N": [
"@cite_14",
"@cite_22",
"@cite_53",
"@cite_24",
"@cite_49",
"@cite_31",
"@cite_16",
"@cite_10",
"@cite_12",
"@cite_20"
],
"mid": [
"2491437874",
"1992873793",
"2023307983",
"2071091379",
"1808775778",
"2000762696",
"1999381810",
"",
"2048207946",
"2963245453"
],
"abstract": [
"We study the density estimation problem with observations generated by certain dynamical systems that admit a unique underlying invariant Lebesgue density. Observations drawn from dynamical systems are not independent and moreover, usual mixing concepts may not be appropriate for measuring the dependence among these observations. By employing the @math -mixing concept to measure the dependence, we conduct statistical analysis on the consistency and convergence of the kernel density estimator. Our main results are as follows: First, we show that with properly chosen bandwidth, the kernel density estimator is universally consistent under @math -norm; Second, we establish convergence rates for the estimator with respect to several classes of dynamical systems under @math -norm. In the analysis, the density function @math is only assumed to be H \" o lder continuous which is a weak assumption in the literature of nonparametric density estimation and also more realistic in the dynamical system context. Last but not least, we prove that the same convergence rates of the estimator under @math -norm and @math -norm can be achieved when the density function is H \" o lder continuous, compactly supported and bounded. The bandwidth selection problem of the kernel density estimator for dynamical system is also discussed in our study via numerical simulations.",
"This paper is concerned with the problem of recovering a finite, deterministic time series from observations that are corrupted by additive, independent noise. A distinctive feature of this problem is that the available data exhibit long-range dependence and, as a consequence, existing statistical theory and methods are not readily applicable. This paper gives an analysis of the denoising problem that extends recent work of Lalley, but begins from first principles. Both positive and negative results are established. The positive results show that denoising is possible under somewhat restrictive conditions on the additive noise. The negative results show that, under more general conditions on the noise, no procedure can recover the underlying deterministic series.",
"The paper uses empirical process techniques to study the asymptotics of the least-squares estimator (LSE) for the fitting of a nonlinear regression function. By combining and extending ideas of Wu and Van de Geer, it establishes new consistency and central limit theorems that hold under only second moment assumptions on the errors. An application to a delicate example of Wu's illustrates the use of the new theorems, leading to a normal approximation to the LSE with unusual logarithmic rescalings.",
"The problem of extracting a signal x n from a noise-corrupted time series y n = x n + e n is considered. The signal x n is assumed to be generated by a discrete-time, deterministic, chaotic dynamical system F, in particular, x n = F n (x 0 ), where the initial point x 0 is assumed to lie in a compact hyperbolic F-invariant set. It is shown that (1) if the noise sequence e n is Gaussian then it is impossible to consistently recover the signal x n , but (2) if the noise sequence consists of i.i.d. random vectors uniformly bounded by a constant δ > 0, then it is possible to recover the signal x n provided δ < 5Δ, where Δ is a separation threshold for F. A filtering algorithm for the latter situation is presented.",
"",
"We consider the asymptotic consistency of maximum likelihood parameter estimation for dynamical systems observed with noise. Under suitable conditions on the dynamical systems and the observations, we show that maximum likelihood parameter estimation is consistent. Our proof involves ideas from both information theory and dynamical systems. Furthermore, we show how some well-studied properties of dynamical systems imply the general statistical properties related to maximum likelihood estimation. Finally, we exhibit classical families of dynamical systems for which maximum likelihood estimation is consistent. Examples include shifts of finite type with Gibbs measures and Axiom A attractors with SRB measures.",
"Suppose we are given a time series or a signal @math for @math . We consider the problem of predicting the signal in the interval @math based on a knowledge of its history and nothing more. We ask the following question: what is the largest value of @math for which a prediction can be made? We show that the answer to this question is contained in a fundamental result of information theory due to Wyner, Ziv, Ornstein, and Weiss. In particular, for the class of chaotic signals, the upper bound is @math in the limit @math , with @math being entropy in a sense that is explained in the text. If @math is small for @math , where @math is of the order of a characteristic time scale, the pattern of events leading up to @math is similar to the pattern of events leading up to @math . It is reasonable to expect @math to be a good predictor of @math . All existing methods for prediction use this idea...",
"",
"We consider the problem of forecasting the next (observable) state of an unknown ergodic dynamical system from a noisy observation of the present state. Our main result shows, for example, that support vector machines (SVMs) using Gaussian RBF kernels can learn the best forecaster from a sequence of noisy observations if (a) the unknown observational noise process is bounded and has a summable α-mixing rate and (b) the unknown ergodic dynamical system is defined by a Lipschitz continuous function on some compact subset of R d and has a summable decay of correlations for Lipschitz continuous functions. In order to prove this result we first establish a general consistency result for SVMs and all stochastic processes that satisfy a mixing notion that is substantially weaker than α-mixing.",
"We establish a Bernstein-type inequality for a class of stochastic processes that include the classical geometrically φ-mixing processes, Rio’s generalization of these processes, as well as many time-discrete dynamical systems. Modulo a logarithmic factor and some constants, our Bernsteintype inequality coincides with the classical Bernstein inequality for i.i.d. data. We further use this new Bernstein-type inequality to derive an oracle inequality for generic regularized empirical risk minimization algorithms and data generated by such processes. Applying this oracle inequality to support vector machines using the Gaussian kernels for both least squares and quantile regression, it turns out that the resulting learning rates match, up to some arbitrarily small extra term in the exponent, the optimal rates for i.i.d. processes."
]
} |
1611.06314 | 2530362799 | While social networks can provide an ideal platform for up-to-date information from individuals across the world, it has also proved to be a place where rumours fester and accidental or deliberate misinformation often emerges. In this article, we aim to support the task of making sense from social media data, and specifically, seek to build an autonomous message-classifier that filters relevant and trustworthy information from Twitter. For our work, we collected about 100 million public tweets, including users’ past tweets, from which we identified 72 rumours (41 true, 31 false). We considered over 80 trustworthiness measures including the authors’ profile and past behaviour, the social network connections (graphs), and the content of tweets themselves. We ran modern machine-learning classifiers over those measures to produce trustworthiness scores at various time windows from the outbreak of the rumour. Such time-windows were key as they allowed useful insight into the progression of the rumours. From our findings, we identified that our model was significantly more accurate than similar studies in the literature. We also identified critical attributes of the data that give rise to the trustworthiness scores assigned. Finally we developed a software demonstration that provides a visual user interface to allow the user to examine the analysis. | Recently there has been progress from experts from different fields of academia in exploring the characteristics of the data source and content that will enable us to determine the veracity of information in an autonomous and efficient manner @cite_29 @cite_21 @cite_1 . Key concepts include information quality, which can be defined as an assessment or measure of how fit an information object is for use, and information trustworthiness, which is the likelihood that a piece of information will preserve a user's trust, or belief, in it @cite_24 . These concepts may overlap and indeed, increasing one (e.g., quality) may lead to an increase in the other (e.g., trustworthiness). Other relevant factors include Accuracy (Free-of-error), Reliability, Objectivity (Bias), Believability (Likelihood, Plausibility of arguments), Popularity, Competence and Provenance @cite_23 @cite_19 @cite_30 @cite_28 @cite_0 . | {
"cite_N": [
"@cite_30",
"@cite_28",
"@cite_29",
"@cite_21",
"@cite_1",
"@cite_24",
"@cite_19",
"@cite_0",
"@cite_23"
],
"mid": [
"2107219896",
"1998165099",
"2032897813",
"2031490232",
"2095892879",
"1566370201",
"2002842121",
"2039635113",
"1567491469"
],
"abstract": [
"Trust in information is developing into a vitally important topic as the Internet becomes increasingly ubiquitous within society. Although many discussions of trust in this environment focus on issues like security, technical reliability, or e-commerce, few address the problem of trust in the information obtained from the Internet. The authors assert that there is a strong need for theoretical and empirical research on trust within the field of information science. As an initial step, the present study develops a model of trust in digital information by integrating the research on trust from the behavioral and social sciences with the research on information quality and human– computer interaction. The model positions trust as a key mediating variable between information quality and information usage, with important consequences for both the producers and consumers of digital information. The authors close by outlining important directions for future research on trust in information science and technology. © 2008 Wiley Periodicals, Inc.",
"Trust is an integral part of the Semantic Web architecture. Most prior work on trusts focuses on entity-centered issues such as authentication and reputation and does not take into account the content, i.e., the nature and use of the information being exchanged. This paper defines content trust and discusses it in the context of other trust measures that have been previously studied. We introduce several factors that users consider in deciding whether to trust the content provided by a Web resource. Our goal is to discern which of these factors could be captured in practice with minimal user interaction in order to maximize the quality of the system's trust estimates. We present results on a study to determine which factors were more important to capture, and describe a simulation environment that we have designed to study alternative models of content trust.",
"The problem of identifying rumors is of practical importance especially in online social networks, since information can diffuse more rapidly and widely than the offline counterpart. In this paper, we identify characteristics of rumors by examining the following three aspects of diffusion: temporal, structural, and linguistic. For the temporal characteristics, we propose a new periodic time series model that considers daily and external shock cycles, where the model demonstrates that rumor likely have fluctuations over time. We also identify key structural and linguistic differences in the spread of rumors and non-rumors. Our selected features classify rumors with high precision and recall in the range of 87 to 92 , that is higher than other states of the arts on rumor classification.",
"Information that propagates through social networks can carry a lot of false claims. For example, rumors on certain topics can propagate rapidly leading to a large number of nodes reporting the same (incorrect) observations. In this paper, we describe an approach for nding the rumor source and assessing the likelihood that a piece of information is in fact a rumor, in the absence of data provenance information. We model the social network as a directed graph, where vertices represent individuals and directed edges represent information ow (e.g., who follows whom on Twitter). A number of monitor nodes are injected into the network whose job is to report data they receive. Our algorithm identies rumors and their sources by observing which of the monitors received the given piece of information and which did not. We show that, with a sucient number of monitor nodes, it is possible to recognize most rumors and their sources with high accuracy.",
"The spread of false rumours during emergencies can jeopardise the well-being of citizens as they are monitoring the stream of news from social media to stay abreast of the latest updates. In this paper, we describe the methodology we have developed within the PHEME project for the collection and sampling of conversational threads, as well as the tool we have developed to facilitate the annotation of these threads so as to identify rumourous ones. We describe the annotation task conducted on threads collected during the 2014 Ferguson unrest and we present and analyse our findings. Our results show that we can collect effectively social media rumours and identify multiple rumours associated with a range of stories that would have been hard to identify by relying on existing techniques that need manual input of rumour-specific keywords.",
"The importance and value of information cannot be disputed. It is used as basis for menial and mission-critical tasks alike. In a society where information is so easily publicised and freely accessible, however, being able to assess information quality and trustworthiness is paramount. With appreciation of this fact, our paper seeks to navigate these two mature fields and define the latest state-of-the-art. The novelty of this work is found in the provision of an up-to-date review, a research survey which considers and links provenance, quality and trustworthiness, and a literature analysis that includes a first-look review at some of these aspects within the social- media domain. This factor-based review should provide an ideal grounding for future research that assesses interaction between these three topics, which may then also progress to associations with information assurance and security at large. To demonstrate how some of the factors might be considered, we also examine their application to a commonplace scenario.",
"The paper discusses and analyzes the notion of information quality in terms of a pragmatic philosophy of language. It is argued that the notion of information quality is of great importance, and needs to be situated better within a sound philosophy of information to help frame information quality in a broader conceptual light. It is found that much research on information quality conceptualizes information quality as either an inherent property of the information itself, or as an individual mental construct of the users. The notion of information quality is often not situated within a philosophy of information. This paper outlines a conceptual framework in which information is regarded as a semiotic sign, and extends that notion with Paul Grice's pragmatic philosophy of language to provide a conversational notion of information quality that is contextual and tied to the notion of meaning.",
"",
"Poor data quality (DQ) can have substantial social and economic impacts. Although firms are improving data quality with practical approaches and tools, their improvement efforts tend to focus narrowly on accuracy. We believe that data consumers have a much broader data quality conceptualization than IS professionals realize. The purpose of this paper is to develop a framework that captures the aspects of data quality that are important to data consumers.A two-stage survey and a two-phase sorting study were conducted to develop a hierarchical framework for organizing data quality dimensions. This framework captures dimensions of data quality that are important to data consumers. Intrinsic DQ denotes that data have quality in their own right. Contextual DQ highlights the requirement that data quality must be considered within the context of the task at hand. Representational DQ and accessibility DQ emphasize the importance of the role of systems. These findings are consistent with our understanding that high-quality data should be intrinsically good, contextually appropriate for the task, clearly represented, and accessible to the data consumer.Our framework has been used effectively in industry and government. Using this framework, IS managers were able to better understand and meet their data consumers' data quality needs. The salient feature of this research study is that quality attributes of data are collected from data consumers instead of being defined theoretically or based on researchers' experience. Although exploratory, this research provides a basis for future studies that measure data quality along the dimensions of this framework."
]
} |
1611.06314 | 2530362799 | While social networks can provide an ideal platform for up-to-date information from individuals across the world, it has also proved to be a place where rumours fester and accidental or deliberate misinformation often emerges. In this article, we aim to support the task of making sense from social media data, and specifically, seek to build an autonomous message-classifier that filters relevant and trustworthy information from Twitter. For our work, we collected about 100 million public tweets, including users’ past tweets, from which we identified 72 rumours (41 true, 31 false). We considered over 80 trustworthiness measures including the authors’ profile and past behaviour, the social network connections (graphs), and the content of tweets themselves. We ran modern machine-learning classifiers over those measures to produce trustworthiness scores at various time windows from the outbreak of the rumour. Such time-windows were key as they allowed useful insight into the progression of the rumours. From our findings, we identified that our model was significantly more accurate than similar studies in the literature. We also identified critical attributes of the data that give rise to the trustworthiness scores assigned. Finally we developed a software demonstration that provides a visual user interface to allow the user to examine the analysis. | To date, there has been a significant number of articles, in both academia and industry, that have been published on the topic of information quality and trustworthiness online, particularly in the case of Twitter. @cite_2 @cite_18 focus on developing automatic methods for assessing the credibility of posts on Twitter. They utilise a machine learning approach to the problem and for their analysis use a vast range of features grouped according to whether they are user-based, topic-based or propagation-based. @cite_12 @cite_14 have also aimed towards developing a wider framework to support the assessment of the trustworthiness of information. This framework builds on trust and quality metrics such as those already reviewed, and outlines a policy-based approach to measurement. The key aspect of this approach is that it allows organisations and users to set policies to mediate content and secondly, to weight the importance of individual trust factors (e.g., expressing that for a particular context, location is more important than corroboration). The result is a tailored trustworthiness score for information suited to the individual's unique requirements. | {
"cite_N": [
"@cite_18",
"@cite_14",
"@cite_12",
"@cite_2"
],
"mid": [
"1971494700",
"1846948195",
"2113046186",
"2084591134"
],
"abstract": [
"Purpose – Twitter is a popular microblogging service which has proven, in recent years, its potential for propagating news and information about developing events. The purpose of this paper is to focus on the analysis of information credibility on Twitter. The purpose of our research is to establish if an automatic discovery process of relevant and credible news events can be achieved. Design methodology approach – The paper follows a supervised learning approach for the task of automatic classification of credible news events. A first classifier decides if an information cascade corresponds to a newsworthy event. Then a second classifier decides if this cascade can be considered credible or not. The paper undertakes this effort training over a significant amount of labeled data, obtained using crowdsourcing tools. The paper validates these classifiers under two settings: the first, a sample of automatically detected Twitter “trends” in English, and second, the paper tests how well this model transfers to...",
"The vast amount of information available online places decision makers wishing to use this content in an advantageous but also very difficult position. The advantages stem from the volume of content from a variety of sources that is readily available; the difficulties arise because of the often unknown quality and trustworthiness of the information – is it fact, opinion or purely meant to deceive? In this paper we reflect on and extend current work on information trust and quality metrics which can be used to address this difficulty. Specifically, we propose new metrics as worthy of consideration and the new combinatorics required to take measurements of the various trust factors into a single score. These feed into our existing overarching policy-based approach that uses trustworthiness metrics to support decision-making online.",
"Information is the currency of the digital age – it is constantly communicated, exchanged and bartered, most commonly to support human understanding and decision-making. While the Internet and Web 2.0 have been pivotal in streamlining many of the information creation and dissemination processes, they have significantly complicated matters for users as well. Most notably, the substantial increase in the amount of content available online has introduced an information overload problem, while also exposing content with largely unknown levels of quality, leaving many users with the difficult question of, what information to trust? In this article we approach this problem from two perspectives, both aimed at supporting human decision-making using online information. First, we focus on the task of measuring the extent to which individuals should trust a piece of openly-sourced information (e.g., from Twitter, Facebook or a blog); this considers a range of factors and metrics in information provenance, quality and infrastructure integrity, and the person’s own preferences and opinion. Having calculated a measure of trustworthiness for an information item, we then consider how this rating and the related content could be communicated to users in a cognitively-enhanced manner, so as to build confidence in the information only where and when appropriate. This work concentrates on a range of potential visualisation techniques for trust, with special focus on radar graphs, and draws inspiration from the fields of Human-Computer Interaction (HCI), System Usability and Risk Communication. The novelty of our contribution stems from the comprehensive approach taken to address this very topical problem, ensuring that the trustworthiness of openly-sourced information is adequately measured and effectively communicated to users, thus enabling them to make informed decisions.",
"We analyze the information credibility of news propagated through Twitter, a popular microblogging service. Previous research has shown that most of the messages posted on Twitter are truthful, but the service is also used to spread misinformation and false rumors, often unintentionally. On this paper we focus on automatic methods for assessing the credibility of a given set of tweets. Specifically, we analyze microblog postings related to \"trending\" topics, and classify them as credible or not credible, based on features extracted from them. We use features from users' posting and re-posting (\"re-tweeting\") behavior, from the text of the posts, and from citations to external sources. We evaluate our methods using a significant number of human assessments about the credibility of items on a recent sample of Twitter postings. Our results shows that there are measurable differences in the way messages propagate, that can be used to classify them automatically as credible or not credible, with precision and recall in the range of 70 to 80 ."
]
} |
1611.06314 | 2530362799 | While social networks can provide an ideal platform for up-to-date information from individuals across the world, it has also proved to be a place where rumours fester and accidental or deliberate misinformation often emerges. In this article, we aim to support the task of making sense from social media data, and specifically, seek to build an autonomous message-classifier that filters relevant and trustworthy information from Twitter. For our work, we collected about 100 million public tweets, including users’ past tweets, from which we identified 72 rumours (41 true, 31 false). We considered over 80 trustworthiness measures including the authors’ profile and past behaviour, the social network connections (graphs), and the content of tweets themselves. We ran modern machine-learning classifiers over those measures to produce trustworthiness scores at various time windows from the outbreak of the rumour. Such time-windows were key as they allowed useful insight into the progression of the rumours. From our findings, we identified that our model was significantly more accurate than similar studies in the literature. We also identified critical attributes of the data that give rise to the trustworthiness scores assigned. Finally we developed a software demonstration that provides a visual user interface to allow the user to examine the analysis. | Having established a view on the characteristics of information quality and trustworthiness researchers focused on designing systems for rumour detection. @cite_29 examined how rumours spread in social media and which characteristics may provide evidence in identifying rumours. The authors focused on three aspects of diffusion of a rumour, namely the temporal, the structural and the linguistic and identified key differences in the spread of rumours and non-rumours. Their results suggest that they were able to identify rumours with up to 92 Another approach is presented in @cite_1 , where the authors introduce a new definition for rumours and provide a novel methodology on how to collect and annotate tweets associated with an event. In contrast to other approaches which depend on predefining a set of rumours and then associating tweets to these, this methodology involves reading the replies to tweets and categorising them to stories or threads. It is a tool intended to facilitate the process of developing a machine learning approach to automatically identify rumours. | {
"cite_N": [
"@cite_29",
"@cite_1"
],
"mid": [
"2032897813",
"2095892879"
],
"abstract": [
"The problem of identifying rumors is of practical importance especially in online social networks, since information can diffuse more rapidly and widely than the offline counterpart. In this paper, we identify characteristics of rumors by examining the following three aspects of diffusion: temporal, structural, and linguistic. For the temporal characteristics, we propose a new periodic time series model that considers daily and external shock cycles, where the model demonstrates that rumor likely have fluctuations over time. We also identify key structural and linguistic differences in the spread of rumors and non-rumors. Our selected features classify rumors with high precision and recall in the range of 87 to 92 , that is higher than other states of the arts on rumor classification.",
"The spread of false rumours during emergencies can jeopardise the well-being of citizens as they are monitoring the stream of news from social media to stay abreast of the latest updates. In this paper, we describe the methodology we have developed within the PHEME project for the collection and sampling of conversational threads, as well as the tool we have developed to facilitate the annotation of these threads so as to identify rumourous ones. We describe the annotation task conducted on threads collected during the 2014 Ferguson unrest and we present and analyse our findings. Our results show that we can collect effectively social media rumours and identify multiple rumours associated with a range of stories that would have been hard to identify by relying on existing techniques that need manual input of rumour-specific keywords."
]
} |
1611.06224 | 2556607581 | Deep learning has improved state-of-the-art results in many important fields, and has been the subject of much research in recent years, leading to the development of several systems for facilitating deep learning. Current systems, however, mainly focus on model building and training phases, while the issues of data management, model sharing, and lifecycle management are largely ignored. Deep learning modeling lifecycle generates a rich set of data artifacts, such as learned parameters and training logs, and comprises of several frequently conducted tasks, e.g., to understand the model behaviors and to try out new models. Dealing with such artifacts and tasks is cumbersome and largely left to the users. This paper describes our vision and implementation of a data and lifecycle management system for deep learning. First, we generalize model exploration and model enumeration queries from commonly conducted tasks by deep learning modelers, and propose a high-level domain specific language (DSL), inspired by SQL, to raise the abstraction level and accelerate the modeling process. To manage the data artifacts, especially the large amount of checkpointed float parameters, we design a novel model versioning system (dlv), and a read-optimized parameter archival storage system (PAS) that minimizes storage footprint and accelerates query workloads without losing accuracy. PAS archives versioned models using deltas in a multi-resolution fashion by separately storing the less significant bits, and features a novel progressive query (inference) evaluation algorithm. Third, we show that archiving versioned models using deltas poses a new dataset versioning problem and we develop efficient algorithms for solving it. We conduct extensive experiments over several real datasets from computer vision domain to show the efficiency of the proposed techniques. | Machine Learning Systems: There have been several high-profile deep learning systems in recent years, but those typically focus on the training aspects (e.g., on distributed training, how to utilize GPUs or allow symbolic formulas, etc.) @cite_18 @cite_21 @cite_14 @cite_4 @cite_19 . The data management and lifecycle management challenges discussed above have been largely ignored so far, but are becoming critical as the use of deep learning permeates through a variety of application domains, since those pose a high barrier to entry for many potential users. In the database community, there has been increasing work on developing general-purpose systems for supporting machine learning @cite_0 @cite_32 @cite_3 , including pushing predictive models into databases @cite_6 @cite_20 , accelerating tasks using database optimizing methods physical design @cite_17 @cite_0 , and managing modeling lifecycles and serving predictive models in advanced ways @cite_24 @cite_1 . is motivated by similar principles; aside from a focus on DNNs, it also supports versioning as a first-class construct @cite_2 which differentiates it from that work. | {
"cite_N": [
"@cite_18",
"@cite_14",
"@cite_4",
"@cite_21",
"@cite_1",
"@cite_32",
"@cite_3",
"@cite_6",
"@cite_0",
"@cite_19",
"@cite_24",
"@cite_2",
"@cite_20",
"@cite_17"
],
"mid": [
"2155893237",
"2168231600",
"",
"2953384591",
"",
"",
"",
"117630655",
"2032775418",
"1982767656",
"",
"",
"2090850279",
"2099102906"
],
"abstract": [
"Caffe provides multimedia scientists and practitioners with a clean and modifiable framework for state-of-the-art deep learning algorithms and a collection of reference models. The framework is a BSD-licensed C++ library with Python and MATLAB bindings for training and deploying general-purpose convolutional neural networks and other deep models efficiently on commodity architectures. Caffe fits industry and internet-scale media needs by CUDA GPU computation, processing over 40 million images a day on a single K40 or Titan GPU (approx 2 ms per image). By separating model representation from actual implementation, Caffe allows experimentation and seamless switching among platforms for ease of development and deployment from prototyping machines to cloud environments. Caffe is maintained and developed by the Berkeley Vision and Learning Center (BVLC) with the help of an active community of contributors on GitHub. It powers ongoing research projects, large-scale industrial applications, and startup prototypes in vision, speech, and multimedia.",
"Recent work in unsupervised feature learning and deep learning has shown that being able to train large models can dramatically improve performance. In this paper, we consider the problem of training a deep network with billions of parameters using tens of thousands of CPU cores. We have developed a software framework called DistBelief that can utilize computing clusters with thousands of machines to train large models. Within this framework, we have developed two algorithms for large-scale distributed training: (i) Downpour SGD, an asynchronous stochastic gradient descent procedure supporting a large number of model replicas, and (ii) Sandblaster, a framework that supports a variety of distributed batch optimization procedures, including a distributed implementation of L-BFGS. Downpour SGD and Sandblaster L-BFGS both increase the scale and speed of deep network training. We have successfully used our system to train a deep network 30x larger than previously reported in the literature, and achieves state-of-the-art performance on ImageNet, a visual object recognition task with 16 million images and 21k categories. We show that these same techniques dramatically accelerate the training of a more modestly- sized deep network for a commercial speech recognition service. Although we focus on and report performance of these methods as applied to training large neural networks, the underlying algorithms are applicable to any gradient-based machine learning algorithm.",
"",
"TensorFlow is a machine learning system that operates at large scale and in heterogeneous environments. TensorFlow uses dataflow graphs to represent computation, shared state, and the operations that mutate that state. It maps the nodes of a dataflow graph across many machines in a cluster, and within a machine across multiple computational devices, including multicore CPUs, general-purpose GPUs, and custom designed ASICs known as Tensor Processing Units (TPUs). This architecture gives flexibility to the application developer: whereas in previous \"parameter server\" designs the management of shared state is built into the system, TensorFlow enables developers to experiment with novel optimizations and training algorithms. TensorFlow supports a variety of applications, with particularly strong support for training and inference on deep neural networks. Several Google services use TensorFlow in production, we have released it as an open-source project, and it has become widely used for machine learning research. In this paper, we describe the TensorFlow dataflow model in contrast to existing systems, and demonstrate the compelling performance that TensorFlow achieves for several real-world applications.",
"",
"",
"",
"This paper argues that next generation database management systems should incorporate a predictive model management component to effectively support both inward-facing applications, such as self management, and user-facing applications such as data-driven predictive analytics. We draw an analogy between model management and data management functionality and discuss how model management can leverage profiling, physical design and query optimization techniques, as well as the pertinent challenges. We then describe the early design and architecture of Longview, a predictive DBMS prototype that we are building at Brown, along with a case study of how models can be used to predict query execution performance.",
"Enterprise data analytics is a booming area in the data management industry. Many companies are racing to develop toolkits that closely integrate statistical and machine learning techniques with data management systems. Almost all such toolkits assume that the input to a learning algorithm is a single table. However, most relational datasets are not stored as single tables due to normalization. Thus, analysts often perform key-foreign key joins before learning on the join output. This strategy of learning after joins introduces redundancy avoided by normalization, which could lead to poorer end-to-end performance and maintenance overheads due to data duplication. In this work, we take a step towards enabling and optimizing learning over joins for a common class of machine learning techniques called generalized linear models that are solved using gradient descent algorithms in an RDBMS setting. We present alternative approaches to learn over a join that are easy to implement over existing RDBMSs. We introduce a new approach named factorized learning that pushes ML computations through joins and avoids redundancy in both I O and computations. We study the tradeoff space for all our approaches both analytically and empirically. Our results show that factorized learning is often substantially faster than the alternatives, but is not always the fastest, necessitating a cost-based approach. We also discuss extensions of all our approaches to multi-table joins as well as to Hive.",
"Recently, deep learning techniques have enjoyed success in various multimedia applications, such as image classification and multi-modal data analysis. Two key factors behind deep learning's remarkable achievement are the immense computing power and the availability of massive training datasets, which enable us to train large models to capture complex regularities of the data. There are two challenges to overcome before deep learning can be widely adopted in multimedia and other applications. One is usability, namely the implementation of different models and training algorithms must be done by non-experts without much effort. The other is scalability, that is the deep learning system must be able to provision for a huge demand of computing resources for training large models with massive datasets. To address these two challenges, in this paper, we design a distributed deep learning platform called SINGA which has an intuitive programming model and good scalability. Our experience with developing and training deep learning models for real-life multimedia applications in SINGA shows that the platform is both usable and scalable.",
"",
"",
"The increasing use of statistical data analysis in enterprise applications has created an arms race among database vendors to offer ever more sophisticated in-database analytics. One challenge in this race is that each new statistical technique must be implemented from scratch in the RDBMS, which leads to a lengthy and complex development process. We argue that the root cause for this overhead is the lack of a unified architecture for in-database analytics. Our main contribution in this work is to take a step towards such a unified architecture. A key benefit of our unified architecture is that performance optimizations for analytics techniques can be studied generically instead of an ad hoc, per-technique fashion. In particular, our technical contributions are theoretical and empirical studies of two key factors that we found impact performance: the order data is stored, and parallelization of computations on a single-node multicore RDBMS. We demonstrate the feasibility of our architecture by integrating several popular analytics techniques into two commercial and one open-source RDBMS. Our architecture requires changes to only a few dozen lines of code to integrate a new statistical technique. We then compare our approach with the native analytics tools offered by the commercial RDBMSes on various analytics tasks, and validate that our approach achieves competitive or higher performance, while still achieving the same quality.",
"There is an arms race in the data management industry to support analytics, in which one critical step is feature selection, the process of selecting a feature set that will be used to build a statistical model. Analytics is one of the biggest topics in data management, and feature selection is widely regarded as the most critical step of analytics; thus, we argue that managing the feature selection process is a pressing data management challenge. We study this challenge by describing a feature-selection language and a supporting prototype system that builds on top of current industrial, R-integration layers. From our interactions with analysts, we learned that feature selection is an interactive, human-in-the-loop process, which means that feature selection workloads are rife with reuse opportunities. Thus, we study how to materialize portions of this computation using not only classical database materialization optimizations but also methods that have not previously been used in database optimization, including structural decomposition methods (like QR factorization) and warmstart. These new methods have no analog in traditional SQL systems, but they may be interesting for array and scientific database applications. On a diverse set of data sets and programs, we find that traditional database-style approaches that ignore these new opportunities are more than two orders of magnitude slower than an optimal plan in this new tradeoff space across multiple R-backends. Furthermore, we show that it is possible to build a simple cost-based optimizer to automatically select a near-optimal execution plan for feature selection."
]
} |
1611.06224 | 2556607581 | Deep learning has improved state-of-the-art results in many important fields, and has been the subject of much research in recent years, leading to the development of several systems for facilitating deep learning. Current systems, however, mainly focus on model building and training phases, while the issues of data management, model sharing, and lifecycle management are largely ignored. Deep learning modeling lifecycle generates a rich set of data artifacts, such as learned parameters and training logs, and comprises of several frequently conducted tasks, e.g., to understand the model behaviors and to try out new models. Dealing with such artifacts and tasks is cumbersome and largely left to the users. This paper describes our vision and implementation of a data and lifecycle management system for deep learning. First, we generalize model exploration and model enumeration queries from commonly conducted tasks by deep learning modelers, and propose a high-level domain specific language (DSL), inspired by SQL, to raise the abstraction level and accelerate the modeling process. To manage the data artifacts, especially the large amount of checkpointed float parameters, we design a novel model versioning system (dlv), and a read-optimized parameter archival storage system (PAS) that minimizes storage footprint and accelerates query workloads without losing accuracy. PAS archives versioned models using deltas in a multi-resolution fashion by separately storing the less significant bits, and features a novel progressive query (inference) evaluation algorithm. Third, we show that archiving versioned models using deltas poses a new dataset versioning problem and we develop efficient algorithms for solving it. We conduct extensive experiments over several real datasets from computer vision domain to show the efficiency of the proposed techniques. | DNN Compression: There has been increasing interest on compressing models, motivated in part by the need to deploy them on devices with simple instruction sets, low memory, and or energy constraints @cite_36 @cite_27 @cite_8 . However, the goal of those works is simplify the model in a lossy manner with as little loss of accuracy as possible, which makes this work orthogonal to lossless compression the archival approach we take in ; in fact, simplified models are likely to compress much better, magnifying the gains of our approach as our experimental results show. Further, these methods often require heavy retraining or expensive computations ( k -means, SVD, etc.) to derive simpler models, which makes them too heavyweight in an interactive setting for which is designed. | {
"cite_N": [
"@cite_36",
"@cite_27",
"@cite_8"
],
"mid": [
"",
"2167215970",
"2264905057"
],
"abstract": [
"",
"We present techniques for speeding up the test-time evaluation of large convolutional networks, designed for object recognition tasks. These models deliver impressive accuracy, but each image evaluation requires millions of floating point operations, making their deployment on smartphones and Internet-scale clusters problematic. The computation is dominated by the convolution operations in the lower layers of the model. We exploit the redundancy present within the convolutional filters to derive approximations that significantly reduce the required computation. Using large state-of-the-art models, we demonstrate speedups of convolutional layers on both CPU and GPU by a factor of 2 x, while keeping the accuracy within 1 of the original model.",
"The complexity of deep neural network algorithms for hardware implementation can be much lowered by optimizing the word-length of weights and signals. Direct quantization of floating-point weights, however, does not show good performance when the number of bits assigned is small. Retraining of quantized networks has been developed to relieve this problem. In this work, the effects of retraining are analyzed for a feedforward deep neural network (FFDNN) and a convolutional neural network (CNN). The network complexity is controlled to know their effects on the resiliency of quantized networks by retraining. The complexity of the FFDNN is controlled by varying the unit size in each hidden layer and the number of layers, while that of the CNN is done by modifying the feature map configuration. We find that the performance gap between the floating-point and the retrain-based ternary (+1, 0, -1) weight neural networks exists with a fair amount in 'complexity limited' networks, but the discrepancy almost vanishes in fully complex networks whose capability is limited by the training data, rather than by the number of connections. This research shows that highly complex DNNs have the capability of absorbing the effects of severe weight quantization through retraining, but connection limited networks are less resilient. This paper also presents the effective compression ratio to guide the trade-off between the network size and the precision when the hardware resource is limited."
]
} |
1611.06224 | 2556607581 | Deep learning has improved state-of-the-art results in many important fields, and has been the subject of much research in recent years, leading to the development of several systems for facilitating deep learning. Current systems, however, mainly focus on model building and training phases, while the issues of data management, model sharing, and lifecycle management are largely ignored. Deep learning modeling lifecycle generates a rich set of data artifacts, such as learned parameters and training logs, and comprises of several frequently conducted tasks, e.g., to understand the model behaviors and to try out new models. Dealing with such artifacts and tasks is cumbersome and largely left to the users. This paper describes our vision and implementation of a data and lifecycle management system for deep learning. First, we generalize model exploration and model enumeration queries from commonly conducted tasks by deep learning modelers, and propose a high-level domain specific language (DSL), inspired by SQL, to raise the abstraction level and accelerate the modeling process. To manage the data artifacts, especially the large amount of checkpointed float parameters, we design a novel model versioning system (dlv), and a read-optimized parameter archival storage system (PAS) that minimizes storage footprint and accelerates query workloads without losing accuracy. PAS archives versioned models using deltas in a multi-resolution fashion by separately storing the less significant bits, and features a novel progressive query (inference) evaluation algorithm. Third, we show that archiving versioned models using deltas poses a new dataset versioning problem and we develop efficient algorithms for solving it. We conduct extensive experiments over several real datasets from computer vision domain to show the efficiency of the proposed techniques. | DNNs with Low Precision Floats: Low precision floats are exploited in accelerating training and testing systems @cite_11 @cite_8 @cite_22 courbariaux2014training , showing techniques and empirical results when training, testing with limited precisions. differs from their work by exploiting storing large collection of parameters archiving, and use segmented floats to answer lifecycle modeling queries. | {
"cite_N": [
"@cite_8",
"@cite_22",
"@cite_11"
],
"mid": [
"2264905057",
"1841592590",
"587794757"
],
"abstract": [
"The complexity of deep neural network algorithms for hardware implementation can be much lowered by optimizing the word-length of weights and signals. Direct quantization of floating-point weights, however, does not show good performance when the number of bits assigned is small. Retraining of quantized networks has been developed to relieve this problem. In this work, the effects of retraining are analyzed for a feedforward deep neural network (FFDNN) and a convolutional neural network (CNN). The network complexity is controlled to know their effects on the resiliency of quantized networks by retraining. The complexity of the FFDNN is controlled by varying the unit size in each hidden layer and the number of layers, while that of the CNN is done by modifying the feature map configuration. We find that the performance gap between the floating-point and the retrain-based ternary (+1, 0, -1) weight neural networks exists with a fair amount in 'complexity limited' networks, but the discrepancy almost vanishes in fully complex networks whose capability is limited by the training data, rather than by the number of connections. This research shows that highly complex DNNs have the capability of absorbing the effects of severe weight quantization through retraining, but connection limited networks are less resilient. This paper also presents the effective compression ratio to guide the trade-off between the network size and the precision when the hardware resource is limited.",
"Training of large-scale deep neural networks is often constrained by the available computational resources. We study the effect of limited precision data representation and computation on neural network training. Within the context of low-precision fixed-point computations, we observe the rounding scheme to play a crucial role in determining the network's behavior during training. Our results show that deep networks can be trained using only 16-bit wide fixed-point number representation when using stochastic rounding, and incur little to no degradation in the classification accuracy. We also demonstrate an energy-efficient hardware accelerator that implements low-precision fixed-point arithmetic with stochastic rounding.",
"Recent advances in deep learning have made the use of large, deep neural networks with tens of millions of parameters suitable for a number of applications that require real-time processing. The sheer size of these networks can represent a challenging computational burden, even for modern CPUs. For this reason, GPUs are routinely used instead to train and run such networks. This paper is a tutorial for students and researchers on some of the techniques that can be used to reduce this computational cost considerably on modern x86 CPUs. We emphasize data layout, batching of the computation, the use of SSE2 instructions, and particularly leverage SSSE3 and SSE4 fixed-point instructions which provide a 3× improvement over an optimized floating-point baseline. We use speech recognition as an example task, and show that a real-time hybrid hidden Markov model neural network (HMM NN) large vocabulary system can be built with a 10× speedup over an unoptimized baseline and a 4× speedup over an aggressively optimized floating-point baseline at no cost in accuracy. The techniques described extend readily to neural network training and provide an effective alternative to the use of specialized hardware."
]
} |
1611.06224 | 2556607581 | Deep learning has improved state-of-the-art results in many important fields, and has been the subject of much research in recent years, leading to the development of several systems for facilitating deep learning. Current systems, however, mainly focus on model building and training phases, while the issues of data management, model sharing, and lifecycle management are largely ignored. Deep learning modeling lifecycle generates a rich set of data artifacts, such as learned parameters and training logs, and comprises of several frequently conducted tasks, e.g., to understand the model behaviors and to try out new models. Dealing with such artifacts and tasks is cumbersome and largely left to the users. This paper describes our vision and implementation of a data and lifecycle management system for deep learning. First, we generalize model exploration and model enumeration queries from commonly conducted tasks by deep learning modelers, and propose a high-level domain specific language (DSL), inspired by SQL, to raise the abstraction level and accelerate the modeling process. To manage the data artifacts, especially the large amount of checkpointed float parameters, we design a novel model versioning system (dlv), and a read-optimized parameter archival storage system (PAS) that minimizes storage footprint and accelerates query workloads without losing accuracy. PAS archives versioned models using deltas in a multi-resolution fashion by separately storing the less significant bits, and features a novel progressive query (inference) evaluation algorithm. Third, we show that archiving versioned models using deltas poses a new dataset versioning problem and we develop efficient algorithms for solving it. We conduct extensive experiments over several real datasets from computer vision domain to show the efficiency of the proposed techniques. | Stability Analysis Results Stability analysis of is studied in the literature @cite_5 @cite_30 @cite_10 , where the problem setting is perturbation analysis regardless specific data, rather focus on statistical measures of stability. uses basic perturbation analysis techniques and focus on novel progressive query answering in a segmented float storage. | {
"cite_N": [
"@cite_30",
"@cite_5",
"@cite_10"
],
"mid": [
"2107808739",
"2167677193",
"2075126893"
],
"abstract": [
"An important issue in the design and implementation of a neural network is the sensitivity of its output to input and weight perturbations. In this paper, we discuss the sensitivity of the most popular and general feedforward neural networks-multilayer perceptron (MLP). The sensitivity is defined as the mathematical expectation of the output errors of the MLP due to input and weight perturbations with respect to all input and weight values in a given continuous interval. The sensitivity for a single neuron is discussed first and an analytical expression that is a function of the absolute values of input and weight perturbations is approximately derived. Then an algorithm is given to compute the sensitivity for the entire MLP. As intuitively expected, the sensitivity increases with input and weight perturbations, but the increase has an upper bound that is determined by the structural configuration of the MLP, namely the number of neurons per layer and the number of layers. There exists an optimal value for the number of neurons in a layer, which yields the highest sensitivity value. The effect caused by the number of layers is quite unexpected. The sensitivity of a neural network may decrease at first and then almost keeps constant while the number increases.",
"An analysis is made of the sensitivity of feedforward layered networks of Adaline elements (threshold logic units) to weight errors. An approximation is derived which expresses the probability of error for an output neuron of a large network (a network with many neurons per layer) as a function of the percentage change in the weights. As would be expected, the probability of error increases with the number of layers in the network and with the percentage change in the weights. The probability of error is essentially independent of the number of weights per neuron and of the number of neurons per layer, as long as these numbers are large (on the order of 100 or more). >",
"The sensitivity of a neural network's output to its inputs' perturbations is an important measure for evaluating the network's performance. To make the sensitivity be a practical tool for designing and implementing Multilayer Perceptrons (MLPs), this paper proposes a general approach to quantify the sensitivity of MLPs. The sensitivity is defined as the mathematical expectation of absolute output deviations due to input perturbations with respect to all possible inputs, and computed following a bottom-up way, in which the sensitivity of a neuron is first considered and then is that of the entire network. The main contribution of the approach is that it requests a weak assumption on the input, that is, input elements need only to be independent of each other without being restricted to have a certain type of distribution and thus is more applicable to real applications. Some experimental results on artificial datasets and real datasets demonstrate the proposed approach is highly accurate."
]
} |
1611.06391 | 2556016755 | Recently, compressed sensing (CS) computed tomography (CT) using sparse projection views has been extensively investigated to reduce the potential risk of radiation to patient. However, due to the insufficient number of projection views, an analytic reconstruction approach results in severe streaking artifacts and CS-based iterative approach is computationally very expensive. To address this issue, here we propose a novel deep residual learning approach for sparse view CT reconstruction. Specifically, based on a novel persistent homology analysis showing that the manifold of streaking artifacts is topologically simpler than original ones, a deep residual learning architecture that estimates the streaking artifacts is developed. Once a streaking artifact image is estimated, an artifact-free image can be obtained by subtracting the streaking artifacts from the input image. Using extensive experiments with real patient data set, we confirm that the proposed residual learning provides significantly better image reconstruction performance with several orders of magnitude faster computational speed. | The residual learning concept was first introduced by He al @cite_14 for image recognition. In low-level computer vision problems, Kim al @cite_8 employed a residual learning for a super-resolution (SR) method. In these approaches, the residual learning was implemented by a skipped connection corresponding to an identity mapping. Unlike these architectures, Zhang al @cite_4 proposed a direct residual learning architecture for image denoising and super-resolution, which has inspired our method. | {
"cite_N": [
"@cite_14",
"@cite_4",
"@cite_8"
],
"mid": [
"2949650786",
"2508457857",
"2951997238"
],
"abstract": [
"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.",
"The discriminative model learning for image denoising has been recently attracting considerable attentions due to its favorable denoising performance. In this paper, we take one step forward by investigating the construction of feed-forward denoising convolutional neural networks (DnCNNs) to embrace the progress in very deep architecture, learning algorithm, and regularization method into image denoising. Specifically, residual learning and batch normalization are utilized to speed up the training process as well as boost the denoising performance. Different from the existing discriminative denoising models which usually train a specific model for additive white Gaussian noise at a certain noise level, our DnCNN model is able to handle Gaussian denoising with unknown noise level (i.e., blind Gaussian denoising). With the residual learning strategy, DnCNN implicitly removes the latent clean image in the hidden layers. This property motivates us to train a single DnCNN model to tackle with several general image denoising tasks, such as Gaussian denoising, single image super-resolution, and JPEG image deblocking. Our extensive experiments demonstrate that our DnCNN model can not only exhibit high effectiveness in several general image denoising tasks, but also be efficiently implemented by benefiting from GPU computing.",
"We present a highly accurate single-image super-resolution (SR) method. Our method uses a very deep convolutional network inspired by VGG-net used for ImageNet classification simonyan2015very . We find increasing our network depth shows a significant improvement in accuracy. Our final model uses 20 weight layers. By cascading small filters many times in a deep network structure, contextual information over large image regions is exploited in an efficient way. With very deep networks, however, convergence speed becomes a critical issue during training. We propose a simple yet effective training procedure. We learn residuals only and use extremely high learning rates ( @math times higher than SRCNN dong2015image ) enabled by adjustable gradient clipping. Our proposed method performs better than existing methods in accuracy and visual improvements in our results are easily noticeable."
]
} |
1611.06391 | 2556016755 | Recently, compressed sensing (CS) computed tomography (CT) using sparse projection views has been extensively investigated to reduce the potential risk of radiation to patient. However, due to the insufficient number of projection views, an analytic reconstruction approach results in severe streaking artifacts and CS-based iterative approach is computationally very expensive. To address this issue, here we propose a novel deep residual learning approach for sparse view CT reconstruction. Specifically, based on a novel persistent homology analysis showing that the manifold of streaking artifacts is topologically simpler than original ones, a deep residual learning architecture that estimates the streaking artifacts is developed. Once a streaking artifact image is estimated, an artifact-free image can be obtained by subtracting the streaking artifacts from the input image. Using extensive experiments with real patient data set, we confirm that the proposed residual learning provides significantly better image reconstruction performance with several orders of magnitude faster computational speed. | The proposed architecture in Fig. is originated from U-Net developed by Ronneberger al @cite_15 for image segmentation. This architecture was motivated from another deconvolution network for image segmentation by Noh al @cite_6 by adding contracting path and pooling unpooling layers. However, we are not aware of any prior work that employed this architecture beyond the image segmentation. | {
"cite_N": [
"@cite_15",
"@cite_6"
],
"mid": [
"2952232639",
"2952637581"
],
"abstract": [
"There is large consent that successful training of deep networks requires many thousand annotated training samples. In this paper, we present a network and training strategy that relies on the strong use of data augmentation to use the available annotated samples more efficiently. The architecture consists of a contracting path to capture context and a symmetric expanding path that enables precise localization. We show that such a network can be trained end-to-end from very few images and outperforms the prior best method (a sliding-window convolutional network) on the ISBI challenge for segmentation of neuronal structures in electron microscopic stacks. Using the same network trained on transmitted light microscopy images (phase contrast and DIC) we won the ISBI cell tracking challenge 2015 in these categories by a large margin. Moreover, the network is fast. Segmentation of a 512x512 image takes less than a second on a recent GPU. The full implementation (based on Caffe) and the trained networks are available at this http URL .",
"We propose a novel semantic segmentation algorithm by learning a deconvolution network. We learn the network on top of the convolutional layers adopted from VGG 16-layer net. The deconvolution network is composed of deconvolution and unpooling layers, which identify pixel-wise class labels and predict segmentation masks. We apply the trained network to each proposal in an input image, and construct the final semantic segmentation map by combining the results from all proposals in a simple manner. The proposed algorithm mitigates the limitations of the existing methods based on fully convolutional networks by integrating deep deconvolution network and proposal-wise prediction; our segmentation method typically identifies detailed structures and handles objects in multiple scales naturally. Our network demonstrates outstanding performance in PASCAL VOC 2012 dataset, and we achieve the best accuracy (72.5 ) among the methods trained with no external data through ensemble with the fully convolutional network."
]
} |
1611.05838 | 2553159917 | It is well known that an @math Wishart matrix with @math degrees of freedom is close to the appropriately centered and scaled Gaussian Orthogonal Ensemble (GOE) if @math is large enough. Recent work of Bubeck, Ding, Eldan, and Racz, and independently Jiang and Li, shows that the transition happens when @math . Here we consider this critical window and explicitly compute the total variation distance between the Wishart and GOE matrices when @math . This shows, in particular, that the phase transition from Wishart to GOE is smooth. | Theorems and state that as @math , statistics of the Wishart @math and the GOE @math have asymptotically the same distribution, but this is not the case if @math remains bounded. In the random matrix literature there has been lots of work showing that statistics of these ensembles have asymptotically the same distribution even when @math . For instance, when @math , then the limiting empirical spectral distribution of the Wishart is the Marchenko-Pastur law, which shows the difference between the Wishart and GOE, but the largest eigenvalue of the Wishart already behaves like that of the GOE @cite_2 @cite_8 @cite_6 . This naturally raises the question of whether there are other regimes of @math and @math where there are interesting phase transitions. | {
"cite_N": [
"@cite_8",
"@cite_6",
"@cite_2"
],
"mid": [
"",
"2079985400",
"1520752838"
],
"abstract": [
"",
"We consider the asymptotic fluctuation behavior of the largest eigenvalue of certain sample covariance matrices in the asymptotic regime where both dimensions of the corresponding data matrix go to infinity. More precisely, let X be an n x p matrix, and let its rows be i.i.d. complex normal vectors with mean 0 and covariance Σ p . We show that for a large class of covariance matrices £ p, the largest eigenvalue of X*X is asymptotically distributed (after recentering and rescaling) as the Tracy-Widom distribution that appears in the study of the Gaussian unitary ensemble. We give explicit formulas for the centering and scaling sequences that are easy to implement and involve only the spectral distribution of the population covariance, n and p. The main theorem applies to a number of covariance models found in applications. For example, well-behaved Toeplitz matrices as well as covariance matrices whose spectral distribution is a sum of atoms (under some conditions on the mass of the atoms) are among the models the theorem can handle. Generalizations of the theorem to certain spiked versions of our models and a.s. results about the largest eigenvalue are given. We also discuss a simple corollary that does not require normality of the entries of the data matrix and some consequences for applications in multivariate statistics.",
"Let x (1) denote the square of the largest singular value of an n x p matrix X, all of whose entries are independent standard Gaussian variates. Equivalently, x (1) is the largest principal component variance of the covariance matrix X'X, or the largest eigenvalue of a p-variate Wishart distribution on n degrees of freedom with identity covariance. Consider the limit of large p and n with n p = y ≥ 1. When centered by μ p = (√n-1 + √p) 2 and scaled by σ p = (√n-1 + √p)(1 √n-1 + 1 √p) 1 3 , the distribution of x (1) approaches the Tracy-Widom law of order 1, which is defined in terms of the Painleve II differential equation and can be numerically evaluated and tabulated in software. Simulations show the approximation to be informative for n and p as small as 5. The limit is derived via a corresponding result for complex Wishart matrices using methods from random matrix theory. The result suggests that some aspects of large p multivariate distribution theory may be easier to apply in practice than their fixed p counterparts."
]
} |
1611.05607 | 2949452023 | We show that the matching problem that underlies optical flow requires multiple strategies, depending on the amount of image motion and other factors. We then study the implications of this observation on training a deep neural network for representing image patches in the context of descriptor based optical flow. We propose a metric learning method, which selects suitable negative samples based on the nature of the true match. This type of training produces a network that displays multiple strategies depending on the input and leads to state of the art results on the KITTI 2012 and KITTI 2015 optical flow benchmarks. | The need for multiple strategies was found in several vision problems where the basic trained model could not optimize the solution for all sub-categories. An example is the work of Antipov al @cite_30 for age estimation. Unsatisfied by the accuracy of the model for children of age 0-12, they train a sub-model only for those ages and employ it to samples that are classified as this category by another model that is run first. | {
"cite_N": [
"@cite_30"
],
"mid": [
"2485314986"
],
"abstract": [
"This work describes our solution in the second edition of the ChaLearn LAP competition on Apparent Age Estimation. Starting from a pretrained version of the VGG-16 convolutional neural network for face recognition, we train it on the huge IMDB-Wiki dataset for biological age estimation and then fine-tune it for apparent age estimation using the relatively small competition dataset. We show that the precise age estimation of children is the cornerstone of the competition. Therefore, we integrate a separate \"children\" VGG-16 network for apparent age estimation of children between 0 and 12 years old in our final solution. The \"children\" network is fine-tuned from the \"general\" one. We employ different age encoding strategies for training \"general\" and \"children\" networks: the soft one (label distribution encoding) for the \"general\" network and the strict one (0 1 classification encoding) for the \"children\" network. Finally, we highlight the importance of the state-of-the-art face detection and face alignment for the final apparent age estimation. Our resulting solution wins the 1st place in the competition significantly outperforming the runner-up."
]
} |
1611.05607 | 2949452023 | We show that the matching problem that underlies optical flow requires multiple strategies, depending on the amount of image motion and other factors. We then study the implications of this observation on training a deep neural network for representing image patches in the context of descriptor based optical flow. We propose a metric learning method, which selects suitable negative samples based on the nature of the true match. This type of training produces a network that displays multiple strategies depending on the input and leads to state of the art results on the KITTI 2012 and KITTI 2015 optical flow benchmarks. | It was recently proposed to eliminate from the training process samples that are either too easy or too hard @cite_16 . For this purpose, specific percentiles on the loss were employed. Samples which did not meet the loss criteria were put aside for a predefined number of epochs. | {
"cite_N": [
"@cite_16"
],
"mid": [
"2520826941"
],
"abstract": [
"In this paper, we address the task of object counting in images. We follow modern learning approaches in which a density map is estimated directly from the input image. We employ CNNs and incorporate two significant improvements to the state of the art methods: layered boosting and selective sampling. As a result, we manage both to increase the counting accuracy and to reduce processing time. Moreover, we show that the proposed method is effective, even in the presence of labeling errors. Extensive experiments on five different datasets demonstrate the efficacy and robustness of our approach. Mean Absolute error was reduced by 20 to 35 . At the same time, the training time of each CNN has been reduced by 50 ."
]
} |
1611.05896 | 2556088399 | In this work, we explore a genre of puzzles ("image riddles") which involves a set of images and a question. Answering these puzzles require both capabilities involving visual detection (including object, activity recognition) and, knowledge-based or commonsense reasoning. We compile a dataset of over 3k riddles where each riddle consists of 4 images and a groundtruth answer. The annotations are validated using crowd-sourced evaluation. We also define an automatic evaluation metric to track future progress. Our task bears similarity with the commonly known IQ tasks such as analogy solving, sequence filling that are often used to test intelligence. We develop a Probabilistic Reasoning-based approach that utilizes probabilistic commonsense knowledge to answer these riddles with a reasonable accuracy. We demonstrate the results of our approach using both automatic and human evaluations. Our approach achieves some promising results for these riddles and provides a strong baseline for future attempts. We make the entire dataset and related materials publicly available to the community in ImageRiddle Website (this http URL). | Our work is also related to the field of . Very recently, researchers spent a significant amount of efforts on both creating datasets and proposing new models @cite_23 @cite_2 @cite_6 @cite_19 . Interestingly both @cite_23 and @cite_6 adapted MS-COCO @cite_9 images and created an open domain dataset with human generated questions and answers. Both @cite_2 and @cite_6 use recurrent networks to encode the sentence and output the answer. Even though some questions from @cite_23 and @cite_6 are very challenging which actually require logical reasoning in order to answer correctly, popular approaches are still hoping to learn the direct signal-to-signal mapping from image and question to its answer, given a large enough annotated data. The necessity of common-sense reasoning could be easily neglected. Here we introduce the new Image Riddle problem which is 1) a well-defined cognitively challenging task that requires both vision and reasoning capability, 2) it is impossible to model the problem as direct signal-to-signal mapping, due to the data sparsity and 3) system's performance could still be bench-marked automatically for comparison. All these qualities make our Image Riddle dataset a good testbed for vision and reasoning research. | {
"cite_N": [
"@cite_9",
"@cite_6",
"@cite_19",
"@cite_23",
"@cite_2"
],
"mid": [
"2952122856",
"1488163396",
"1606748815",
"2950761309",
"2952246170"
],
"abstract": [
"We present a new dataset with the goal of advancing the state-of-the-art in object recognition by placing the question of object recognition in the context of the broader question of scene understanding. This is achieved by gathering images of complex everyday scenes containing common objects in their natural context. Objects are labeled using per-instance segmentations to aid in precise object localization. Our dataset contains photos of 91 objects types that would be easily recognizable by a 4 year old. With a total of 2.5 million labeled instances in 328k images, the creation of our dataset drew upon extensive crowd worker involvement via novel user interfaces for category detection, instance spotting and instance segmentation. We present a detailed statistical analysis of the dataset in comparison to PASCAL, ImageNet, and SUN. Finally, we provide baseline performance analysis for bounding box and segmentation detection results using a Deformable Parts Model.",
"In this paper, we present the mQA model, which is able to answer questions about the content of an image. The answer can be a sentence, a phrase or a single word. Our model contains four components: a Long Short-Term Memory (LSTM) to extract the question representation, a Convolutional Neural Network (CNN) to extract the visual representation, an LSTM for storing the linguistic context in an answer, and a fusing component to combine the information from the first three components and generate the answer. We construct a Freestyle Multilingual Image Question Answering (FM-IQA) dataset to train and evaluate our mQA model. It contains over 150,000 images and 310,000 freestyle Chinese question-answer pairs and their English translations. The quality of the generated answers of our mQA model on this dataset is evaluated by human judges through a Turing Test. Specifically, we mix the answers provided by humans and our model. The human judges need to distinguish our model from the human. They will also provide a score (i.e. 0, 1, 2, the larger the better) indicating the quality of the answer. We propose strategies to monitor the quality of this evaluation process. The experiments show that in 64.7 of cases, the human judges cannot distinguish our model from humans. The average score is 1.454 (1.918 for human). The details of this work, including the FM-IQA dataset, can be found on the project page: this http URL",
"In this paper, we propose to employ the convolutional neural network (CNN) for the image question answering (QA). Our proposed CNN provides an end-to-end framework with convolutional architectures for learning not only the image and question representations, but also their inter-modal interactions to produce the answer. More specifically, our model consists of three CNNs: one image CNN to encode the image content, one sentence CNN to compose the words of the question, and one multimodal convolution layer to learn their joint representation for the classification in the space of candidate answer words. We demonstrate the efficacy of our proposed model on the DAQUAR and COCO-QA datasets, which are two benchmark datasets for the image QA, with the performances significantly outperforming the state-of-the-art.",
"We propose the task of free-form and open-ended Visual Question Answering (VQA). Given an image and a natural language question about the image, the task is to provide an accurate natural language answer. Mirroring real-world scenarios, such as helping the visually impaired, both the questions and answers are open-ended. Visual questions selectively target different areas of an image, including background details and underlying context. As a result, a system that succeeds at VQA typically needs a more detailed understanding of the image and complex reasoning than a system producing generic image captions. Moreover, VQA is amenable to automatic evaluation, since many open-ended answers contain only a few words or a closed set of answers that can be provided in a multiple-choice format. We provide a dataset containing 0.25M images, 0.76M questions, and 10M answers (www.visualqa.org), and discuss the information it provides. Numerous baselines and methods for VQA are provided and compared with human performance. Our VQA demo is available on CloudCV (this http URL).",
"We address a question answering task on real-world images that is set up as a Visual Turing Test. By combining latest advances in image representation and natural language processing, we propose Neural-Image-QA, an end-to-end formulation to this problem for which all parts are trained jointly. In contrast to previous efforts, we are facing a multi-modal problem where the language output (answer) is conditioned on visual and natural language input (image and question). Our approach Neural-Image-QA doubles the performance of the previous best approach on this problem. We provide additional insights into the problem by analyzing how much information is contained only in the language part for which we provide a new human baseline. To study human consensus, which is related to the ambiguities inherent in this challenging task, we propose two novel metrics and collect additional answers which extends the original DAQUAR dataset to DAQUAR-Consensus."
]
} |
1611.05943 | 2790172107 | P4 is an emergent packet-processing language with which the user can describe how the packets are to be processed in a switching element. This paper presents a way to implement complex operations that are not natively supported in P4. In this work, we explored two different methods to add extensions to P4: i) using new native primitives and ii) using extern instances. As a case study, an ROHC entity was implemented and invoked in a P4 program. The tests showed similar relative performance in both methods in terms of normalized packet latency. However, extern instances appear to be more suitable for target-specific switching applications, where the manufacturer vendor can specify its own specific operations without changes in the P4 syntax and semantics. Extern instances only require changes in the target-specific backend compiler while keeping the P4 frontend compiler unchanged. The use of externs also results in a more elegant code solution since they are implemented outside the switch-core, thus reducing side effects risks that can be caused by a modification in a switch pipeline implementation. | In @cite_22 , Bosshart propose the P4 language, a protocol independent packet processing language. Fig. presents the abstract forwarding model on which P4 was built upon. A P4 program consists of a parser state machine (PSM) followed by a set of match-action tables in the ingress and egress pipelines. The processing flow is controlled by an imperative control program. | {
"cite_N": [
"@cite_22"
],
"mid": [
"1994926493"
],
"abstract": [
"P4 is a high-level language for programming protocol-independent packet processors. P4 works in conjunction with SDN control protocols like OpenFlow. In its current form, OpenFlow explicitly specifies protocol headers on which it operates. This set has grown from 12 to 41 fields in a few years, increasing the complexity of the specification while still not providing the flexibility to add new headers. In this paper we propose P4 as a strawman proposal for how OpenFlow should evolve in the future. We have three goals: (1) Reconfigurability in the field: Programmers should be able to change the way switches process packets once they are deployed. (2) Protocol independence: Switches should not be tied to any specific network protocols. (3) Target independence: Programmers should be able to describe packet-processing functionality independently of the specifics of the underlying hardware. As an example, we describe how to use P4 to configure a switch to add a new hierarchical label."
]
} |
1611.05943 | 2790172107 | P4 is an emergent packet-processing language with which the user can describe how the packets are to be processed in a switching element. This paper presents a way to implement complex operations that are not natively supported in P4. In this work, we explored two different methods to add extensions to P4: i) using new native primitives and ii) using extern instances. As a case study, an ROHC entity was implemented and invoked in a P4 program. The tests showed similar relative performance in both methods in terms of normalized packet latency. However, extern instances appear to be more suitable for target-specific switching applications, where the manufacturer vendor can specify its own specific operations without changes in the P4 syntax and semantics. Extern instances only require changes in the target-specific backend compiler while keeping the P4 frontend compiler unchanged. The use of externs also results in a more elegant code solution since they are implemented outside the switch-core, thus reducing side effects risks that can be caused by a modification in a switch pipeline implementation. | Due to its simplicity, portability, and device agnosticism, P4 has gained popularity as a packet processing language for programmable forwarding elements in both academia and industry @cite_10 @cite_9 @cite_3 @cite_1 @cite_17 . | {
"cite_N": [
"@cite_9",
"@cite_1",
"@cite_3",
"@cite_10",
"@cite_17"
],
"mid": [
"2301990415",
"1608999459",
"2491015902",
"2297371674",
"2607067159"
],
"abstract": [
"The P4 language is an emerging domain-specific language for describing the data plane processing at a network device. P4 has been mapped to a wide range of forwarding devices including NPUs, programmable NICs and FPGAs, except for General Purpose Graphics Processing Unit (GPGPU) which is a salient parallel architecture for processing network flows. In this work, we design a heterogeneous architecture with both CPU and GPU as a P4 programming target, and present a toolset to map a P4 program onto the proposed architecture. Our evaluation reveals that a P4 program can render promising performance on such architecture by parallelizing its \"match+action\" engine with the GPGPU accelerator. The experiment results show that the auto-configured GPU kernels achieve scalable lookup and classification speeds: the prototype system can reach up to 580 Gbps for IP lookups (64-byte packets) and 60 million classifications per second for 4k firewall rules, respectively.",
"We describe the design and implementation of Open vSwitch, a multi-layer, open source virtual switch for all major hypervisor platforms. Open vSwitch was designed de novo for networking in virtual environments, resulting in major design departures from traditional software switching architectures. We detail the advanced flow classification and caching techniques that Open vSwitch uses to optimize its operations and conserve hypervisor resources. We evaluate Open vSwitch performance, drawing from our deployment experiences over the past seven years of using and improving Open vSwitch.",
"Hypervisors use software switches to steer packets to and from virtual machines (VMs). These switches frequently need upgrading and customization—to support new protocol headers or encapsulations for tunneling and overlays, to improve measurement and debugging features, and even to add middlebox-like functions. Software switches are typically based on a large body of code, including kernel code, and changing the switch is a formidable undertaking requiring domain mastery of network protocol design and developing, testing, and maintaining a large, complex codebase. Changing how a software switch forwards packets should not require intimate knowledge of its implementation. Instead, it should be possible to specify how packets are processed and forwarded in a high-level domain-specific language (DSL) such as P4, and compiled to run on a software switch. We present PISCES, a software switch derived from Open vSwitch (OVS), a hard-wired hypervisor switch, whose behavior is customized using P4. PISCES is not hard-wired to specific protocols; this independence makes it easy to add new features. We also show how the compiler can analyze the high-level specification to optimize forwarding performance. Our evaluation shows that PISCES performs comparably to OVS and that PISCES programs are about 40 times shorter than equivalent changes to OVS source code.",
"In this paper, we introduce PFPSim, a host-compiled simulator for early validation and analysis of packet processing applications on programmable forwarding plane architectures. The simulation model is automatically generated from a high-level description of the hardware software architecture of the forwarding device and the behavioral description of the various modules in the architecture. Our high-level architectural description language is capable of defining many-core network processors as well as reconfigurable pipelines. The behavior of the fixed-function processing elements in the architecture is defined in C++. The code targeted for the processor cores, or reconfigurable pipeline stages, is compiled from P4, an emerging programming language for packet processing applications. Application developers can use PFPSim as a virtual prototype to simulate and debug their applications before hardware availability. Moreover, forwarding device architects can use PFPSim to evaluate the trade-offs between different hardware software design decisions.",
"This paper presents P4FPGA, a new tool for developing and evaluating data plane applications. P4FPGA is an open-source compiler and runtime. The compiler extends the P4.org reference compiler with a custom backend that generates FPGA code. P4FPGA supports different architecture configurations, depending on the needs of the particular application. We have benchmarked several representative P4 programs, and our experiments show that code generated by P4FPGA runs at line-rate at all packet sizes with latencies comparable to commercial ASICs. By combining high-level programming abstractions offered by P4 with a flexible and powerful hardware target, P4FPGA allows developers to rapidly prototype and deploy new data plane applications."
]
} |
1611.05774 | 2950191064 | Recurrent neural network grammars (RNNG) are a recently proposed probabilistic generative modeling family for natural language. They show state-of-the-art language modeling and parsing performance. We investigate what information they learn, from a linguistic perspective, through various ablations to the model and the data, and by augmenting the model with an attention mechanism (GA-RNNG) to enable closer inspection. We find that explicit modeling of composition is crucial for achieving the best performance. Through the attention mechanism, we find that headedness plays a central role in phrasal representation (with the model's latent attention largely agreeing with predictions made by hand-crafted head rules, albeit with some important differences). By training grammars without nonterminal labels, we find that phrasal representations depend minimally on nonterminals, providing support for the endocentricity hypothesis. | The problem of understanding neural network models in NLP has been previously studied for sequential RNNs @cite_11 @cite_16 . showed that sequence-to-sequence neural translation models capture a certain degree of syntactic knowledge of the source language, such as voice (active or passive) and tense information, as a by-product of the translation objective. Our experiment on the importance of composition function was motivated by and , who achieved competitive parsing accuracy without explicit composition. In another work, investigated the importance of recursive tree structures (as opposed to linear recurrent models) in four different tasks, including sentiment and semantic relation classification. Their findings suggest that recursive tree structures are beneficial for tasks that require identifying long-range relations, such as semantic relationship classification, with no conclusive advantage for sentiment classification and discourse parsing. Through the stack-only ablation we demonstrate that the RNNG composition function is crucial to obtaining state-of-the-art parsing performance. Extensive prior work on phrase-structure parsing typically employs the probabilistic context-free grammar formalism, with lexicalized @cite_2 and nonterminal @cite_5 @cite_19 augmentations. The conjecture that fine-grained nonterminal rules and labels can be discovered given weaker bracketing structures was based on several studies @cite_18 @cite_6 @cite_12 . | {
"cite_N": [
"@cite_18",
"@cite_6",
"@cite_19",
"@cite_2",
"@cite_5",
"@cite_16",
"@cite_12",
"@cite_11"
],
"mid": [
"2079326125",
"2129882630",
"2097606805",
"2950064328",
"1551104980",
"1601924930",
"2139621418",
"1951216520"
],
"abstract": [
"Many recent statistical parsers rely on a preprocessing step which uses hand-written, corpus-specific rules to augment the training data with extra information. For example, head-finding rules are used to augment node labels with lexical heads. In this paper, we provide machinery to reduce the amount of human effort needed to adapt existing models to new corpora: first, we propose a flexible notation for specifying these rules that would allow them to be shared by different models; second, we report on an experiment to see whether we can use Expectation-Maximization to automatically fine-tune a set of hand-written rules to a particular corpus.",
"We present a generative distributional model for the unsupervised induction of natural language syntax which explicitly models constituent yields and contexts. Parameter search with EM produces higher quality analyses than previously exhibited by unsupervised systems, giving the best published un-supervised parsing results on the ATIS corpus. Experiments on Penn treebank sentences of comparable length show an even higher F1 of 71 on non-trivial brackets. We compare distributionally induced and actual part-of-speech tags as input data, and examine extensions to the basic model. We discuss errors made by the system, compare the system to previous models, and discuss upper bounds, lower bounds, and stability for this task.",
"We demonstrate that an unlexicalized PCFG can parse much more accurately than previously shown, by making use of simple, linguistically motivated state splits, which break down false independence assumptions latent in a vanilla treebank grammar. Indeed, its performance of 86.36 (LP LR F1) is better than that of early lexicalized PCFG models, and surprisingly close to the current state-of-the-art. This result has potential uses beyond establishing a strong lower bound on the maximum possible accuracy of unlexicalized models: an unlexicalized PCFG is much more compact, easier to replicate, and easier to interpret than more complex lexical models, and the parsing algorithms are simpler, more widely understood, of lower asymptotic complexity, and easier to optimize.",
"In this paper we first propose a new statistical parsing model, which is a generative model of lexicalised context-free grammar. We then extend the model to include a probabilistic treatment of both subcategorisation and wh-movement. Results on Wall Street Journal text show that the parser performs at 88.1 87.5 constituent precision recall, an average improvement of 2.3 over (Collins 96).",
"The kinds of tree representations used in a treebank corpus can have a dramatic effect on performance of a parser based on the PCFG estimated from that corpus, causing the estimated likelihood of a tree to differ substantially from its frequency in the training corpus. This paper points out that the Penn II treebank representations are of the kind predicted to have such an effect, and describes a simple node relabeling transformation that improves a treebank PCFG-based parser's average precision and recall by around 8 , or approximately half of the performance difference between a simple PCFG model and the best broad-coverage parsers available today. This performance variation comes about because any PCFG, and hence the corpus of trees from which the PCFG is induced, embodies independence assumptions about the distribution of words and phrases. The particular independence assumptions implicit in a tree representation can be studied theoretically and investigated empirically by means of a tree transformation detransformation process.",
"While neural networks have been successfully applied to many NLP tasks the resulting vector-based models are very difficult to interpret. For example it's not clear how they achieve compositionality , building sentence meaning from the meanings of words and phrases. In this paper we describe four strategies for visualizing compositionality in neural models for NLP, inspired by similar work in computer vision. We first plot unit values to visualize compositionality of negation, intensification, and concessive clauses, allow us to see well-known markedness asymmetries in negation. We then introduce three simple and straightforward methods for visualizing a unit's salience , the amount it contributes to the final composed meaning: (1) gradient back-propagation, (2) the variance of a token from the average word node, (3) LSTM-style gates that measure information flow. We test our methods on sentiment using simple recurrent nets and LSTMs. Our general-purpose methods may have wide applications for understanding compositionality and other semantic properties of deep networks , and also shed light on why LSTMs outperform simple recurrent nets,",
"We present an automatic approach to tree annotation in which basic nonterminal symbols are alternately split and merged to maximize the likelihood of a training treebank. Starting with a simple X-bar grammar, we learn a new grammar whose nonterminals are subsymbols of the original nonterminals. In contrast with previous work, we are able to split various terminals to different degrees, as appropriate to the actual complexity in the data. Our grammars automatically learn the kinds of linguistic distinctions exhibited in previous work on manual tree annotation. On the other hand, our grammars are much more compact and substantially more accurate than previous work on automatic annotation. Despite its simplicity, our best grammar achieves an F1 of 90.2 on the Penn Treebank, higher than fully lexicalized systems.",
"Recurrent Neural Networks (RNNs), and specifically a variant with Long Short-Term Memory (LSTM), are enjoying renewed interest as a result of successful applications in a wide range of machine learning problems that involve sequential data. However, while LSTMs provide exceptional results in practice, the source of their performance and their limitations remain rather poorly understood. Using character-level language models as an interpretable testbed, we aim to bridge this gap by providing an analysis of their representations, predictions and error types. In particular, our experiments reveal the existence of interpretable cells that keep track of long-range dependencies such as line lengths, quotes and brackets. Moreover, our comparative analysis with finite horizon n-gram models traces the source of the LSTM improvements to long-range structural dependencies. Finally, we provide analysis of the remaining errors and suggests areas for further study."
]
} |
1611.05603 | 2950413999 | State-of-the-art methods treat pedestrian attribute recognition as a multi-label image classification problem. The location information of person attributes is usually eliminated or simply encoded in the rigid splitting of whole body in previous work. In this paper, we formulate the task in a weakly-supervised attribute localization framework. Based on GoogLeNet, firstly, a set of mid-level attribute features are discovered by novelly designed detection layers, where a max-pooling based weakly-supervised object detection technique is used to train these layers with only image-level labels without the need of bounding box annotations of pedestrian attributes. Secondly, attribute labels are predicted by regression of the detection response magnitudes. Finally, the locations and rough shapes of pedestrian attributes can be inferred by performing clustering on a fusion of activation maps of the detection layers, where the fusion weights are estimated as the correlation strengths between each attribute and its relevant mid-level features. Extensive experiments are performed on the two currently largest pedestrian attribute datasets, i.e. the PETA dataset and the RAP dataset. Results show that the proposed method has achieved competitive performance on attribute recognition, compared to other state-of-the-art methods. Moreover, the results of attribute localization are visualized to understand the characteristics of the proposed method. | It is yet another a popular idea to make use of part information to help improving the attribute recognition accuracy. In @cite_13 , part models like DPM and poselets are used for aligning input patches for CNNs. propose an expanded parts model in @cite_17 to learn a collection of part templates which can score an image partially with most discriminative regions for classification. The MLCNN in @cite_0 divides a human body into 15 parts and train CNN models for each of them, then choose part of the models to contribute to the recognition of an attribute, according to the spatial constraint prior of it. The DeepMAR* model described in @cite_21 takes three block images as input in addition to the whole body image, which correspond to the head-shoulder part, upper body and lower body of a pedestrian respectively. The idea of dividing the image into parts is adopted in the design of the WPAL-network, which drives us to make use of flexible spatial pyramid pooling layers to help locating mid-level features of some attributes in only local patches rather than the whole image. | {
"cite_N": [
"@cite_0",
"@cite_21",
"@cite_13",
"@cite_17"
],
"mid": [
"1522973599",
"2308869522",
"",
"2117103983"
],
"abstract": [
"Recently, pedestrian attributes like gender, age and clothing etc., have been used as soft biometric traits for recognizing people. Unlike existing methods that assume the independence of attributes during their prediction, we propose a multi-label convolutional neural network (MLCNN) to predict multiple attributes together in a unified framework. Firstly, a pedestrian image is roughly divided into multiple overlapping body parts, which are simultaneously integrated in the multi-label convolutional neural network. Secondly, these parts are filtered independently and aggregated in the cost layer. The cost function is a combination of multiple binary attribute classification cost functions. Moreover, we propose an attribute assisted person re-identification method, which fuses attribute distances and low-level feature distances between pairs of person images to improve person re-identification performance. Extensive experiments show: 1) the average attribute classification accuracy of the proposed method is 5.2 and 9.3 higher than the SVM-based method on three public databases, VIPeR and GRID, respectively; 2) the proposed attribute assisted person re-identification method is superior to existing approaches.",
"In this paper, we aim to improve the dataset foundation for pedestrian attribute recognition in real surveillance scenarios. Recognition of human attributes, such as gender, and clothes types, has great prospects in real applications. However, the development of suitable benchmark datasets for attribute recognition remains lagged behind. Existing human attribute datasets are collected from various sources or an integration of pedestrian re-identification datasets. Such heterogeneous collection poses a big challenge on developing high quality fine-grained attribute recognition algorithms. Furthermore, human attribute recognition are generally severely affected by environmental or contextual factors, such as viewpoints, occlusions and body parts, while existing attribute datasets barely care about them. To tackle these problems, we build a Richly Annotated Pedestrian (RAP) dataset from real multi-camera surveillance scenarios with long term collection, where data samples are annotated with not only fine-grained human attributes but also environmental and contextual factors. RAP has in total 41,585 pedestrian samples, each of which is annotated with 72 attributes as well as viewpoints, occlusions, body parts information. To our knowledge, the RAP dataset is the largest pedestrian attribute dataset, which is expected to greatly promote the study of large-scale attribute recognition systems. Furthermore, we empirically analyze the effects of different environmental and contextual factors on pedestrian attribute recognition. Experimental results demonstrate that viewpoints, occlusions and body parts information could assist attribute recognition a lot in real applications.",
"",
"We propose a new model for recognizing human attributes (e.g. wearing a suit, sitting, short hair) and actions (e.g. running, riding a horse) in still images. The proposed model relies on a collection of part templates which are learnt discriminatively to explain specific scale-space locations in the images (in human centric coordinates). It avoids the limitations of highly structured models, which consist of a few (i.e. a mixture of) 'average' templates. To learn our model, we propose an algorithm which automatically mines out parts and learns corresponding discriminative templates with their respective locations from a large number of candidate parts. We validate the method on recent challenging datasets: (i) Willow 7 actions [7], (ii) 27 Human Attributes (HAT) [25], and (iii) Stanford 40 actions [37]. We obtain convincing qualitative and state-of-the-art quantitative results on the three datasets."
]
} |
1611.05603 | 2950413999 | State-of-the-art methods treat pedestrian attribute recognition as a multi-label image classification problem. The location information of person attributes is usually eliminated or simply encoded in the rigid splitting of whole body in previous work. In this paper, we formulate the task in a weakly-supervised attribute localization framework. Based on GoogLeNet, firstly, a set of mid-level attribute features are discovered by novelly designed detection layers, where a max-pooling based weakly-supervised object detection technique is used to train these layers with only image-level labels without the need of bounding box annotations of pedestrian attributes. Secondly, attribute labels are predicted by regression of the detection response magnitudes. Finally, the locations and rough shapes of pedestrian attributes can be inferred by performing clustering on a fusion of activation maps of the detection layers, where the fusion weights are estimated as the correlation strengths between each attribute and its relevant mid-level features. Extensive experiments are performed on the two currently largest pedestrian attribute datasets, i.e. the PETA dataset and the RAP dataset. Results show that the proposed method has achieved competitive performance on attribute recognition, compared to other state-of-the-art methods. Moreover, the results of attribute localization are visualized to understand the characteristics of the proposed method. | To avoid the high-cost of labeling bounding boxes of objects, researchers proposed various weakly-supervised learning approaches for object detection and localization. In @cite_20 , demonstrate capability of SVM and deformable part models on weakly-supervised object detection. In @cite_7 , proposed unsupervised latent category learning, which can discover latent information in backgrounds to help object localization in cluttered backgrounds. proposed in @cite_3 a multi-fold multiple-instance learning procedure featuring prevention of weakly-supervised training from prematurely locking onto erroneous object locations. | {
"cite_N": [
"@cite_3",
"@cite_7",
"@cite_20"
],
"mid": [
"2016016818",
"318792885",
"2099528205"
],
"abstract": [
"Object category localization is a challenging problem in computer vision. Standard supervised training requires bounding box annotations of object instances. This time-consuming annotation process is sidestepped in weakly supervised learning. In this case, the supervised information is restricted to binary labels that indicate the absence presence of object instances in the image, without their locations. We follow a multiple-instance learning approach that iteratively trains the detector and infers the object locations in the positive training images. Our main contribution is a multi-fold multiple instance learning procedure, which prevents training from prematurely locking onto erroneous object locations. This procedure is particularly important when high-dimensional representations, such as the Fisher vectors, are used. We present a detailed experimental evaluation using the PASCAL VOC 2007 dataset. Compared to state-of-the-art weakly supervised detectors, our approach better localizes objects in the training images, which translates into improved detection performance.",
"Localizing objects in cluttered backgrounds is a challenging task in weakly supervised localization. Due to large object variations in cluttered images, objects have large ambiguity with backgrounds. However, backgrounds contain useful latent information, e.g., the sky for aeroplanes. If we can learn this latent information, object-background ambiguity can be reduced to suppress the background. In this paper, we propose the latent category learning (LCL), which is an unsupervised learning problem given only image-level class labels. Firstly, inspired by the latent semantic discovery, we use the typical probabilistic Latent Semantic Analysis (pLSA) to learn the latent categories, which can represent objects, object parts or backgrounds. Secondly, to determine which category contains the target object, we propose a category selection method evaluating each category’s discrimination. We evaluate the method on the PASCAL VOC 2007 database and ILSVRC 2013 detection challenge. On VOC 2007, the proposed method yields the annotation accuracy of 48 , which outperforms previous results by 10 . More importantly, we achieve the detection average precision of 30.9 , which improves previous results by 8 and can be competitive with the supervised deformable part model (DPM) 5.0 baseline 33.7 . On ILSVRC 2013 detection, the method yields the precision of 6.0 , which is also competitive with the DPM 5.0.",
"Weakly supervised discovery of common visual structure in highly variable, cluttered images is a key problem in recognition. We address this problem using deformable part-based models (DPM's) with latent SVM training [6]. These models have been introduced for fully supervised training of object detectors, but we demonstrate that they are also capable of more open-ended learning of latent structure for such tasks as scene recognition and weakly supervised object localization. For scene recognition, DPM's can capture recurring visual elements and salient objects; in combination with standard global image features, they obtain state-of-the-art results on the MIT 67-category indoor scene dataset. For weakly supervised object localization, optimization over latent DPM parameters can discover the spatial extent of objects in cluttered training images without ground-truth bounding boxes. The resulting method outperforms a recent state-of-the-art weakly supervised object localization approach on the PASCAL-07 dataset."
]
} |
1611.05603 | 2950413999 | State-of-the-art methods treat pedestrian attribute recognition as a multi-label image classification problem. The location information of person attributes is usually eliminated or simply encoded in the rigid splitting of whole body in previous work. In this paper, we formulate the task in a weakly-supervised attribute localization framework. Based on GoogLeNet, firstly, a set of mid-level attribute features are discovered by novelly designed detection layers, where a max-pooling based weakly-supervised object detection technique is used to train these layers with only image-level labels without the need of bounding box annotations of pedestrian attributes. Secondly, attribute labels are predicted by regression of the detection response magnitudes. Finally, the locations and rough shapes of pedestrian attributes can be inferred by performing clustering on a fusion of activation maps of the detection layers, where the fusion weights are estimated as the correlation strengths between each attribute and its relevant mid-level features. Extensive experiments are performed on the two currently largest pedestrian attribute datasets, i.e. the PETA dataset and the RAP dataset. Results show that the proposed method has achieved competitive performance on attribute recognition, compared to other state-of-the-art methods. Moreover, the results of attribute localization are visualized to understand the characteristics of the proposed method. | In @cite_10 , the proposed network has convolution layers followed by a global max-pooling layer. Each channel of the global max-pooling layer is viewed as a detector for a certain class of object. It is assumed that the positions of max value point in the feature map correspond to the locations where the objects of the target class exist in. However, this method cannot be directly applied to our attribute localization task. Firstly, different from objects, some attributes are abstract concepts, such as gender, orientation and age, which do not correspond to certain regions. Secondly, some attributes such as hat wearing or shoe style are expected to appear within a certain partition in a pedestrian sample, which can be used to improve the localization of those attributes. Thus, to better fit the task of attribute localization, we embed this structure in the middle stage of the network to discover mid-level features relevant to attributes rather than attributes themselves, and propose to use FSPP layers instead of a single global max-pooling layer to help constraining location of certain attributes. | {
"cite_N": [
"@cite_10"
],
"mid": [
"1994488211"
],
"abstract": [
"Successful methods for visual object recognition typically rely on training datasets containing lots of richly annotated images. Detailed image annotation, e.g. by object bounding boxes, however, is both expensive and often subjective. We describe a weakly supervised convolutional neural network (CNN) for object classification that relies only on image-level labels, yet can learn from cluttered scenes containing multiple objects. We quantify its object classification and object location prediction performance on the Pascal VOC 2012 (20 object classes) and the much larger Microsoft COCO (80 object classes) datasets. We find that the network (i) outputs accurate image-level labels, (ii) predicts approximate locations (but not extents) of objects, and (iii) performs comparably to its fully-supervised counterparts using object bounding box annotation for training."
]
} |
1611.05689 | 2952346959 | We present a new deep learning-based approach for dense stereo matching. Compared to previous works, our approach does not use deep learning of pixel appearance descriptors, employing very fast classical matching scores instead. At the same time, our approach uses a deep convolutional network to predict the local parameters of cost volume aggregation process, which in this paper we implement using differentiable domain transform. By treating such transform as a recurrent neural network, we are able to train our whole system that includes cost volume computation, cost-volume aggregation (smoothing), and winner-takes-all disparity selection end-to-end. The resulting method is highly efficient at test time, while achieving good matching accuracy. On the KITTI 2015 benchmark, it achieves a result of 6.34 error rate while running at 29 frames per second rate on a modern GPU. | As discussed above, our work is also related to preceding approaches that use deep learning for stereo. Our approach differs markedly from @cite_13 @cite_24 @cite_26 as we use deep learning within the cost aggregation rather than to compute the matching costs themselves. Unlike @cite_13 @cite_24 @cite_26 and similarly to @cite_20 we also use end-to-end learning that encompasses all stages of depthmap computation within our method. Unlike @cite_20 , which uses a rather generic feed-forward convolutional network trained on a massive amount of synthetic stereo pairs, our method employs classical stereo matching algorithms such as census transform as modules within a more specific architecture that combines convolutional networks with a gated recurrent neural network module (which is equivalent to the domain transform operation). | {
"cite_N": [
"@cite_24",
"@cite_26",
"@cite_13",
"@cite_20"
],
"mid": [
"2963502507",
"2440384215",
"2144041313",
""
],
"abstract": [
"We present a method for extracting depth information from a rectified image pair. Our approach focuses on the first stage of many stereo algorithms: the matching cost computation. We approach the problem by learning a similarity measure on small image patches using a convolutional neural network. Training is carried out in a supervised manner by constructing a binary classification data set with examples of similar and dissimilar pairs of patches. We examine two network architectures for this task: one tuned for speed, the other for accuracy. The output of the convolutional neural network is used to initialize the stereo matching cost. A series of post-processing steps follow: cross-based cost aggregation, semiglobal matching, a left-right consistency check, subpixel enhancement, a median filter, and a bilateral filter. We evaluate our method on the KITTI 2012, KITTI 2015, and Middlebury stereo data sets and show that it outperforms other approaches on all three data sets.",
"In the past year, convolutional neural networks have been shown to perform extremely well for stereo estimation. However, current architectures rely on siamese networks which exploit concatenation followed by further processing layers, requiring a minute of GPU computation per image pair. In contrast, in this paper we propose a matching network which is able to produce very accurate results in less than a second of GPU computation. Towards this goal, we exploit a product layer which simply computes the inner product between the two representations of a siamese architecture. We train our network by treating the problem as multi-class classification, where the classes are all possible disparities. This allows us to get calibrated scores, which result in much better matching performance when compared to existing approaches.",
"We present a method for extracting depth information from a rectified image pair. We train a convolutional neural network to predict how well two image patches match and use it to compute the stereo matching cost. The cost is refined by cross-based cost aggregation and semiglobal matching, followed by a left-right consistency check to eliminate errors in the occluded regions. Our stereo method achieves an error rate of 2.61 on the KITTI stereo dataset and is currently (August 2014) the top performing method on this dataset.",
""
]
} |
1611.05372 | 2951155117 | We study the sensitivity of optimal solutions of convex separable optimization problems over an integral polymatroid base polytope with respect to parameters determining both the cost of each element and the polytope. Under convexity and a regularity assumption on the functional dependency of the cost function with respect to the parameters, we show that reoptimization after a change in parameters can be done by elementary local operations. Applying this result, we derive that starting from any optimal solution there is a new optimal solution to new parameters such that the L1-norm of the difference of the two solutions is at most two times the L1 norm of the difference of the parameters. We apply these sensitivity results to a class of non-cooperative polymatroid games and derive the existence of pure Nash equilibria. We complement our results by showing that polymatroids are the maximal combinatorial structure enabling these results. For any non-polymatroid region, there is a corresponding optimization problem for which the sensitivity results do not hold. In addition, there is a game where the players strategies are isomorphic to the non-polymatroid region and that does not admit a pure Nash equilibrium. | Parts of the results of this paper have been presented by the authors in less general and preliminary form in the Proceedings of the 10th Conference on Web and Internet Economics @cite_35 . | {
"cite_N": [
"@cite_35"
],
"mid": [
"16002891"
],
"abstract": [
"We study competitive resource allocation problems in which players distribute their demands integrally over a set of resources subject to player-specific submodular capacity constraints. Each player has to pay for each unit of demand a cost that is a non-decreasing and convex function of the total allocation of that resource. This general model of resource allocation generalizes both singleton congestion games with integer-splittable demands and matroid congestion games with player-specific costs. As our main result, we show that in such general resource allocation problems a pure Nash equilibrium is guaranteed to exist by giving a pseudo-polynomial algorithm computing a pure Nash equilibrium."
]
} |
1611.05358 | 2952746495 | The goal of this work is to recognise phrases and sentences being spoken by a talking face, with or without the audio. Unlike previous works that have focussed on recognising a limited number of words or phrases, we tackle lip reading as an open-world problem - unconstrained natural language sentences, and in the wild videos. Our key contributions are: (1) a 'Watch, Listen, Attend and Spell' (WLAS) network that learns to transcribe videos of mouth motion to characters; (2) a curriculum learning strategy to accelerate training and to reduce overfitting; (3) a 'Lip Reading Sentences' (LRS) dataset for visual speech recognition, consisting of over 100,000 natural sentences from British television. The WLAS model trained on the LRS dataset surpasses the performance of all previous work on standard lip reading benchmark datasets, often by a significant margin. This lip reading performance beats a professional lip reader on videos from BBC television, and we also demonstrate that visual information helps to improve speech recognition performance even when the audio is available. | For recognising full words, Petridis @cite_11 trains an LSTM classifier on a discrete cosine transform (DCT) and deep bottleneck features (DBF). Similarly, Wand @cite_38 uses an LSTM with HOG input features to recognise short phrases. The shortage of training data in lip reading presumably contributes to the continued use of shallow features. Existing datasets consist of videos with only a small number of subjects, and also a very limited vocabulary ( @math 60 words), which is also an obstacle to progress. The recent paper of Chung and Zisserman @cite_25 tackles the small-lexicon problem by using faces in television broadcasts to assemble a dataset for 500 words. However, as with any word-level classification task, the setting is still distant from the real-world, given that the word boundaries must be known beforehand. A very recent work @cite_24 (under submission to ICLR 2017) uses a CNN and LSTM-based network and Connectionist Temporal Classification (CTC) @cite_14 to compute the labelling. This reports strong speaker-independent performance on the constrained grammar and 51 word vocabulary of the GRID dataset @cite_23 . However, the method, suitably modified, should be applicable to longer, more general sentences. | {
"cite_N": [
"@cite_38",
"@cite_14",
"@cite_24",
"@cite_23",
"@cite_25",
"@cite_11"
],
"mid": [
"2951015274",
"2127141656",
"2578229578",
"2015143272",
"2594690981",
"2404704342"
],
"abstract": [
"Lipreading, i.e. speech recognition from visual-only recordings of a speaker's face, can be achieved with a processing pipeline based solely on neural networks, yielding significantly better accuracy than conventional methods. Feed-forward and recurrent neural network layers (namely Long Short-Term Memory; LSTM) are stacked to form a single structure which is trained by back-propagating error gradients through all the layers. The performance of such a stacked network was experimentally evaluated and compared to a standard Support Vector Machine classifier using conventional computer vision features (Eigenlips and Histograms of Oriented Gradients). The evaluation was performed on data from 19 speakers of the publicly available GRID corpus. With 51 different words to classify, we report a best word accuracy on held-out evaluation speakers of 79.6 using the end-to-end neural network-based solution (11.6 improvement over the best feature-based solution evaluated).",
"Many real-world sequence learning tasks require the prediction of sequences of labels from noisy, unsegmented input data. In speech recognition, for example, an acoustic signal is transcribed into words or sub-word units. Recurrent neural networks (RNNs) are powerful sequence learners that would seem well suited to such tasks. However, because they require pre-segmented training data, and post-processing to transform their outputs into label sequences, their applicability has so far been limited. This paper presents a novel method for training RNNs to label unsegmented sequences directly, thereby solving both problems. An experiment on the TIMIT speech corpus demonstrates its advantages over both a baseline HMM and a hybrid HMM-RNN.",
"Lipreading is the task of decoding text from the movement of a speaker's mouth. Traditional approaches separated the problem into two stages: designing or learning visual features, and prediction. More recent deep lipreading approaches are end-to-end trainable (, 2016; Chung & Zisserman, 2016a). However, existing work on models trained end-to-end perform only word classification, rather than sentence-level sequence prediction. Studies have shown that human lipreading performance increases for longer words (Easton & Basala, 1982), indicating the importance of features capturing temporal context in an ambiguous communication channel. Motivated by this observation, we present LipNet, a model that maps a variable-length sequence of video frames to text, making use of spatiotemporal convolutions, a recurrent network, and the connectionist temporal classification loss, trained entirely end-to-end. To the best of our knowledge, LipNet is the first end-to-end sentence-level lipreading model that simultaneously learns spatiotemporal visual features and a sequence model. On the GRID corpus, LipNet achieves 95.2 accuracy in sentence-level, overlapped speaker split task, outperforming experienced human lipreaders and the previous 86.4 word-level state-of-the-art accuracy (, 2016).",
"An audio-visual corpus has been collected to support the use of common material in speech perception and automatic speech recognition studies. The corpus consists of high-quality audio and video recordings of 1000 sentences spoken by each of 34 talkers. Sentences are simple, syntactically identical phrases such as “place green at B 4 now.” Intelligibility tests using the audio signals suggest that the material is easily identifiable in quiet and low levels of stationary noise. The annotated corpus is available on the web for research use.",
"Our aim is to recognise the words being spoken by a talking face, given only the video but not the audio. Existing works in this area have focussed on trying to recognise a small number of utterances in controlled environments (e.g. digits and alphabets), partially due to the shortage of suitable datasets.",
"Deep bottleneck features (DBNFs) have been used successfully in the past for acoustic speech recognition from audio. However, research on extracting DBNFs for visual speech recognition is very limited. In this work, we present an approach to extract deep bottleneck visual features based on deep autoencoders. To the best of our knowledge, this is the first work that extracts DBNFs for visual speech recognition directly from pixels. We first train a deep autoencoder with a bottleneck layer in order to reduce the dimensionality of the image. Then the autoencoder's decoding layers are replaced by classification layers which make the bottleneck features more discriminative. Discrete Cosine Transform (DCT) features are also appended in the bottleneck layer during training in order to make the bottleneck features complementary to DCT features. Long-Short Term Memory (LSTM) networks are used to model the temporal dynamics and the performance is evaluated on the OuluVS and AVLetters databases. The extracted complementary DBNF in combination with DCT features achieve the best performance resulting in an absolute improvement of up to 5 over the DCT baseline."
]
} |
1611.05358 | 2952746495 | The goal of this work is to recognise phrases and sentences being spoken by a talking face, with or without the audio. Unlike previous works that have focussed on recognising a limited number of words or phrases, we tackle lip reading as an open-world problem - unconstrained natural language sentences, and in the wild videos. Our key contributions are: (1) a 'Watch, Listen, Attend and Spell' (WLAS) network that learns to transcribe videos of mouth motion to characters; (2) a curriculum learning strategy to accelerate training and to reduce overfitting; (3) a 'Lip Reading Sentences' (LRS) dataset for visual speech recognition, consisting of over 100,000 natural sentences from British television. The WLAS model trained on the LRS dataset surpasses the performance of all previous work on standard lip reading benchmark datasets, often by a significant margin. This lip reading performance beats a professional lip reader on videos from BBC television, and we also demonstrate that visual information helps to improve speech recognition performance even when the audio is available. | Audio-visual speech recognition. The problems of audio-visual speech recognition (AVSR) and lip reading are closely linked. Mroueh @cite_6 employs feed-forward Deep Neural Networks (DNNs) to perform phoneme classification using a large non-public audio-visual dataset. The use of HMMs together with hand-crafted or pre-trained visual features have proved popular -- @cite_32 encodes input images using DBF; @cite_40 used DCT; and @cite_34 uses a CNN pre-trained to classify phonemes; all three combine these features with HMMs to classify spoken digits or isolated words. As with lip reading, there has been little attempt to develop AVSR systems that generalise to real-world settings. | {
"cite_N": [
"@cite_40",
"@cite_34",
"@cite_32",
"@cite_6"
],
"mid": [
"1736374775",
"2076462394",
"2289925289",
"2949547965"
],
"abstract": [
"We investigate the use of facial depth data of a speaking subject, captured by the Kinect device, as an additional speechinformative modality to incorporate to a traditional audiovisual automatic speech recognizer. We present our feature extraction algorithm for both visual and accompanying depth modalities, based on a discrete cosine transform of the mouth region-of-interest data, further transformed by a two-stage linear discriminant analysis projection to incorporate speech dynamics and improve classification. For automatic speech recognition utilizing the three available data streams (audio, visual, and depth), we consider both the feature and decision fusion paradigms, the latter via a state-synchronous tri-stream hidden Markov model. We report multi-speaker recognition results on a small-vocabulary task employing our recently collected bilingual audio-visual corpus with depth information, demonstrating improved recognition performance by the addition of the proposed depth stream, across a wide range of audio conditions.",
"Audio-visual speech recognition (AVSR) system is thought to be one of the most promising solutions for reliable speech recognition, particularly when the audio is corrupted by noise. However, cautious selection of sensory features is crucial for attaining high recognition performance. In the machine-learning community, deep learning approaches have recently attracted increasing attention because deep neural networks can effectively extract robust latent features that enable various recognition algorithms to demonstrate revolutionary generalization capabilities under diverse application conditions. This study introduces a connectionist-hidden Markov model (HMM) system for noise-robust AVSR. First, a deep denoising autoencoder is utilized for acquiring noise-robust audio features. By preparing the training data for the network with pairs of consecutive multiple steps of deteriorated audio features and the corresponding clean features, the network is trained to output denoised audio features from the corresponding features deteriorated by noise. Second, a convolutional neural network (CNN) is utilized to extract visual features from raw mouth area images. By preparing the training data for the CNN as pairs of raw images and the corresponding phoneme label outputs, the network is trained to predict phoneme labels from the corresponding mouth area input images. Finally, a multi-stream HMM (MSHMM) is applied for integrating the acquired audio and visual HMMs independently trained with the respective features. By comparing the cases when normal and denoised mel-frequency cepstral coefficients (MFCCs) are utilized as audio features to the HMM, our unimodal isolated word recognition results demonstrate that approximately 65 word recognition rate gain is attained with denoised MFCCs under 10 dB signal-to-noise-ratio (SNR) for the audio signal input. Moreover, our multimodal isolated word recognition results utilizing MSHMM with denoised MFCCs and acquired visual features demonstrate that an additional word recognition rate gain is attained for the SNR conditions below 10 dB.",
"This paper develops an Audio-Visual Speech Recognition (AVSR) method, by (1) exploring high-performance visual features, (2) applying audio and visual deep bottleneck features to improve AVSR performance, and (3) investigating effectiveness of voice activity detection in a visual modality. In our approach, many kinds of visual features are incorporated, subsequently converted into bottleneck features by deep learning technology. By using proposed features, we successfully achieved 73.66 lipreading accuracy in speaker-independent open condition, and about 90 AVSR accuracy on average in noisy environments. In addition, we extracted speech segments from visual features, resulting 77.80 lipreading accuracy. It is found VAD is useful in both audio and visual modalities, for better lipreading and AVSR.",
"In this paper, we present methods in deep multimodal learning for fusing speech and visual modalities for Audio-Visual Automatic Speech Recognition (AV-ASR). First, we study an approach where uni-modal deep networks are trained separately and their final hidden layers fused to obtain a joint feature space in which another deep network is built. While the audio network alone achieves a phone error rate (PER) of @math under clean condition on the IBM large vocabulary audio-visual studio dataset, this fusion model achieves a PER of @math demonstrating the tremendous value of the visual channel in phone classification even in audio with high signal to noise ratio. Second, we present a new deep network architecture that uses a bilinear softmax layer to account for class specific correlations between modalities. We show that combining the posteriors from the bilinear networks with those from the fused model mentioned above results in a further significant phone error rate reduction, yielding a final PER of @math ."
]
} |
1611.05358 | 2952746495 | The goal of this work is to recognise phrases and sentences being spoken by a talking face, with or without the audio. Unlike previous works that have focussed on recognising a limited number of words or phrases, we tackle lip reading as an open-world problem - unconstrained natural language sentences, and in the wild videos. Our key contributions are: (1) a 'Watch, Listen, Attend and Spell' (WLAS) network that learns to transcribe videos of mouth motion to characters; (2) a curriculum learning strategy to accelerate training and to reduce overfitting; (3) a 'Lip Reading Sentences' (LRS) dataset for visual speech recognition, consisting of over 100,000 natural sentences from British television. The WLAS model trained on the LRS dataset surpasses the performance of all previous work on standard lip reading benchmark datasets, often by a significant margin. This lip reading performance beats a professional lip reader on videos from BBC television, and we also demonstrate that visual information helps to improve speech recognition performance even when the audio is available. | For the most part, prior work can be divided into two types. The first type uses CTC @cite_14 , where the model typically predicts framewise labels and then looks for the optimal alignment between the framewise predictions and the output sequence. The weakness is that the output labels are not conditioned on each other. | {
"cite_N": [
"@cite_14"
],
"mid": [
"2127141656"
],
"abstract": [
"Many real-world sequence learning tasks require the prediction of sequences of labels from noisy, unsegmented input data. In speech recognition, for example, an acoustic signal is transcribed into words or sub-word units. Recurrent neural networks (RNNs) are powerful sequence learners that would seem well suited to such tasks. However, because they require pre-segmented training data, and post-processing to transform their outputs into label sequences, their applicability has so far been limited. This paper presents a novel method for training RNNs to label unsegmented sequences directly, thereby solving both problems. An experiment on the TIMIT speech corpus demonstrates its advantages over both a baseline HMM and a hybrid HMM-RNN."
]
} |
1611.05358 | 2952746495 | The goal of this work is to recognise phrases and sentences being spoken by a talking face, with or without the audio. Unlike previous works that have focussed on recognising a limited number of words or phrases, we tackle lip reading as an open-world problem - unconstrained natural language sentences, and in the wild videos. Our key contributions are: (1) a 'Watch, Listen, Attend and Spell' (WLAS) network that learns to transcribe videos of mouth motion to characters; (2) a curriculum learning strategy to accelerate training and to reduce overfitting; (3) a 'Lip Reading Sentences' (LRS) dataset for visual speech recognition, consisting of over 100,000 natural sentences from British television. The WLAS model trained on the LRS dataset surpasses the performance of all previous work on standard lip reading benchmark datasets, often by a significant margin. This lip reading performance beats a professional lip reader on videos from BBC television, and we also demonstrate that visual information helps to improve speech recognition performance even when the audio is available. | The second type is sequence-to-sequence models @cite_31 that first read all of the input sequence before starting to predict the output sentence. A number of papers have adopted this approach for speech recognition @cite_19 @cite_26 , and the most related work to ours is that of Chan @cite_33 which proposes an elegant sequence-to-sequence method to transcribe audio signal to characters. They utilise a number of the latest sequence learning tricks such as scheduled sampling @cite_21 and attention @cite_19 ; we take many inspirations from this work. | {
"cite_N": [
"@cite_26",
"@cite_33",
"@cite_21",
"@cite_19",
"@cite_31"
],
"mid": [
"1586532344",
"",
"2950304420",
"2953022181",
"2949888546"
],
"abstract": [
"We replace the Hidden Markov Model (HMM) which is traditionally used in in continuous speech recognition with a bi-directional recurrent neural network encoder coupled to a recurrent neural network decoder that directly emits a stream of phonemes. The alignment between the input and output sequences is established using an attention mechanism: the decoder emits each symbol based on a context created with a subset of input symbols elected by the attention mechanism. We report initial results demonstrating that this new approach achieves phoneme error rates that are comparable to the state-of-the-art HMM-based decoders, on the TIMIT dataset.",
"",
"Recurrent Neural Networks can be trained to produce sequences of tokens given some input, as exemplified by recent results in machine translation and image captioning. The current approach to training them consists of maximizing the likelihood of each token in the sequence given the current (recurrent) state and the previous token. At inference, the unknown previous token is then replaced by a token generated by the model itself. This discrepancy between training and inference can yield errors that can accumulate quickly along the generated sequence. We propose a curriculum learning strategy to gently change the training process from a fully guided scheme using the true previous token, towards a less guided scheme which mostly uses the generated token instead. Experiments on several sequence prediction tasks show that this approach yields significant improvements. Moreover, it was used successfully in our winning entry to the MSCOCO image captioning challenge, 2015.",
"Recurrent sequence generators conditioned on input data through an attention mechanism have recently shown very good performance on a range of tasks in- cluding machine translation, handwriting synthesis and image caption gen- eration. We extend the attention-mechanism with features needed for speech recognition. We show that while an adaptation of the model used for machine translation in reaches a competitive 18.7 phoneme error rate (PER) on the TIMIT phoneme recognition task, it can only be applied to utterances which are roughly as long as the ones it was trained on. We offer a qualitative explanation of this failure and propose a novel and generic method of adding location-awareness to the attention mechanism to alleviate this issue. The new method yields a model that is robust to long inputs and achieves 18 PER in single utterances and 20 in 10-times longer (repeated) utterances. Finally, we propose a change to the at- tention mechanism that prevents it from concentrating too much on single frames, which further reduces PER to 17.6 level.",
"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."
]
} |
1611.05550 | 2554773066 | Many applications, such as photon-limited imaging and genomics, involve large datasets with noisy entries from exponential family distributions. It is of interest to estimate the covariance structure and principal components of the noiseless distribution. Principal Component Analysis (PCA), the standard method for this setting, can be inefficient when the noise is non-Gaussian. We develop @math PCA (exponential family PCA), a new methodology for PCA on exponential family distributions. @math PCA can be used for dimensionality reduction and denoising of large data matrices. @math PCA involves the eigendecomposition of a new covariance matrix estimator, constructed in a simple and deterministic way using moment calculations, shrinkage, and random matrix theory. We provide several theoretical justifications for our estimator, including the finite-sample convergence rate, and the Marchenko-Pastur law in high dimensions. @math PCA compares favorably to PCA and various PCA alternatives for exponential families, in simulations as well as in XFEL and SNP data analysis. An open-source implementation is available. | To give context for our method, we review related work. The reader intersted in the methodology can skip directly to Section . We refer to @cite_5 for a detailed overview of PCA methodology, to @cite_12 for a more general overview of multivariate statistical analysis including PCA, and to @cite_0 for discussions of high-dimensional statistics, random matrix theory and PCA. | {
"cite_N": [
"@cite_0",
"@cite_5",
"@cite_12"
],
"mid": [
"143531921",
"2148694408",
"2063698478"
],
"abstract": [
"1. Introduction 2. Limiting spectral distributions 3. CLT for linear spectral statistics 4. The generalised variance and multiple correlation coefficient 5. The T2-statistic 6. Classification of data 7. Testing the general linear hypothesis 8. Testing independence of sets of variates 9. Testing hypotheses of equality of covariance matrices 10. Estimation of the population spectral distribution 11. Large-dimensional spiked population models 12. Efficient optimisation of a large financial portfolio.",
"Introduction * Properties of Population Principal Components * Properties of Sample Principal Components * Interpreting Principal Components: Examples * Graphical Representation of Data Using Principal Components * Choosing a Subset of Principal Components or Variables * Principal Component Analysis and Factor Analysis * Principal Components in Regression Analysis * Principal Components Used with Other Multivariate Techniques * Outlier Detection, Influential Observations and Robust Estimation * Rotation and Interpretation of Principal Components * Principal Component Analysis for Time Series and Other Non-Independent Data * Principal Component Analysis for Special Types of Data * Generalizations and Adaptations of Principal Component Analysis",
"Preface to the Third Edition.Preface to the Second Edition.Preface to the First Edition.1. Introduction.2. The Multivariate Normal Distribution.3. Estimation of the Mean Vector and the Covariance Matrix.4. The Distributions and Uses of Sample Correlation Coefficients.5. The Generalized T2-Statistic.6. Classification of Observations.7. The Distribution of the Sample Covariance Matrix and the Sample Generalized Variance.8. Testing the General Linear Hypothesis: Multivariate Analysis of Variance9. Testing Independence of Sets of Variates.10. Testing Hypotheses of Equality of Covariance Matrices and Equality of Mean Vectors and Covariance Matrices.11. Principal Components.12. Cononical Correlations and Cononical Variables.13. The Distributions of Characteristic Roots and Vectors.14. Factor Analysis.15. Pattern of Dependence Graphical Models.Appendix A: Matrix Theory.Appendix B: Tables.References.Index."
]
} |
1611.05550 | 2554773066 | Many applications, such as photon-limited imaging and genomics, involve large datasets with noisy entries from exponential family distributions. It is of interest to estimate the covariance structure and principal components of the noiseless distribution. Principal Component Analysis (PCA), the standard method for this setting, can be inefficient when the noise is non-Gaussian. We develop @math PCA (exponential family PCA), a new methodology for PCA on exponential family distributions. @math PCA can be used for dimensionality reduction and denoising of large data matrices. @math PCA involves the eigendecomposition of a new covariance matrix estimator, constructed in a simple and deterministic way using moment calculations, shrinkage, and random matrix theory. We provide several theoretical justifications for our estimator, including the finite-sample convergence rate, and the Marchenko-Pastur law in high dimensions. @math PCA compares favorably to PCA and various PCA alternatives for exponential families, in simulations as well as in XFEL and SNP data analysis. An open-source implementation is available. | There have been several approaches suggested for extending PCA to non-Gaussian distributions, see. e.g., @cite_5 , Sec. 14.4. One possibility is to use robust estimates of the covariance matrix [see][Sec. 14.4, for references] jolliffe2002principal . Another approach assumes that the natural parameter lies in a low dimensional space , and then attempts to maximize the log-likelihood. This leads to a non-convex optimization problem for which an alternating maximization method is proposed, without global convergence guarantees. More recently, described a similar generalization of PCA, while proposed another likelihood-based method, both without global convergence guarantees. Scalable methods include @cite_10 , albeit without precise performance guarantees in high dimensions. | {
"cite_N": [
"@cite_5",
"@cite_10"
],
"mid": [
"2148694408",
"2963481531"
],
"abstract": [
"Introduction * Properties of Population Principal Components * Properties of Sample Principal Components * Interpreting Principal Components: Examples * Graphical Representation of Data Using Principal Components * Choosing a Subset of Principal Components or Variables * Principal Component Analysis and Factor Analysis * Principal Components in Regression Analysis * Principal Components Used with Other Multivariate Techniques * Outlier Detection, Influential Observations and Robust Estimation * Rotation and Interpretation of Principal Components * Principal Component Analysis for Time Series and Other Non-Independent Data * Principal Component Analysis for Special Types of Data * Generalizations and Adaptations of Principal Component Analysis",
"We develop a flexible framework for low-rank matrix estimation that allows us to transform noise models into regularization schemes via a simple bootstrap algorithm. Effectively, our procedure seeks an autoencoding basis for the observed matrix that is stable with respect to the specified noise model; we call the resulting procedure a stable autoencoder. In the simplest case, with an isotropic noise model, our method is equivalent to a classical singular value shrinkage estimator. For non-isotropic noise models--e.g., Poisson noise-- the method does not reduce to singular value shrinkage, and instead yields new estimators that perform well in experiments. Moreover, by iterating our stable autoencoding scheme, we can automatically generate low-rank estimates without specifying the target rank as a tuning parameter."
]
} |
1611.05550 | 2554773066 | Many applications, such as photon-limited imaging and genomics, involve large datasets with noisy entries from exponential family distributions. It is of interest to estimate the covariance structure and principal components of the noiseless distribution. Principal Component Analysis (PCA), the standard method for this setting, can be inefficient when the noise is non-Gaussian. We develop @math PCA (exponential family PCA), a new methodology for PCA on exponential family distributions. @math PCA can be used for dimensionality reduction and denoising of large data matrices. @math PCA involves the eigendecomposition of a new covariance matrix estimator, constructed in a simple and deterministic way using moment calculations, shrinkage, and random matrix theory. We provide several theoretical justifications for our estimator, including the finite-sample convergence rate, and the Marchenko-Pastur law in high dimensions. @math PCA compares favorably to PCA and various PCA alternatives for exponential families, in simulations as well as in XFEL and SNP data analysis. An open-source implementation is available. | The recent work of @cite_13 develops a generalized Stein's Unbiased Risk Estimation (SURE) approach for singular value shrinkage denoising of low-rank matrices in exponential families. However, their shrinkage formulas become numerically intractable for Frobenius norm beyond Gaussian errors, and they instead introduce a heuristic algorithm. Their work is geared towards higher signal-to-noise ratio settings. | {
"cite_N": [
"@cite_13"
],
"mid": [
"2404992943"
],
"abstract": [
"We consider the problem of estimating a low-rank signal matrix from noisy measurements under the assumption that the distribution of the data matrix belongs to an exponential family. In this setting, we derive generalized Stein's unbiased risk estimation (SURE) formulas that hold for any spectral estimators which shrink or threshold the singular values of the data matrix. This leads to new data-driven spectral estimators, whose optimality is discussed using tools from random matrix theory and through numerical experiments. Under the spiked population model and in the asymptotic setting where the dimensions of the data matrix are let going to infinity, some theoretical properties of our approach are compared to recent results on asymptotically optimal shrinking rules for Gaussian noise. It also leads to new procedures for singular values shrinkage in finite-dimensional matrix denoising for Gamma-distributed and Poisson-distributed measurements."
]
} |
1611.05550 | 2554773066 | Many applications, such as photon-limited imaging and genomics, involve large datasets with noisy entries from exponential family distributions. It is of interest to estimate the covariance structure and principal components of the noiseless distribution. Principal Component Analysis (PCA), the standard method for this setting, can be inefficient when the noise is non-Gaussian. We develop @math PCA (exponential family PCA), a new methodology for PCA on exponential family distributions. @math PCA can be used for dimensionality reduction and denoising of large data matrices. @math PCA involves the eigendecomposition of a new covariance matrix estimator, constructed in a simple and deterministic way using moment calculations, shrinkage, and random matrix theory. We provide several theoretical justifications for our estimator, including the finite-sample convergence rate, and the Marchenko-Pastur law in high dimensions. @math PCA compares favorably to PCA and various PCA alternatives for exponential families, in simulations as well as in XFEL and SNP data analysis. An open-source implementation is available. | There are many approaches to denoising in image and signal processing, the majority designed for Gaussian noise [see e.g.,][] starck2010sparse . Most classical methods are designed for single-image denoising'', and do not share information across multiple images. Our setting is different, because we have many very noisy samples---e.g., XFEL images. @cite_1 Sec. 6.5. provides an overview of the classical methods for Poisson noise. Popular approaches reduce to the Gaussian case by a wavelet transform such as a Haar transform ; by adaptive wavelet shrinkage; or by approximate variance stabilization such as the Anscombe transform. The latter is known to work well for Poisson signals with large parameters, due to approximate normality. However, the normal approximation breaks down for the Poisson with a small parameter, such as photon-limited XFEL [see e.g.,][Sec. 6.6] starck2010sparse . | {
"cite_N": [
"@cite_1"
],
"mid": [
"228380312"
],
"abstract": [
"Mallat's book is the undisputed reference in this field - it is the only one that covers the essential material in such breadth and depth. - Laurent Demanet, Stanford University The new edition of this classic book gives all the major concepts, techniques and applications of sparse representation, reflecting the key role the subject plays in today's signal processing. The book clearly presents the standard representations with Fourier, wavelet and time-frequency transforms, and the construction of orthogonal bases with fast algorithms. The central concept of sparsity is explained and applied to signal compression, noise reduction, and inverse problems, while coverage is given to sparse representations in redundant dictionaries, super-resolution and compressive sensing applications. Features: * Balances presentation of the mathematics with applications to signal processing * Algorithms and numerical examples are implemented in WaveLab, a MATLAB toolbox * Companion website for instructors and selected solutions and code available for students New in this edition * Sparse signal representations in dictionaries * Compressive sensing, super-resolution and source separation * Geometric image processing with curvelets and bandlets * Wavelets for computer graphics with lifting on surfaces * Time-frequency audio processing and denoising * Image compression with JPEG-2000 * New and updated exercises A Wavelet Tour of Signal Processing: The Sparse Way, third edition, is an invaluable resource for researchers and R&D engineers wishing to apply the theory in fields such as image processing, video processing and compression, bio-sensing, medical imaging, machine vision and communications engineering. Stephane Mallat is Professor in Applied Mathematics at cole Polytechnique, Paris, France. From 1986 to 1996 he was a Professor at the Courant Institute of Mathematical Sciences at New York University, and between 2001 and 2007, he co-founded and became CEO of an image processing semiconductor company. Companion website: A Numerical Tour of Signal Processing * Includes all the latest developments since the book was published in 1999, including its application to JPEG 2000 and MPEG-4 * Algorithms and numerical examples are implemented in Wavelab, a MATLAB toolbox * Balances presentation of the mathematics with applications to signal processing"
]
} |
1611.05368 | 2550944152 | The artistic style of a painting is a subtle aesthetic judgment used by art historians for grouping and classifying artwork. The recently introduced neural-style' algorithm substantially succeeds in merging the perceived artistic style of one image or set of images with the perceived content of another. In light of this and other recent developments in image analysis via convolutional neural networks, we investigate the effectiveness of a neural-style' representation for classifying the artistic style of paintings. | Algorithmic determination of artistic style in paintings has only been considered sporadically in the past. Examples of early efforts at style classification are @cite_9 and @cite_3 , where the datasets used are quite small, and only a handful of very distinct artistic style categories considered. Several complex models are constructed in @cite_11 by hand-engineering features on a large dataset similar to the one used for this work. And in @cite_10 , it is demonstrated that convolutional neural networks may be effective for understanding image style in general, including artistic style in paintings. In the papers just mentioned the number of artistic style categories is held to a relatively small 25 and 27 broadly defined style categories arespectively. | {
"cite_N": [
"@cite_9",
"@cite_10",
"@cite_3",
"@cite_11"
],
"mid": [
"2104813487",
"2166242527",
"2128907386",
""
],
"abstract": [
"The goal of this paper is to offer a framework for classification of images and video according to their \"type\", or \"style\"--a problem which is hard to define, but easy to illustrate; for example, identifying an artist by the Style of his her painting, or determining the activity in a video sequence. The paper offers a simple classification paradigm based on local properties of spatial or spatio-temporal blocks. The learning and classification are based on the naive Bayes classifier. A few experimental results are presented.",
"The style of an image plays a significant role in how it is viewed, but style has received little attention in computer vision research. We describe an approach to predicting style of images, and perform a thorough evaluation of different image features for these tasks. We find that features learned in a multi-layer network generally perform best -- even when trained with object class (not style) labels. Our large-scale learning methods results in the best published performance on an existing dataset of aesthetic ratings and photographic style annotations. We present two novel datasets: 80K Flickr photographs annotated with 20 curated style labels, and 85K paintings annotated with 25 style genre labels. Our approach shows excellent classification performance on both datasets. We use the learned classifiers to extend traditional tag-based image search to consider stylistic constraints, and demonstrate cross-dataset understanding of style.",
"We describe a method for automated recognition of painters and schools of art based on their signature styles and studied the computer-based perception of visual art. Paintings of nine artists, representing three different schools of art—impressionism, surrealism and abstract expressionism—were analyzed using a large set of image features and image transforms. The computed image descriptors were assessed using Fisher scores, and the most informative features were used for the classification and similarity measurements of paintings, painters, and schools of art. Experimental results show that the classification accuracy when classifying paintings into nine painter classes is 77p, and the accuracy of associating a given painting with its school of art is 91p. An interesting feature of the proposed method is its ability to automatically associate different artists that share the same school of art in an unsupervised fashion. The source code used for the image classification and image similarity described in this article is available for free download.",
""
]
} |
1611.05368 | 2550944152 | The artistic style of a painting is a subtle aesthetic judgment used by art historians for grouping and classifying artwork. The recently introduced neural-style' algorithm substantially succeeds in merging the perceived artistic style of one image or set of images with the perceived content of another. In light of this and other recent developments in image analysis via convolutional neural networks, we investigate the effectiveness of a neural-style' representation for classifying the artistic style of paintings. | In the paper A Neural Algorithm of Artistic Style'', it is demonstrated that the correlations between the low-level feature activations in a deep convolutional neural network encode sufficient information about the style of the input image to permit a tranfer of the visual style of the input image onto a new image via an algorithm informally referred to as the neural-style'' algorithm @cite_14 . An example of the output of this algorithm is presented in Figure . Several authors have built upon the work of Gatys et. al. in the past year @cite_4 , @cite_0 , @cite_0 , @cite_13 . These investigations have primarily focused on ways to improve either the quality of the style transfer or the efficiency of the algorithm. To the best of our knowledge the only other look at the use of the style representation of an image as a classifier is in @cite_7 . | {
"cite_N": [
"@cite_14",
"@cite_4",
"@cite_7",
"@cite_0",
"@cite_13"
],
"mid": [
"2475287302",
"2344328033",
"2415436437",
"2275086408",
"2950689937"
],
"abstract": [
"Rendering the semantic content of an image in different styles is a difficult image processing task. Arguably, a major limiting factor for previous approaches has been the lack of image representations that explicitly represent semantic information and, thus, allow to separate image content from style. Here we use image representations derived from Convolutional Neural Networks optimised for object recognition, which make high level image information explicit. We introduce A Neural Algorithm of Artistic Style that can separate and recombine the image content and style of natural images. The algorithm allows us to produce new images of high perceptual quality that combine the content of an arbitrary photograph with the appearance of numerous wellknown artworks. Our results provide new insights into the deep image representations learned by Convolutional Neural Networks and demonstrate their potential for high level image synthesis and manipulation.",
"In the past, manually re-drawing an image in a certain artistic style required a professional artist and a long time. Doing this for a video sequence single-handed was beyond imagination. Nowadays computers provide new possibilities. We present an approach that transfers the style from one image (for example, a painting) to a whole video sequence. We make use of recent advances in style transfer in still images and propose new initializations and loss functions applicable to videos. This allows us to generate consistent and stable stylized video sequences, even in cases with large motion and strong occlusion. We show that the proposed method clearly outperforms simpler baselines both qualitatively and quantitatively.",
"In this paper, we have examined the effectiveness of \"style matrix\" which is used in the works on style transfer and texture synthesis by in the context of image retrieval as image features. A style matrix is presented by Gram matrix of the feature maps in a deep convolutional neural network. We proposed a style vector which are generated from a style matrix with PCA dimension reduction. In the experiments, we evaluate image retrieval performance using artistic images downloaded from Wikiarts.org regarding both artistic styles ans artists. We have obtained 40.64 and 70.40 average precision for style search and artist search, respectively, both of which outperformed the results by common CNN features. In addition, we found PCA-compression boosted the performance.",
"We explore the method of style transfer presented in the article \"A Neural Algorithm of Artistic Style\" by Leon A. Gatys, Alexander S. Ecker and Matthias Bethge (arXiv:1508.06576). We first demonstrate the power of the suggested style space on a few examples. We then vary different hyper-parameters and program properties that were not discussed in the original paper, among which are the recognition network used, starting point of the gradient descent and different ways to partition style and content layers. We also give a brief comparison of some of the existing algorithm implementations and deep learning frameworks used. To study the style space further we attempt to generate synthetic images by maximizing a single entry in one of the Gram matrices @math and some interesting results are observed. Next, we try to mimic the sparsity and intensity distribution of Gram matrices obtained from a real painting and generate more complex textures. Finally, we propose two new style representations built on top of network's features and discuss how one could be used to achieve local and potentially content-aware style transfer.",
"We consider image transformation problems, where an input image is transformed into an output image. Recent methods for such problems typically train feed-forward convolutional neural networks using a loss between the output and ground-truth images. Parallel work has shown that high-quality images can be generated by defining and optimizing loss functions based on high-level features extracted from pretrained networks. We combine the benefits of both approaches, and propose the use of perceptual loss functions for training feed-forward networks for image transformation tasks. We show results on image style transfer, where a feed-forward network is trained to solve the optimization problem proposed by in real-time. Compared to the optimization-based method, our network gives similar qualitative results but is three orders of magnitude faster. We also experiment with single-image super-resolution, where replacing a per-pixel loss with a perceptual loss gives visually pleasing results."
]
} |
1611.05428 | 2549141685 | Compression can sometimes improve performance by making more of the data available to the processors faster. We consider the compression of integer keys in a B+-tree index. For this purpose, systems such as IBM DB2 use variable-byte compression over differentially coded keys. We revisit this problem with various compression alternatives such as Google's VarIntGB, Binary Packing and Frame-of-Reference. In all cases, we describe algorithms that can operate directly on compressed data. Many of our alternatives exploit the single-instruction-multiple-data (SIMD) instructions supported by modern CPUs. We evaluate our techniques in a database environment provided by Upscaledb, a production-quality key-value database. Our best techniques are SIMD accelerated: they simultaneously reduce memory usage while improving single-threaded speeds. In particular, a differentially coded SIMD binary-packing techniques (BP128) can offer a superior query speed (e.g., 40 better than an uncompressed database) while providing the best compression (e.g., by a factor of ten). For analytic workloads, our fast compression techniques offer compelling benefits. Our software is available as open source. | There has been much research dedicated to improving the performance of tree data structures. For example, cache-conscious trees can reduce the number of cache misses to improve performance @cite_1 @cite_0 . In particular, @cite_0 propose the Cache Sensitive T-Trees (CST-Trees). | {
"cite_N": [
"@cite_0",
"@cite_1"
],
"mid": [
"2145148249",
"2127542829"
],
"abstract": [
"Researchers have modified existing index structures into ones optimized for CPU cache performance in main memory database environments. A Cache Sensitive B+-Tree is one of them. It is designed to minimize the impact of cache misses for B+-Trees and it has been known to be more effective than other types of main memory index structure including T-Trees. In this paper, we introduce a Cache Sensitive T-Tree (CST-Tree) and show how T-Trees can also be redesigned to be cache sensitive. We present an experimental performance study which shows that our Cache Sensitive T-Trees can outperform the original T-Trees and Cache Sensitive B+-Trees on average 65 percent and 17 percent, respectively.",
"The performance of in-memory databases is significantly affected by the number of data blocks fetched from memory into the processor-resident cache. In recent years, various tree-based indexes have been proposed for main memory databases. A common assumption in the analysis of these indexes is that there is no data in the cache that can be reused between key lookups, i.e. the system has a cold cache for each lookup. In practice, though, the ?temperature? of the cache is strongly dependent on the application. For example, a warm cache is typical for OLTP applications that query the same index over and over with little computation in between lookups. In this paper, we present a comparative study of the cache behavior of various B+-tree-based indexes which shows that none of them performs best in all cases. Also, we propose a lightweight technique for improving the cache behavior of any B+-tree based index that performs best in all settings."
]
} |
1611.05428 | 2549141685 | Compression can sometimes improve performance by making more of the data available to the processors faster. We consider the compression of integer keys in a B+-tree index. For this purpose, systems such as IBM DB2 use variable-byte compression over differentially coded keys. We revisit this problem with various compression alternatives such as Google's VarIntGB, Binary Packing and Frame-of-Reference. In all cases, we describe algorithms that can operate directly on compressed data. Many of our alternatives exploit the single-instruction-multiple-data (SIMD) instructions supported by modern CPUs. We evaluate our techniques in a database environment provided by Upscaledb, a production-quality key-value database. Our best techniques are SIMD accelerated: they simultaneously reduce memory usage while improving single-threaded speeds. In particular, a differentially coded SIMD binary-packing techniques (BP128) can offer a superior query speed (e.g., 40 better than an uncompressed database) while providing the best compression (e.g., by a factor of ten). For analytic workloads, our fast compression techniques offer compelling benefits. Our software is available as open source. | The application of SIMD instructions to accelerate B+-tree operations (without compression) is reviewed by Zhou and Ross @cite_36 . @cite_15 describe how to scan quickly column stores using SIMD instructions. @cite_16 show how to accelerate K-ary search on modern processors. @cite_40 describe the IBM DB2 column store that makes extensive use of SIMD instructions and compression. | {
"cite_N": [
"@cite_36",
"@cite_40",
"@cite_15",
"@cite_16"
],
"mid": [
"2147076738",
"1967601791",
"2096496252",
"2073216502"
],
"abstract": [
"Modern CPUs have instructions that allow basic operations to be performed on several data elements in parallel. These instructions are called SIMD instructions, since they apply a single instruction to multiple data elements. SIMD technology was initially built into commodity processors in order to accelerate the performance of multimedia applications. SIMD instructions provide new opportunities for database engine design and implementation. We study various kinds of operations in a database context, and show how the inner loop of the operations can be accelerated using SIMD instructions. The use of SIMD instructions has two immediate performance benefits: It allows a degree of parallelism, so that many operands can be processed at once. It also often leads to the elimination of conditional branch instructions, reducing branch mispredictions.We consider the most important database operations, including sequential scans, aggregation, index operations, and joins. We present techniques for implementing these using SIMD instructions. We show that there are significant benefits in redesigning traditional query processing algorithms so that they can make better use of SIMD technology. Our study shows that using a SIMD parallelism of four, the CPU time for the new algorithms is from 10 to more than four times less than for the traditional algorithms. Superlinear speedups are obtained as a result of the elimination of branch misprediction effects.",
"DB2 with BLU Acceleration deeply integrates innovative new techniques for defining and processing column-organized tables that speed read-mostly Business Intelligence queries by 10 to 50 times and improve compression by 3 to 10 times, compared to traditional row-organized tables, without the complexity of defining indexes or materialized views on those tables. But DB2 BLU is much more than just a column store. Exploiting frequency-based dictionary compression and main-memory query processing technology from the Blink project at IBM Research - Almaden, DB2 BLU performs most SQL operations - predicate application (even range predicates and IN-lists), joins, and grouping - on the compressed values, which can be packed bit-aligned so densely that multiple values fit in a register and can be processed simultaneously via SIMD (single-instruction, multipledata) instructions. Designed and built from the ground up to exploit modern multi-core processors, DB2 BLU's hardware-conscious algorithms are carefully engineered to maximize parallelism by using novel data structures that need little latching, and to minimize data-cache and instruction-cache misses. Though DB2 BLU is optimized for in-memory processing, database size is not limited by the size of main memory. Fine-grained synopses, late materialization, and a new probabilistic buffer pool protocol for scans minimize disk I Os, while aggressive prefetching reduces I O stalls. Full integration with DB2 ensures that DB2 with BLU Acceleration benefits from the full functionality and robust utilities of a mature product, while still enjoying order-of-magnitude performance gains from revolutionary technology without even having to change the SQL, and can mix column-organized and row-organized tables in the same tablespace and even within the same query.",
"The availability of huge system memory, even on standard servers, generated a lot of interest in main memory database engines. In data warehouse systems, highly compressed column-oriented data structures are quite prominent. In order to scale with the data volume and the system load, many of these systems are highly distributed with a shared-nothing approach. The fundamental principle of all systems is a full table scan over one or multiple compressed columns. Recent research proposed different techniques to speedup table scans like intelligent compression or using an additional hardware such as graphic cards or FPGAs. In this paper, we show that utilizing the embedded Vector Processing Units (VPUs) found in standard superscalar processors can speed up the performance of mainmemory full table scan by factors. This is achieved without changing the hardware architecture and thereby without additional power consumption. Moreover, as on-chip VPUs directly access the system's RAM, no additional costly copy operations are needed for using the new SIMD-scan approach in standard main memory database engines. Therefore, we propose this scan approach to be used as the standard scan operator for compressed column-oriented main memory storage. We then discuss how well our solution scales with the number of processor cores; consequently, to what degree it can be applied in multi-threaded environments. To verify the feasibility of our approach, we implemented the proposed techniques on a modern Intel multi-core processor using Intel® Streaming SIMD Extensions (Intel® SSE). In addition, we integrated the new SIMD-scan approach into SAP® Netweaver® Business Warehouse Accelerator. We conclude with describing the performance benefits of using our approach for processing and scanning compressed data using VPUs in column-oriented main memory database systems.",
"This paper presents novel tree-based search algorithms that exploit the SIMD instructions found in virtually all modern processors. The algorithms are a natural extension of binary search: While binary search performs one comparison at each iteration, thereby cutting the search space in two halves, our algorithms perform k comparisons at a time and thus cut the search space into k pieces. On traditional processors, this so-called k-ary search procedure is not beneficial because the cost increase per iteration offsets the cost reduction due to the reduced number of iterations. On modern processors, however, multiple scalar operations can be executed simultaneously, which makes k-ary search attractive. In this paper, we provide two different search algorithms that differ in terms of efficiency and memory access patterns. Both algorithms are first described in a platform independent way and then evaluated on various state-of-the-art processors. Our experiments suggest that k-ary search provides significant performance improvements (factor two and more) on most platforms."
]
} |
1611.05428 | 2549141685 | Compression can sometimes improve performance by making more of the data available to the processors faster. We consider the compression of integer keys in a B+-tree index. For this purpose, systems such as IBM DB2 use variable-byte compression over differentially coded keys. We revisit this problem with various compression alternatives such as Google's VarIntGB, Binary Packing and Frame-of-Reference. In all cases, we describe algorithms that can operate directly on compressed data. Many of our alternatives exploit the single-instruction-multiple-data (SIMD) instructions supported by modern CPUs. We evaluate our techniques in a database environment provided by Upscaledb, a production-quality key-value database. Our best techniques are SIMD accelerated: they simultaneously reduce memory usage while improving single-threaded speeds. In particular, a differentially coded SIMD binary-packing techniques (BP128) can offer a superior query speed (e.g., 40 better than an uncompressed database) while providing the best compression (e.g., by a factor of ten). For analytic workloads, our fast compression techniques offer compelling benefits. Our software is available as open source. | Compression in databases has a long history @cite_28 . Compression techniques such as run-length encoding and differential coding are common, for example, in column-oriented databases @cite_20 . IBM DB2 compresses integer keys using variable-byte compression and differential coding @cite_17 . Graefe @cite_6 describes the compression opportunity when keys are consecutive. | {
"cite_N": [
"@cite_28",
"@cite_6",
"@cite_20",
"@cite_17"
],
"mid": [
"1993819379",
"2074125039",
"1791587242",
"2151131744"
],
"abstract": [
"In this paper, we show how compression can be integrated into a relational database system. Specifically, we describe how the storage manager, the query execution engine, and the query optimizer of a database system can be extended to deal with compressed data. Our main result is that compression can significantly improve the response time of queries if very light-weight compression techniques are used. We will present such light-weight compression techniques and give the results of running the TPC-D benchmark on a so compressed database and a non-compressed database using the AODB database system, an experimental database system that was developed at the Universities of Mannheim and Passau. Our benchmark results demonstrate that compression indeed offers high performance gains (up to 50 ) for IO-intensive queries and moderate gains for CPU-intensive queries. Compression can, however, also increase the running time of certain update operations. In all, we recommend to extend today's database systems with light-weight compression techniques and to make extensive use of this feature.",
"Column-oriented storage formats have been proposed for query processing in relational data warehouses, specifically for fast scans over non-indexed columns. This short note proposes a data compression method that reuses traditional on-disk B-tree structures with only minor changes yet achieves storage density and scan performance comparable to specialized columnar designs. The advantage of the proposed method over alternative storage structures is that traditional algorithms can be reused, e.g., for assembling rows with multiple columns, bulk insertion and deletion, logging and recovery, consistency checking, etc.",
"This paper presents the design of a read-optimized relational DBMS that contrasts sharply with most current systems, which are write-optimized. Among the many differences in its design are: storage of data by column rather than by row, careful coding and packing of objects into storage including main memory during query processing, storing an overlapping collection of column-oriented projections, rather than the current fare of tables and indexes, a non-traditional implementation of transactions which includes high availability and snapshot isolation for read-only transactions, and the extensive use of bitmap indexes to complement B-tree structures.We present preliminary performance data on a subset of TPC-H and show that the system we are building, C-Store, is substantially faster than popular commercial products. Hence, the architecture looks very encouraging.",
"In database systems, the cost of data storage and retrieval are important components of the total cost and response time of the system. A popular mechanism to reduce the storage footprint is by compressing the data residing in tables and indexes. Compressing indexes efficiently, while maintaining response time requirements, is known to be challenging. This is especially true when designing for a workload spectrum covering both data warehousing and transaction processing environments. DB2 Linux, UNIX, Windows (LUW) recently introduced index compression for use in both environments. This uses techniques that are able to compress index data efficiently while incurring virtually no performance penalty for query processing. On the contrary, for certain operations, the performance is actually better. In this paper, we detail the design of index compression in DB2 LUW and discuss the challenges that were encountered in meeting the design goals. We also demonstrate its effectiveness by showing performance results on typical customer scenarios."
]
} |
1611.05428 | 2549141685 | Compression can sometimes improve performance by making more of the data available to the processors faster. We consider the compression of integer keys in a B+-tree index. For this purpose, systems such as IBM DB2 use variable-byte compression over differentially coded keys. We revisit this problem with various compression alternatives such as Google's VarIntGB, Binary Packing and Frame-of-Reference. In all cases, we describe algorithms that can operate directly on compressed data. Many of our alternatives exploit the single-instruction-multiple-data (SIMD) instructions supported by modern CPUs. We evaluate our techniques in a database environment provided by Upscaledb, a production-quality key-value database. Our best techniques are SIMD accelerated: they simultaneously reduce memory usage while improving single-threaded speeds. In particular, a differentially coded SIMD binary-packing techniques (BP128) can offer a superior query speed (e.g., 40 better than an uncompressed database) while providing the best compression (e.g., by a factor of ten). For analytic workloads, our fast compression techniques offer compelling benefits. Our software is available as open source. | Jin and Chung improve the CST-Trees by using FOR compression @cite_18 . Similarly, @cite_12 propose a SIMD-accelerated in-memory tree index (FAST) where they use FOR @. In related work, @cite_13 propose the VAST-Tree: it improves upon FAST in several ways. In particular, it offers better compression ratios of the keys than FAST by using differential coding and the PFOR compression scheme @cite_42 . Though fast, PFOR does not exploit SIMD instructions: Lemire and Boytsov @cite_35 found that SIMD-accelerated binary packing (i.e., BP128) could be 2 to 3 times faster with little difference in the compression ratios. | {
"cite_N": [
"@cite_35",
"@cite_18",
"@cite_42",
"@cite_13",
"@cite_12"
],
"mid": [
"1791987072",
"2099958087",
"",
"1991780427",
"2151224499"
],
"abstract": [
"In many important applications-such as search engines and relational database systems-data are stored in the form of arrays of integers. Encoding and, most importantly, decoding of these arrays consumes considerable CPUtime. Therefore, substantial effort has been made to reduce costs associated with compression and decompression. In particular, researchers have exploited the superscalar nature of modern processors and single-instruction, multiple-data SIMD instructions. Nevertheless, we introduce a novel vectorized scheme called SIMD-BP128i¾? that improves over previously proposed vectorized approaches. It is nearly twice as fast as the previously fastest schemes on desktop processors varint-G8IU and PFOR. At the same time, SIMD-BP128i¾? saves up to 2bits int. For even better compression, we propose another new vectorized scheme SIMD-FastPFOR that has a compression ratio within 10 of a state-of-the-art scheme Simple-8b while being two times faster during decoding. Copyright © 2013 John Wiley & Sons, Ltd.",
"Cache-conscious indices have been researched for a decade. These index structures can enhance data locality and reduce cache misses. Cache-Sensitive B+-tree (CSB+-tree) is a state-of-the-art, high performance index tree for main-memory database systems. However, in the worst case, the space utilization of CSB+-tree can fall to 25 . The main contribution of this paper is to improve space efficiency of CSB+-tree by using compression mechanisms. We have conducted experimental studies. Results show that the compressed CSB+-trees can yield significant space savings with additional benefits for search and delete operations by utilizing the superiority of modern hardware.",
"",
"We propose a compact and efficient index structure for massive data sets. Several indexing techniques are widely-used and well-known such as binary trees and B+trees. Unfortunately, we find that these techniques suffer major two shortcomings when applied to massive sets; first, their indices are so large they could overflow regular main memory, and, second, they suffer from a variety of penalties (e.g., conditional branches, low cache hits, and TLB misses), which restricts the number of instructions executed per processor cycle. Our state-of-the-art index structure, called VAST-Tree, classifies branch nodes into multiple layers. It applies existing techniques such as cache-conscious, aligned, and branch-free structures to the top layers of branch nodes in trees. Next, it applies the adaptive compression technique to save space and harness data parallelism with SIMD instructions to the middle and bottom layers of branch nodes. Moreover, a processor-friendly compression technique is applied to leaf nodes. The end result is that trees are much more compact and traversal efficiency is high. We implement a prototype and show its resulting index size and performance as compared to binary trees, and the hardware-conscious technique called FAST which currently offers the highest performance. Compared to current alternatives, VAST-Tree compacts the branch nodes by more than 95 , and the overall index size by 47-84 given that there are 230 keys. With 228 keys, it has roughly 6.0-times and 1.24-times throughput and saves the memory consumption by more than 94.7 and 40.5 as compared to binary trees and FAST, respectively.",
"In-memory tree structured index search is a fundamental database operation. Modern processors provide tremendous computing power by integrating multiple cores, each with wide vector units. There has been much work to exploit modern processor architectures for database primitives like scan, sort, join and aggregation. However, unlike other primitives, tree search presents significant challenges due to irregular and unpredictable data accesses in tree traversal. In this paper, we present FAST, an extremely fast architecture sensitive layout of the index tree. FAST is a binary tree logically organized to optimize for architecture features like page size, cache line size, and SIMD width of the underlying hardware. FAST eliminates impact of memory latency, and exploits thread-level and datalevel parallelism on both CPUs and GPUs to achieve 50 million (CPU) and 85 million (GPU) queries per second, 5X (CPU) and 1.7X (GPU) faster than the best previously reported performance on the same architectures. FAST supports efficient bulk updates by rebuilding index trees in less than 0.1 seconds for datasets as large as 64Mkeys and naturally integrates compression techniques, overcoming the memory bandwidth bottleneck and achieving a 6X performance improvement over uncompressed index search for large keys on CPUs."
]
} |
1611.05428 | 2549141685 | Compression can sometimes improve performance by making more of the data available to the processors faster. We consider the compression of integer keys in a B+-tree index. For this purpose, systems such as IBM DB2 use variable-byte compression over differentially coded keys. We revisit this problem with various compression alternatives such as Google's VarIntGB, Binary Packing and Frame-of-Reference. In all cases, we describe algorithms that can operate directly on compressed data. Many of our alternatives exploit the single-instruction-multiple-data (SIMD) instructions supported by modern CPUs. We evaluate our techniques in a database environment provided by Upscaledb, a production-quality key-value database. Our best techniques are SIMD accelerated: they simultaneously reduce memory usage while improving single-threaded speeds. In particular, a differentially coded SIMD binary-packing techniques (BP128) can offer a superior query speed (e.g., 40 better than an uncompressed database) while providing the best compression (e.g., by a factor of ten). For analytic workloads, our fast compression techniques offer compelling benefits. Our software is available as open source. | Random access in differentially-coded compressed arrays is often made possible with auxiliary data structures that allow skipping @cite_9 @cite_5 . However, there are alternatives to differential coding that offer more convenient random access. @cite_3 propose differentially encoded search trees; Teuhola @cite_19 adapts interpolative coding @cite_7 so that it can support logarithmic search. @cite_38 modify variable-byte encoding to create Directly Addressable Codes (DACs)---so that one can have access to individual coded value in constant time using rank select dictionaries. This strategy is applied to other compression schemes by K "ulekci @cite_44 . There has also been much interest in variations on the Elias-Fano representation @cite_39 @cite_37 , as it can provide good compression and fast random access to the encoded values. Other techniques such as wavelet trees @cite_2 or bitmap indexes @cite_4 @cite_22 @cite_31 @cite_21 can also be used for similar purposes. | {
"cite_N": [
"@cite_38",
"@cite_37",
"@cite_4",
"@cite_22",
"@cite_7",
"@cite_9",
"@cite_21",
"@cite_3",
"@cite_39",
"@cite_44",
"@cite_19",
"@cite_2",
"@cite_5",
"@cite_31"
],
"mid": [
"2157714561",
"",
"2303665562",
"1984614894",
"1559631118",
"1988679864",
"2131143419",
"2053518549",
"",
"2037892395",
"2093408358",
"2086536051",
"2125203709",
"2084965869"
],
"abstract": [
"We present a new variable-length encoding scheme for sequences of integers, Directly Addressable Codes (DACs), which enables direct access to any element of the encoded sequence without the need of any sampling method. Our proposal is a kind of implicit data structure that introduces synchronism in the encoded sequence without using asymptotically any extra space. We show some experiments demonstrating that the technique is not only simple, but also competitive in time and space with existing solutions in several applications, such as the representation of LCP arrays or high-order entropy-compressed sequences.",
"",
"Compressed bitmap indexes are used in databases and search engines. Many bitmap compression techniques have been proposed, almost all relying primarily on run-length encoding RLE. However, on unsorted data, we can get superior performance with a hybrid compression technique that uses both uncompressed bitmaps and packed arrays inside a two-level tree. An instance of this technique, Roaring, has recently been proposed. Due to its good performance, it has been adopted by several production platforms e.g., Apache Lucene, Apache Spark, Apache Kylin, and Druid. Yet there are cases where run-length-encoded bitmaps are smaller than the original Roaring bitmaps-typically when the data are sorted so that the bitmaps contain long compressible runs. To better handle these cases, we build a new Roaring hybrid that combines uncompressed bitmaps, packed arrays, and RLE-compressed segments. The result is a new Roaring format that compresses better. Overall, our new implementation of Roaring can be several times faster up to two orders of magnitude than the implementations of traditional RLE-based alternatives WAH, Concise, and EWAH while compressing better. We review the design choices and optimizations that make these good results possible. Copyright © 2016 John Wiley & Sons, Ltd.",
"Conjunctive Boolean queries are a key component of modern information retrieval systems, especially when Web-scale repositories are being searched. A conjunctive query q is equivalent to a vqv-way intersection over ordered sets of integers, where each set represents the documents containing one of the terms, and each integer in each set is an ordinal document identifier. As is the case with many computing applications, there is tension between the way in which the data is represented, and the ways in which it is to be manipulated. In particular, the sets representing index data for typical document collections are highly compressible, but are processed using random access techniques, meaning that methods for carrying out set intersections must be alert to issues to do with access patterns and data representation. Our purpose in this article is to explore these trade-offs, by investigating intersection techniques that make use of both uncompressed “integer” representations, as well as compressed arrangements. We also propose a simple hybrid method that provides both compact storage, and also faster intersection computations for conjunctive querying than is possible even with uncompressed representations.",
"Information retrieval systems contain large volumes of text, and currently have typical sizes into the gigabyte range. Inverted indexes are one important method for providing search facilities into these collections, but unless compressed require a great deal of space. In this paper we introduce a new method for compressing inverted indexes that yields excellent compression, fast decoding, and exploits clustering—the tendency for words to appear relatively frequently in some parts of the collection and infrequently in others. We also describe two other quite separate applications for the same compression method: representing the MTF list positions generated by the Burrows-Wheeler Block Sorting transformations and transmitting the codebook for semi-static block-based minimum-redundancy coding.",
"Inverted index data structures are the key to fast text search engines. We first investigate one of the predominant operation on inverted indexes, which asks for intersecting two sorted lists of document IDs of different lengths. We explore compression and performance of different inverted list data structures. In particular, we present Lookup, a new data structure that allows intersection in expected time linear in the smaller list. Based on this result, we present the algorithmic core of a full text data base that allows fast Boolean queries, phrase queries, and document reporting using less space than the input text. The system uses a carefully choreographed combination of classical data compression techniques and inverted-index-based search data structures. Our experiments show that inverted indexes are preferable over purely suffix-array-based techniques for in-memory (English) text search engines. A similar system is now running in practice in each core of the distributed data base engine TREX of SAP.",
"Sorted lists of integers are commonly used in inverted indexes and database systems. They are often compressed in memory. We can use the single-instruction, multiple data SIMD instructions available in common processors to boost the speed of integer compression schemes. Our S4-BP128-D4 scheme uses as little as 0.7CPU cycles per decoded 32-bit integer while still providing state-of-the-art compression. However, if the subsequent processing of the integers is slow, the effort spent on optimizing decompression speed can be wasted. To show that it does not have to be so, we 1 vectorize and optimize the intersection of posting lists; 2 introduce the SIMD GALLOPING algorithm. We exploit the fact that one SIMD instruction can compare four pairs of 32-bit integers at once. We experiment with two Text REtrieval Conference TREC text collections, GOV2 and ClueWeb09 category B, using logs from the TREC million-query track. We show that using only the SIMD instructions ubiquitous in all modern CPUs, our techniques for conjunctive queries can double the speed of a state-of-the-art approach. Copyright © 2015 John Wiley & Sons, Ltd.",
"Let X=x\"1,x\"2,...,x\"n be a sequence of non-decreasing integer values. Storing a compressed representation of X that supports access and search is a problem that occurs in many domains. The most common solution to this problem uses a linear list and encodes the differences between consecutive values with encodings that favor small numbers. This solution includes additional information (i.e. samples) to support efficient searching on the encoded values. We introduce a completely different alternative that achieves compression by encoding the differences in a search tree. Our proposal has many applications, such as the representation of posting lists, geographic data, sparse bitmaps, and compressed suffix arrays, to name just a few. The structure is practical and we provide an experimental evaluation to show that it is competitive with the existing techniques.",
"",
"We investigate the usage of the wavelet tree and the rank select-dictionary data structures on hybrid-structured variable-length codes, which represent an integer in the form of a unary code section followed by a binary section. We propose to handle unary and binary partitions as separate streams and create wavelet trees or R S dictionaries over the unary streams, which grants us the opportunity to directly access any codeword. Particularly concentrating on Elias and Rice schemes, we introduce several solutions that i) improve the compression significantly, and ii) provide random access in constant or logarithmic time. Experiments are conducted to compare the performances of the proposed codes against Elias Rice schemes and more recent state-of-the-art codings such as Simple9, PForDelta,DACs, and improved-AC techniques. We observed that the newly introduced methods outperform the original Elias Rice codecs by approximately 30 and the others by approximately 10 in terms of compression ratios. The methods described in this study are generic and may further be extended to some other hybrid structure (unary binary) variable-length codes as well.",
"Sequences of integers are common data types, occurring either as primary data or ancillary structures. The sizes of sequences can be large, making compression an interesting option. Effective compression presupposes variable-length coding, which destroys the regular alignment of values. Yet it would often be desirable to access only a small subset of the entries, either by position (ordinal number) or by content (element value), without having to decode most of the sequence from the start. Here such a random access technique for compressed integers is described, with the special feature that no auxiliary index is needed. The solution applies a method called interpolative coding, which is one of the most efficient non-statistical codes for integers. Indexing is avoided by address calculation guaranteeing sufficient space for codes even in the worst case. The additional redundancy, compared to regular interpolative coding, is only about 1bit per source integer for uniform distribution. The time complexity of random access is logarithmic with respect to the source size for both position-based and content-based retrieval. According to experiments, random access is faster than full decoding when the number of accessed integers is not more than approximately 0.75.n log\"2n for sequence length n. The tests also confirm that the method is quite competitive with other approaches to random access coding, suggested in the literature.",
"The wavelet tree is a versatile data structure that serves a number of purposes, from string processing to computational geometry. It can be regarded as a device that represents a sequence, a reordering, or a grid of points. In addition, its space adapts to various entropy measures of the data it encodes, enabling compressed representations. New competitive solutions to a number of problems, based on wavelet trees, are appearing every year. In this survey we give an overview of wavelet trees and the surprising number of applications in which we have found them useful: basic and weighted point grids, sets of rectangles, strings, permutations, binary relations, graphs, inverted indexes, document retrieval indexes, full-text indexes, XML indexes, and general numeric sequences.",
"Query-processing costs on large text databases are dominated by the need to retrieve and scan the inverted list of each query term. Retrieval time for inverted lists can be greatly reduced by the use of compression, but this adds to the CPU time required. Here we show that the CPU component of query response time for conjunctive Boolean queries and for informal ranked queries can be similarly reduced, at little cost in terms of storage, by the inclusion of an internal index in each compressed inverted list. This method has been applied in a retrieval system for a collection of nearly two million short documents. Our experimental results show that the self-indexing strategy adds less than 20 to the size of the compressed inverted file, which itself occupies less than 10 of the indexed text, yet can reduce processing time for Boolean queries of 5-10 terms to under one fifth of the previous cost. Similarly, ranked queries of 40-50 terms can be evaluated in as little as 25 of the previous time, with little or no loss of retrieval effectiveness.",
"This paper examines the space-time performance of in-memory conjunctive list intersection algorithms, as used in search engines, where integers represent document identifiers. We demonstrate that the combination of bitvectors, large skips, delta compressed lists and URL ordering produces superior results to using skips or bitvectors alone. We define semi-bitvectors, a new partial bitvector data structure that stores the front of the list using a bitvector and the remainder using skips and delta compression. To make it particularly effective, we propose that documents be ordered so as to skew the postings lists to have dense regions at the front. This can be accomplished by grouping documents by their size in a descending manner and then reordering within each group using URL ordering. In each list, the division point between bitvector and delta compression can occur at any group boundary. We explore the performance of semi-bitvectors using the GOV2 dataset for various numbers of groups, resulting in significant space-time improvements over existing approaches. Semi-bitvectors do not directly support ranking. Indeed, bitvectors are not believed to be useful for ranking based search systems, because frequencies and offsets cannot be included in their structure. To refute this belief, we propose several approaches to improve the performance of ranking-based search systems using bitvectors, and leave their verification for future work. These proposals suggest that bitvectors, and more particularly semi-bitvectors, warrant closer examination by the research community."
]
} |
1611.05428 | 2549141685 | Compression can sometimes improve performance by making more of the data available to the processors faster. We consider the compression of integer keys in a B+-tree index. For this purpose, systems such as IBM DB2 use variable-byte compression over differentially coded keys. We revisit this problem with various compression alternatives such as Google's VarIntGB, Binary Packing and Frame-of-Reference. In all cases, we describe algorithms that can operate directly on compressed data. Many of our alternatives exploit the single-instruction-multiple-data (SIMD) instructions supported by modern CPUs. We evaluate our techniques in a database environment provided by Upscaledb, a production-quality key-value database. Our best techniques are SIMD accelerated: they simultaneously reduce memory usage while improving single-threaded speeds. In particular, a differentially coded SIMD binary-packing techniques (BP128) can offer a superior query speed (e.g., 40 better than an uncompressed database) while providing the best compression (e.g., by a factor of ten). For analytic workloads, our fast compression techniques offer compelling benefits. Our software is available as open source. | Our work should be applicable to other B-trees and related data structures, i.e., Log-Structured Merge-Trees @cite_25 @cite_43 (LSM). | {
"cite_N": [
"@cite_43",
"@cite_25"
],
"mid": [
"1981420413",
"2068739275"
],
"abstract": [
"Bigtable is a distributed storage system for managing structured data that is designed to scale to a very large size: petabytes of data across thousands of commodity servers. Many projects at Google store data in Bigtable, including web indexing, Google Earth, and Google Finance. These applications place very different demands on Bigtable, both in terms of data size (from URLs to web pages to satellite imagery) and latency requirements (from backend bulk processing to real-time data serving). Despite these varied demands, Bigtable has successfully provided a flexible, high-performance solution for all of these Google products. In this article, we describe the simple data model provided by Bigtable, which gives clients dynamic control over data layout and format, and we describe the design and implementation of Bigtable.",
"High-performance transaction system applications typically insert rows in a History table to provide an activity trace; at the same time the transaction system generates log records for purposes of system recovery. Both types of generated information can benefit from efficient indexing. An example in a well-known setting is the TPC-A benchmark application, modified to support efficient queries on the history for account activity for specific accounts. This requires an index by account-id on the fast-growing History table. Unfortunately, standard disk-based index structures such as the B-tree will effectively double the I O cost of the transaction to maintain an index such as this in real time, increasing the total system cost up to fifty percent. Clearly a method for maintaining a real-time index at low cost is desirable. The log-structured mergetree (LSM-tree) is a disk-based data structure designed to provide low-cost indexing for a file experiencing a high rate of record inserts (and deletes) over an extended period. The LSM-tree uses an algorithm that defers and batches index changes, cascading the changes from a memory-based component through one or more disk components in an efficient manner reminiscent of merge sort. During this process all index values are continuously accessible to retrievals (aside from very short locking periods), either through the memory component or one of the disk components. The algorithm has greatly reduced disk arm movements compared to a traditional access methods such as B-trees, and will improve cost-performance in domains where disk arm costs for inserts with traditional access methods overwhelm storage media costs. The LSM-tree approach also generalizes to operations other than insert and delete. However, indexed finds requiring immediate response will lose I O efficiency in some cases, so the LSM-tree is most useful in applications where index inserts are more common than finds that retrieve the entries. This seems to be a common property for history tables and log files, for example. The conclusions of Sect. 6 compare the hybrid use of memory and disk components in the LSM-tree access method with the commonly understood advantage of the hybrid method to buffer disk pages in memory."
]
} |
1611.05335 | 2626262222 | A first-person camera, placed at a person's head, captures, which objects are important to the camera wearer. Most prior methods for this task learn to detect such important objects from the manually labeled first-person data in a supervised fashion. However, important objects are strongly related to the camera wearer's internal state such as his intentions and attention, and thus, only the person wearing the camera can provide the importance labels. Such a constraint makes the annotation process costly and limited in scalability. In this work, we show that we can detect important objects in first-person images without the supervision by the camera wearer or even third-person labelers. We formulate an important detection problem as an interplay between the 1) segmentation and 2) recognition agents. The segmentation agent first proposes a possible important object segmentation mask for each image, and then feeds it to the recognition agent, which learns to predict an important object mask using visual semantics and spatial features. We implement such an interplay between both agents via an alternating cross-pathway supervision scheme inside our proposed Visual-Spatial Network (VSN). Our VSN consists of spatial ("where") and visual ("what") pathways, one of which learns common visual semantics while the other focuses on the spatial location cues. Our unsupervised learning is accomplished via a cross-pathway supervision, where one pathway feeds its predictions to a segmentation agent, which proposes a candidate important object segmentation mask that is then used by the other pathway as a supervisory signal. We show our method's success on two different important object datasets, where our method achieves similar or better results as the supervised methods. | There have been a number of first-person methods that explored important object detection task either as a main task @cite_18 @cite_24 @cite_16 , or as an auxiliary task for an activity recognition @cite_20 @cite_9 @cite_4 @cite_6 or video summarization @cite_25 @cite_27 . The work in @cite_25 @cite_16 @cite_9 @cite_20 employ hand-crafted appearance features, egocentric and optical flow features to describe a first-person image, and then train a discriminative classifier to detect the regions that correspond to the important objects. The more recent work @cite_4 @cite_18 use FCNs @cite_1 to predict important objects end-to-end. Whereas the method in @cite_18 employs a two stream visual appearance and 3D network, the work in @cite_4 exploits the connection between the activities and objects and proposes a two stream appearance and optical flow network with a multi-loss objective function. | {
"cite_N": [
"@cite_18",
"@cite_4",
"@cite_9",
"@cite_1",
"@cite_6",
"@cite_24",
"@cite_27",
"@cite_16",
"@cite_25",
"@cite_20"
],
"mid": [
"2296893412",
"",
"1947050545",
"2952632681",
"2149276562",
"2160254296",
"2120645068",
"2031688197",
"2071711566",
""
],
"abstract": [
"Unlike traditional third-person cameras mounted on robots, a first-person camera, captures a person's visual sensorimotor object interactions from up close. In this paper, we study the tight interplay between our momentary visual attention and motor action with objects from a first-person camera. We propose a concept of action-objects---the objects that capture person's conscious visual (watching a TV) or tactile (taking a cup) interactions. Action-objects may be task-dependent but since many tasks share common person-object spatial configurations, action-objects exhibit a characteristic 3D spatial distance and orientation with respect to the person. We design a predictive model that detects action-objects using EgoNet, a joint two-stream network that holistically integrates visual appearance (RGB) and 3D spatial layout (depth and height) cues to predict per-pixel likelihood of action-objects. Our network also incorporates a first-person coordinate embedding, which is designed to learn a spatial distribution of the action-objects in the first-person data. We demonstrate EgoNet's predictive power, by showing that it consistently outperforms previous baseline approaches. Furthermore, EgoNet also exhibits a strong generalization ability, i.e., it predicts semantically meaningful objects in novel first-person datasets. Our method's ability to effectively detect action-objects could be used to improve robots' understanding of human-object interactions.",
"",
"We address the challenging problem of recognizing the camera wearer's actions from videos captured by an egocentric camera. Egocentric videos encode a rich set of signals regarding the camera wearer, including head movement, hand pose and gaze information. We propose to utilize these mid-level egocentric cues for egocentric action recognition. We present a novel set of egocentric features and show how they can be combined with motion and object features. The result is a compact representation with superior performance. In addition, we provide the first systematic evaluation of motion, object and egocentric cues in egocentric action recognition. Our benchmark leads to several surprising findings. These findings uncover the best practices for egocentric actions, with a significant performance boost over all previous state-of-the-art methods on three publicly available datasets.",
"Convolutional networks are powerful visual models that yield hierarchies of features. We show that convolutional networks by themselves, trained end-to-end, pixels-to-pixels, exceed the state-of-the-art in semantic segmentation. Our key insight is to build \"fully convolutional\" networks that take input of arbitrary size and produce correspondingly-sized output with efficient inference and learning. We define and detail the space of fully convolutional networks, explain their application to spatially dense prediction tasks, and draw connections to prior models. We adapt contemporary classification networks (AlexNet, 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 novel architecture that combines semantic information from a deep, coarse layer with appearance information from a shallow, fine layer to produce accurate and detailed segmentations. Our fully convolutional network achieves state-of-the-art segmentation of PASCAL VOC (20 relative improvement to 62.2 mean IU on 2012), NYUDv2, and SIFT Flow, while inference takes one third of a second for a typical image.",
"We present a method to analyze daily activities, such as meal preparation, using video from an egocentric camera. Our method performs inference about activities, actions, hands, and objects. Daily activities are a challenging domain for activity recognition which are well-suited to an egocentric approach. In contrast to previous activity recognition methods, our approach does not require pre-trained detectors for objects and hands. Instead we demonstrate the ability to learn a hierarchical model of an activity by exploiting the consistent appearance of objects, hands, and actions that results from the egocentric context. We show that joint modeling of activities, actions, and objects leads to superior performance in comparison to the case where they are considered independently. We introduce a novel representation of actions based on object-hand interactions and experimentally demonstrate the superior performance of our representation in comparison to standard activity representations such as bag of words.",
"In this work we introduce a novel approach to object categorization that incorporates two types of context-co-occurrence and relative location - with local appearance-based features. Our approach, named CoLA (for co-occurrence, location and appearance), uses a conditional random field (CRF) to maximize object label agreement according to both semantic and spatial relevance. We model relative location between objects using simple pairwise features. By vector quantizing this feature space, we learn a small set of prototypical spatial relationships directly from the data. We evaluate our results on two challenging datasets: PASCAL 2007 and MSRC. The results show that combining co-occurrence and spatial context improves accuracy in as many as half of the categories compared to using co-occurrence alone.",
"We present a video summarization approach that discovers the story of an egocentric video. Given a long input video, our method selects a short chain of video sub shots depicting the essential events. Inspired by work in text analysis that links news articles over time, we define a random-walk based metric of influence between sub shots that reflects how visual objects contribute to the progression of events. Using this influence metric, we define an objective for the optimal k-subs hot summary. Whereas traditional methods optimize a summary's diversity or representative ness, ours explicitly accounts for how one sub-event \"leads to\" another-which, critically, captures event connectivity beyond simple object co-occurrence. As a result, our summaries provide a better sense of story. We apply our approach to over 12 hours of daily activity video taken from 23 unique camera wearers, and systematically evaluate its quality compared to multiple baselines with 34 human subjects.",
"This paper addresses the problem of learning object models from egocentric video of household activities, using extremely weak supervision. For each activity sequence, we know only the names of the objects which are present within it, and have no other knowledge regarding the appearance or location of objects. The key to our approach is a robust, unsupervised bottom up segmentation method, which exploits the structure of the egocentric domain to partition each frame into hand, object, and background categories. By using Multiple Instance Learning to match object instances across sequences, we discover and localize object occurrences. Object representations are refined through transduction and object-level classifiers are trained. We demonstrate encouraging results in detecting novel object instances using models produced by weakly-supervised learning.",
"We present a video summarization approach for egocentric or \"wearable\" camera data. Given hours of video, the proposed method produces a compact storyboard summary of the camera wearer's day. In contrast to traditional keyframe selection techniques, the resulting summary focuses on the most important objects and people with which the camera wearer interacts. To accomplish this, we develop region cues indicative of high-level saliency in egocentric video--such as the nearness to hands, gaze, and frequency of occurrence--and learn a regressor to predict the relative importance of any new region based on these cues. Using these predictions and a simple form of temporal event detection, our method selects frames for the storyboard that reflect the key object-driven happenings. We adjust the compactness of the final summary given either an importance selection criterion or a length budget; for the latter, we design an efficient dynamic programming solution that accounts for importance, visual uniqueness, and temporal displacement. Critically, the approach is neither camera-wearer-specific nor object-specific; that means the learned importance metric need not be trained for a given user or context, and it can predict the importance of objects and people that have never been seen previously. Our results on two egocentric video datasets show the method's promise relative to existing techniques for saliency and summarization.",
""
]
} |
1611.05335 | 2626262222 | A first-person camera, placed at a person's head, captures, which objects are important to the camera wearer. Most prior methods for this task learn to detect such important objects from the manually labeled first-person data in a supervised fashion. However, important objects are strongly related to the camera wearer's internal state such as his intentions and attention, and thus, only the person wearing the camera can provide the importance labels. Such a constraint makes the annotation process costly and limited in scalability. In this work, we show that we can detect important objects in first-person images without the supervision by the camera wearer or even third-person labelers. We formulate an important detection problem as an interplay between the 1) segmentation and 2) recognition agents. The segmentation agent first proposes a possible important object segmentation mask for each image, and then feeds it to the recognition agent, which learns to predict an important object mask using visual semantics and spatial features. We implement such an interplay between both agents via an alternating cross-pathway supervision scheme inside our proposed Visual-Spatial Network (VSN). Our VSN consists of spatial ("where") and visual ("what") pathways, one of which learns common visual semantics while the other focuses on the spatial location cues. Our unsupervised learning is accomplished via a cross-pathway supervision, where one pathway feeds its predictions to a segmentation agent, which proposes a candidate important object segmentation mask that is then used by the other pathway as a supervisory signal. We show our method's success on two different important object datasets, where our method achieves similar or better results as the supervised methods. | Recently, there have been several deep learning approaches that proposed learning with weakly labeled or unlabeled datasets @cite_13 @cite_26 @cite_12 @cite_21 @cite_5 @cite_15 @cite_11 @cite_10 @cite_3 . Due to the high cost of obtaining per-pixel labels, this has been a particularly relevant problem for semantic segmentation. The weakest form of supervision for semantic segmentation includes image-level labels, which were used to train FCNs in several prior approaches @cite_15 @cite_10 @cite_3 @cite_21 . Some recent work @cite_12 used point supervision, which requires almost as much effort as the image-level labels but also provides some spatial information. Several approaches employed free form squiggles as a supervisory signal @cite_5 @cite_11 which provides even more information, and are still easy enough to annotate. Furthermore, several approaches utilized bounding box level annotations for FCN training @cite_21 @cite_26 . Finally, recent work achieved excellent edge detection results without using any annotations at all @cite_13 . | {
"cite_N": [
"@cite_26",
"@cite_10",
"@cite_21",
"@cite_3",
"@cite_5",
"@cite_15",
"@cite_13",
"@cite_12",
"@cite_11"
],
"mid": [
"2949086864",
"1931270512",
"1529410181",
"2952004933",
"1927251054",
"1945608308",
"2204233249",
"",
""
],
"abstract": [
"Recent leading approaches to semantic segmentation rely on deep convolutional networks trained with human-annotated, pixel-level segmentation masks. Such pixel-accurate supervision demands expensive labeling effort and limits the performance of deep networks that usually benefit from more training data. In this paper, we propose a method that achieves competitive accuracy but only requires easily obtained bounding box annotations. The basic idea is to iterate between automatically generating region proposals and training convolutional networks. These two steps gradually recover segmentation masks for improving the networks, and vise versa. Our method, called BoxSup, produces competitive results supervised by boxes only, on par with strong baselines fully supervised by masks under the same setting. By leveraging a large amount of bounding boxes, BoxSup further unleashes the power of deep convolutional networks and yields state-of-the-art results on PASCAL VOC 2012 and PASCAL-CONTEXT.",
"Multiple instance learning (MIL) can reduce the need for costly annotation in tasks such as semantic segmentation by weakening the required degree of supervision. We propose a novel MIL formulation of multi-class semantic segmentation learning by a fully convolutional network. In this setting, we seek to learn a semantic segmentation model from just weak image-level labels. The model is trained end-to-end to jointly optimize the representation while disambiguating the pixel-image label assignment. Fully convolutional training accepts inputs of any size, does not need object proposal pre-processing, and offers a pixelwise loss map for selecting latent instances. Our multi-class MIL loss exploits the further supervision given by images with multiple labels. We evaluate this approach through preliminary experiments on the PASCAL VOC segmentation challenge.",
"Deep convolutional neural networks (DCNNs) trained on a large number of images with strong pixel-level annotations have recently significantly pushed the state-of-art in semantic image segmentation. We study the more challenging problem of learning DCNNs for semantic image segmentation from either (1) weakly annotated training data such as bounding boxes or image-level labels or (2) a combination of few strongly labeled and many weakly labeled images, sourced from one or multiple datasets. We develop Expectation-Maximization (EM) methods for semantic image segmentation model training under these weakly supervised and semi-supervised settings. Extensive experimental evaluation shows that the proposed techniques can learn models delivering competitive results on the challenging PASCAL VOC 2012 image segmentation benchmark, while requiring significantly less annotation effort. We share source code implementing the proposed system at this https URL",
"We present an approach to learn a dense pixel-wise labeling from image-level tags. Each image-level tag imposes constraints on the output labeling of a Convolutional Neural Network (CNN) classifier. We propose Constrained CNN (CCNN), a method which uses a novel loss function to optimize for any set of linear constraints on the output space (i.e. predicted label distribution) of a CNN. Our loss formulation is easy to optimize and can be incorporated directly into standard stochastic gradient descent optimization. The key idea is to phrase the training objective as a biconvex optimization for linear models, which we then relax to nonlinear deep networks. Extensive experiments demonstrate the generality of our new learning framework. The constrained loss yields state-of-the-art results on weakly supervised semantic image segmentation. We further demonstrate that adding slightly more supervision can greatly improve the performance of the learning algorithm.",
"Despite the promising performance of conventional fully supervised algorithms, semantic segmentation has remained an important, yet challenging task. Due to the limited availability of complete annotations, it is of great interest to design solutions for semantic segmentation that take into account weakly labeled data, which is readily available at a much larger scale. Contrasting the common theme to develop a different algorithm for each type of weak annotation, in this work, we propose a unified approach that incorporates various forms of weak supervision - image level tags, bounding boxes, and partial labels - to produce a pixel-wise labeling. We conduct a rigorous evaluation on the challenging Siftflow dataset for various weakly labeled settings, and show that our approach outperforms the state-of-the-art by 12 on per-class accuracy, while maintaining comparable per-pixel accuracy.",
"We are interested in inferring object segmentation by leveraging only object class information, and by considering only minimal priors on the object segmentation task. This problem could be viewed as a kind of weakly supervised segmentation task, and naturally fits the Multiple Instance Learning (MIL) framework: every training image is known to have (or not) at least one pixel corresponding to the image class label, and the segmentation task can be rewritten as inferring the pixels belonging to the class of the object (given one image, and its object class). We propose a Convolutional Neural Network-based model, which is constrained during training to put more weight on pixels which are important for classifying the image. We show that at test time, the model has learned to discriminate the right pixels well enough, such that it performs very well on an existing segmentation benchmark, by adding only few smoothing priors. Our system is trained using a subset of the Imagenet dataset and the segmentation experiments are performed on the challenging Pascal VOC dataset (with no fine-tuning of the model on Pascal VOC). Our model beats the state of the art results in weakly supervised object segmentation task by a large margin. We also compare the performance of our model with state of the art fully-supervised segmentation approaches.",
"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.",
"",
""
]
} |
1611.05396 | 2951244631 | We present a new Cascaded Shape Regression (CSR) architecture, namely Dynamic Attention-Controlled CSR (DAC-CSR), for robust facial landmark detection on unconstrained faces. Our DAC-CSR divides facial landmark detection into three cascaded sub-tasks: face bounding box refinement, general CSR and attention-controlled CSR. The first two stages refine initial face bounding boxes and output intermediate facial landmarks. Then, an online dynamic model selection method is used to choose appropriate domain-specific CSRs for further landmark refinement. The key innovation of our DAC-CSR is the fault-tolerant mechanism, using fuzzy set sample weighting for attention-controlled domain-specific model training. Moreover, we advocate data augmentation with a simple but effective 2D profile face generator, and context-aware feature extraction for better facial feature representation. Experimental results obtained on challenging datasets demonstrate the merits of our DAC-CSR over the state-of-the-art. | Facial landmark detection can trace its history to the nineteen nineties. The representative FLD methods making the early milestones include Active Shape Model (ASM) @cite_7 , Active Appearance Model (AAM) @cite_12 and Constrained Local Model (CLM) @cite_29 . These algorithms and their extensions have achieved excellent FLD results in constrained scenarios @cite_10 . As a result, the current trend is to develop a more robust FLD for unconstrained faces that are rich in appearance variations. The leading algorithms for unconstrained FLD are CSR-based approaches @cite_34 @cite_41 @cite_23 @cite_13 @cite_27 . In contrast to the classical methods such as ASM, AAM and CLM that rely on a generative PCA-based shape model, CSR directly positions facial landmarks on their optimal locations based on image features. The shape update is achieved in a discriminative way by constructing a mapping function from robust shape-related local features to shape updates. The secret of the success of CSR is the architecture that cascades a set of weak regressors in series to form a strong regressor. | {
"cite_N": [
"@cite_13",
"@cite_7",
"@cite_41",
"@cite_29",
"@cite_27",
"@cite_23",
"@cite_34",
"@cite_10",
"@cite_12"
],
"mid": [
"1998294030",
"2073039128",
"2136000821",
"1977821862",
"2462523589",
"1832881114",
"1990937109",
"1572092093",
""
],
"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.",
"We describe ‘Active Shape Models’ which iteratively adapt to refine estimates of the pose, scale and shape of models of image objects. The method uses flexible models derived from sets of training examples. These models, known as Point Distribution Models, represent objects as sets of labelled points. An initial estimate of the location of the model points in an image is improved by attempting to move each point to a better position nearby. Adjustments to the pose variables and shape parameters are calculated. Limits are placed on the shape parameters ensuring that the example can only deform into shapes conforming to global constraints imposed by the training set. An iterative procedure deforms the model example to find the best fit to the image object. Results of applying the method are described. The technique is shown to be a powerful method for refining estimates of object shape and location.",
"We present a fast and accurate algorithm for computing the 2D pose of objects in images called cascaded pose regression (CPR). CPR progressively refines a loosely specified initial guess, where each refinement is carried out by a different regressor. Each regressor performs simple image measurements that are dependent on the output of the previous regressors; the entire system is automatically learned from human annotated training examples. CPR is not restricted to rigid transformations: ‘pose’ is any parameterized variation of the object's appearance such as the degrees of freedom of deformable and articulated objects. We compare CPR against both standard regression techniques and human performance (computed from redundant human annotations). Experiments on three diverse datasets (mice, faces, fish) suggest CPR is fast (2–3ms per pose estimate), accurate (approaching human performance), and easy to train from small amounts of labeled data.",
"We present an efficient and robust model matching method which uses a joint shape and texture appearance model to generate a set of region template detectors. The model is fitted to an unseen image in an iterative manner by generating templates using the joint model and the current parameter estimates, correlating the templates with the target image to generate response images and optimising the shape parameters so as to maximise the sum of responses. The appearance model is similar to that used in the Active Appearance Model due to However in our approach the appearance model is used to generate likely feature templates, instead of trying to approximate the image pixels directly. We show that when applied to human faces, our constrained local model (CLM) algorithm is more robust and more accurate than the original AAM search method, which relies on the image reconstruction error to update the model parameters. We demonstrate improved localisation accuracy on two publicly available face data sets and improved tracking on a challenging set of in-car face sequences.",
"We present a practical approach to address the problem of unconstrained face alignment for a single image. In our unconstrained problem, we need to deal with large shape and appearance variations under extreme head poses and rich shape deformation. To equip cascaded regressors with the capability to handle global shape variation and irregular appearance-shape relation in the unconstrained scenario, we partition the optimisation space into multiple domains of homogeneous descent, and predict a shape as a composition of estimations from multiple domain-specific regressors. With a specially formulated learning objective and a novel tree splitting function, our approach is capable of estimating a robust and meaningful composition. In addition to achieving state-of-the-art accuracy over existing approaches, our framework is also an efficient solution (350 FPS), thanks to the on-the-fly domain exclusion mechanism and the capability of leveraging the fast pixel feature.",
"A large amount of training data is usually crucial for successful supervised learning. However, the task of providing training samples is often time-consuming, involving a considerable amount of tedious manual work. In addition, the amount of training data available is often limited. As an alternative, in this paper, we discuss how best to augment the available data for the application of automatic facial landmark detection. We propose the use of a 3D morphable face model to generate synthesized faces for a regression-based detector training. Benefiting from the large synthetic training data, the learned detector is shown to exhibit a better capability to detect the landmarks of a face with pose variations. Furthermore, the synthesized training data set provides accurate and consistent landmarks automatically as compared to the landmarks annotated manually, especially for occluded facial parts. The synthetic data and real data are from different domains; hence the detector trained using only synthesized faces does not generalize well to real faces. To deal with this problem, we propose a cascaded collaborative regression algorithm, which generates a cascaded shape updater that has the ability to overcome the difficulties caused by pose variations, as well as achieving better accuracy when applied to real faces. The training is based on a mix of synthetic and real image data with the mixing controlled by a dynamic mixture weighting schedule. Initially, the training uses heavily the synthetic data, as this can model the gross variations between the various poses. As the training proceeds, progressively more of the natural images are incorporated, as these can model finer detail. To improve the performance of the proposed algorithm further, we designed a dynamic multi-scale local feature extraction method, which captures more informative local features for detector training. An extensive evaluation on both controlled and uncontrolled face data sets demonstrates the merit of the proposed algorithm.",
"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.",
"It has been shown that multilinear subspace analysis is a powerful tool to overcome difficulties posed by viewpoint, illumination and expression variations in Active Appearance Model(AAM). However, the Higher Order Singular Value Decomposition (HOSVD) in multilinear analysis requires training samples to build the training tensor, which include face images under all different variations. It is hard to obtain such a complete training tensor in practical applications. In this paper, we propose a multilinear AAM which can be generated from an incomplete training tensor using Multilinear Subspace Analysis with Missing Values (M2SA). Also, the 2D appearance is used for training appearance tensor directly to reduce the memory requirements. Experimental results on the Multi-PIE face database show the efficiency of the proposed method.",
""
]
} |
1611.05396 | 2951244631 | We present a new Cascaded Shape Regression (CSR) architecture, namely Dynamic Attention-Controlled CSR (DAC-CSR), for robust facial landmark detection on unconstrained faces. Our DAC-CSR divides facial landmark detection into three cascaded sub-tasks: face bounding box refinement, general CSR and attention-controlled CSR. The first two stages refine initial face bounding boxes and output intermediate facial landmarks. Then, an online dynamic model selection method is used to choose appropriate domain-specific CSRs for further landmark refinement. The key innovation of our DAC-CSR is the fault-tolerant mechanism, using fuzzy set sample weighting for attention-controlled domain-specific model training. Moreover, we advocate data augmentation with a simple but effective 2D profile face generator, and context-aware feature extraction for better facial feature representation. Experimental results obtained on challenging datasets demonstrate the merits of our DAC-CSR over the state-of-the-art. | There have been a number of improvements to the performance of CSR-based FLD. One category of these improvements is to enhance some components of the existing CSR architecture. For example, the use of more robust shape-related local features, Scale-Invariant Feature Transform (SIFT) @cite_43 @cite_37 @cite_1 , Histogram of Oriented Gradients (HOG) @cite_30 @cite_23 @cite_24 @cite_33 , Sparse Auto-Encoder (SAE) @cite_14 , Local Binary Features (LBF) @cite_34 @cite_13 and Convolutional Neural Networks (CNN-) based features @cite_6 @cite_44 , has been suggested. Another example is to use more powerful regression methods as weak regressors in CSR, such as random forests @cite_34 @cite_13 and deep neural networks @cite_21 @cite_6 @cite_44 @cite_37 @cite_1 @cite_31 . Lately, 3D face models have been shown to positively impact FLD in challenging benchmarking datasets, especially in relation to faces with extreme poses @cite_23 @cite_16 @cite_3 . | {
"cite_N": [
"@cite_30",
"@cite_37",
"@cite_14",
"@cite_33",
"@cite_21",
"@cite_1",
"@cite_6",
"@cite_16",
"@cite_24",
"@cite_43",
"@cite_44",
"@cite_3",
"@cite_23",
"@cite_31",
"@cite_34",
"@cite_13"
],
"mid": [
"",
"2437557374",
"2015229219",
"2090530238",
"1976948919",
"2166694921",
"2474575620",
"2520331172",
"1487748529",
"2157285372",
"2519753233",
"2964014798",
"1832881114",
"1896424170",
"1990937109",
"1998294030"
],
"abstract": [
"",
"Face alignment or facial landmark detection plays an important role in many computer vision applications, e.g., face recognition, facial expression recognition, face animation, etc. However, the performance of face alignment system degenerates severely when occlusions occur. In this work, we propose a novel face alignment method, which cascades several Deep Regression networks coupled with De-corrupt Autoencoders (denoted as DRDA) to explicitly handle partial occlusion problem. Different from the previous works that can only detect occlusions and discard the occluded parts, our proposed de-corrupt autoencoder network can automatically recover the genuine appearance for the occluded parts and the recovered parts can be leveraged together with those non-occluded parts for more accurate alignment. By coupling de-corrupt autoencoders with deep regression networks, a deep alignment model robust to partial occlusions is achieved. Besides, our method can localize occluded regions rather than merely predict whether the landmarks are occluded. Experiments on two challenging occluded face datasets demonstrate that our method significantly outperforms the state-of-the-art methods.",
"In this letter, we present a random cascaded-regression copse (R-CR-C) for robust facial landmark detection. Its key innovations include a new parallel cascade structure design, and an adaptive scheme for scale-invariant shape update and local feature extraction. Evaluation on two challenging benchmarks shows the superiority of the proposed algorithm to state-of-the-art methods. © 1994-2012 IEEE.",
"In this paper, we present the details of our method in attending the 300 Faces in-the-wild (300W) challenge. We build our method on cascade regression framework, where a series of regressors are utilized to progressively refine the shape initialized by face detector. In cascade regression, we use the HOG feature in a multi-scale manner, where the large pose validation is handled in early stages by HOG feature at large scale, and then shape is refined at later stages with HOG feature at small scale. We observe that the performance of the cascade regression method decreases when the initialization provided by face detector is not accurate enough (for faces with large appearance variations, face detection is still a challenging problem). To handle the problem, we propose to generate multiple hypotheses, and then learn to rank or combine these hypotheses to get the final result. The parameters in both learn to rank and learn to combine can be learned in a structural SVM framework. Despite the simplicity of our method, it achieves state-of-the-art performance on LFPW, and dramatically outperforms the baseline AAM on the 300-W challenge.",
"We propose a new approach for estimation of the positions of facial key points with three-level carefully designed convolutional networks. At each level, the outputs of multiple networks are fused for robust and accurate estimation. Thanks to the deep structures of convolutional networks, global high-level features are extracted over the whole face region at the initialization stage, which help to locate high accuracy key points. There are two folds of advantage for this. First, the texture context information over the entire face is utilized to locate each key point. Second, since the networks are trained to predict all the key points simultaneously, the geometric constraints among key points are implicitly encoded. The method therefore can avoid local minimum caused by ambiguity and data corruption in difficult image samples due to occlusions, large pose variations, and extreme lightings. The networks at the following two levels are trained to locally refine initial predictions and their inputs are limited to small regions around the initial predictions. Several network structures critical for accurate and robust facial point detection are investigated. Extensive experiments show that our approach outperforms state-of-the-art methods in both detection accuracy and reliability.",
"Accurate face alignment is a vital prerequisite step for most face perception tasks such as face recognition, facial expression analysis and non-realistic face re-rendering. It can be formulated as the nonlinear inference of the facial landmarks from the detected face region. Deep network seems a good choice to model the nonlinearity, but it is nontrivial to apply it directly. In this paper, instead of a straightforward application of deep network, we propose a Coarse-to-Fine Auto-encoder Networks (CFAN) approach, which cascades a few successive Stacked Auto-encoder Networks (SANs). Specifically, the first SAN predicts the landmarks quickly but accurately enough as a preliminary, by taking as input a low-resolution version of the detected face holistically. The following SANs then progressively refine the landmark by taking as input the local features extracted around the current landmarks (output of the previous SAN) with higher and higher resolution. Extensive experiments conducted on three challenging datasets demonstrate that our CFAN outperforms the state-of-the-art methods and performs in real-time(40+fps excluding face detection on a desktop).",
"Cascaded regression has recently become the method of choice for solving non-linear least squares problems such as deformable image alignment. Given a sizeable training set, cascaded regression learns a set of generic rules that are sequentially applied to minimise the least squares problem. Despite the success of cascaded regression for problems such as face alignment and head pose estimation, there are several shortcomings arising in the strategies proposed thus far. Specifically, (a) the regressors are learnt independently, (b) the descent directions may cancel one another out and (c) handcrafted features (e.g., HoGs, SIFT etc.) are mainly used to drive the cascade, which may be sub-optimal for the task at hand. In this paper, we propose a combined and jointly trained convolutional recurrent neural network architecture that allows the training of an end-to-end to system that attempts to alleviate the aforementioned drawbacks. The recurrent module facilitates the joint optimisation of the regressors by assuming the cascades form a nonlinear dynamical system, in effect fully utilising the information between all cascade levels by introducing a memory unit that shares information across all levels. The convolutional module allows the network to extract features that are specialised for the task at hand and are experimentally shown to outperform hand-crafted features. We show that the application of the proposed architecture for the problem of face alignment results in a strong improvement over the current state-of-the-art.",
"We present an approach to simultaneously solve the two problems of face alignment and 3D face reconstruction from an input 2D face image of arbitrary poses and expressions. The proposed method iteratively and alternately applies two sets of cascaded regressors, one for updating 2D landmarks and the other for updating reconstructed pose-expression-normalized (PEN) 3D face shape. The 3D face shape and the landmarks are correlated via a 3D-to-2D mapping matrix. In each iteration, adjustment to the landmarks is firstly estimated via a landmark regressor, and this landmark adjustment is also used to estimate 3D face shape adjustment via a shape regressor. The 3D-to-2D mapping is then computed based on the adjusted 3D face shape and 2D landmarks, and it further refines the 2D landmarks. An effective algorithm is devised to learn these regressors based on a training dataset of pairing annotated 3D face shapes and 2D face images. Compared with existing methods, the proposed method can fully automatically generate PEN 3D face shapes in real time from a single 2D face image and locate both visible and invisible 2D landmarks. Extensive experiments show that the proposed method can achieve the state-of-the-art accuracy in both face alignment and 3D face reconstruction, and benefit face recognition owing to its reconstructed PEN 3D face shapes.",
"In this paper, we propose a novel fitting method that uses local image features to fit a 3D Morphable Face Model to 2D images. To overcome the obstacle of optimising a cost function that contains a non-differentiable feature extraction operator, we use a learning-based cascaded regression method that learns the gradient direction from data. The method allows to simultaneously solve for shape and pose parameters. Our method is thoroughly evaluated on Morphable Model generated data and first results on real data are presented. Compared to traditional fitting methods, which use simple raw features like pixel colour or edge maps, local features have been shown to be much more robust against variations in imaging conditions. Our approach is unique in that we are the first to use local features to fit a 3D Morphable Model. Because of the speed of our method, it is applicable for real-time applications. Our cascaded regression framework is available as an open source library at github.com patrikhuber superviseddescent.",
"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.",
"In this work, we introduce a novel Recurrent Attentive-Refinement (RAR) network for facial landmark detection under unconstrained conditions, suffering from challenges like facial occlusions and or pose variations. RAR follows the pipeline of cascaded regressions that refines landmark locations progressively. However, instead of updating all the landmark locations together, RAR refines the landmark locations sequentially at each recurrent stage. In this way, more reliable landmark points are refined earlier and help to infer locations of other challenging landmarks that may stay with occlusions and or extreme poses. RAR can thus effectively control detection errors from those challenging landmarks and improve overall performance even in presence of heavy occlusions and or extreme conditions. To determine the sequence of landmarks, RAR employs an attentive-refinement mechanism. The attention LSTM (A-LSTM) and refinement LSTM (R-LSTM) models are introduced in RAR. At each recurrent stage, A-LSTM implicitly identifies a reliable landmark as the attention center. Following the sequence of attention centers, R-LSTM sequentially refines the landmarks near or correlated with the attention centers and provides ultimate detection results finally. To further enhance algorithmic robustness, instead of using mean shape for initialization, RAR adaptively determines the initialization by selecting from a pool of shape centers clustered from all training shapes. As an end-to-end trainable model, RAR demonstrates superior performance in detecting challenging landmarks in comprehensive experiments and it also establishes new state-of-the-arts on the 300-W, COFW and AFLW benchmark datasets.",
"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.",
"A large amount of training data is usually crucial for successful supervised learning. However, the task of providing training samples is often time-consuming, involving a considerable amount of tedious manual work. In addition, the amount of training data available is often limited. As an alternative, in this paper, we discuss how best to augment the available data for the application of automatic facial landmark detection. We propose the use of a 3D morphable face model to generate synthesized faces for a regression-based detector training. Benefiting from the large synthetic training data, the learned detector is shown to exhibit a better capability to detect the landmarks of a face with pose variations. Furthermore, the synthesized training data set provides accurate and consistent landmarks automatically as compared to the landmarks annotated manually, especially for occluded facial parts. The synthetic data and real data are from different domains; hence the detector trained using only synthesized faces does not generalize well to real faces. To deal with this problem, we propose a cascaded collaborative regression algorithm, which generates a cascaded shape updater that has the ability to overcome the difficulties caused by pose variations, as well as achieving better accuracy when applied to real faces. The training is based on a mix of synthetic and real image data with the mixing controlled by a dynamic mixture weighting schedule. Initially, the training uses heavily the synthetic data, as this can model the gross variations between the various poses. As the training proceeds, progressively more of the natural images are incorporated, as these can model finer detail. To improve the performance of the proposed algorithm further, we designed a dynamic multi-scale local feature extraction method, which captures more informative local features for detector training. An extensive evaluation on both controlled and uncontrolled face data sets demonstrates the merit of the proposed algorithm.",
"Facial landmark detection has long been impeded by the problems of occlusion and pose variation. Instead of treating the detection task as a single and independent problem, we investigate the possibility of improving detection robustness through multi-task learning. Specifically, we wish to optimize facial landmark detection together with heterogeneous but subtly correlated tasks, e.g. head pose estimation and facial attribute inference. This is non-trivial since different tasks have different learning difficulties and convergence rates. To address this problem, we formulate a novel tasks-constrained deep model, with task-wise early stopping to facilitate learning convergence. Extensive evaluations show that the proposed task-constrained learning (i) outperforms existing methods, especially in dealing with faces with severe occlusion and pose variation, and (ii) reduces model complexity drastically compared to the state-of-the-art method based on cascaded deep model [21].",
"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.",
"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."
]
} |
1611.05396 | 2951244631 | We present a new Cascaded Shape Regression (CSR) architecture, namely Dynamic Attention-Controlled CSR (DAC-CSR), for robust facial landmark detection on unconstrained faces. Our DAC-CSR divides facial landmark detection into three cascaded sub-tasks: face bounding box refinement, general CSR and attention-controlled CSR. The first two stages refine initial face bounding boxes and output intermediate facial landmarks. Then, an online dynamic model selection method is used to choose appropriate domain-specific CSRs for further landmark refinement. The key innovation of our DAC-CSR is the fault-tolerant mechanism, using fuzzy set sample weighting for attention-controlled domain-specific model training. Moreover, we advocate data augmentation with a simple but effective 2D profile face generator, and context-aware feature extraction for better facial feature representation. Experimental results obtained on challenging datasets demonstrate the merits of our DAC-CSR over the state-of-the-art. | Another important approach is to adopt advanced CSR architectures, such as the use of multiple CSR models or constructing multi-view models. Feng al @cite_14 constructed multiple CSR models by randomly selecting subsets from the original training set and fusing multiple outputs to produce the final FLD result. A similar idea has also been used in @cite_39 . As an alternative, a multi-view FLD system employs a set of view-specific models that are able to achieve more accurate landmark detection for faces exhibiting specific views @cite_0 @cite_19 @cite_27 . | {
"cite_N": [
"@cite_14",
"@cite_39",
"@cite_0",
"@cite_19",
"@cite_27"
],
"mid": [
"2015229219",
"",
"2172803778",
"2241943627",
"2462523589"
],
"abstract": [
"In this letter, we present a random cascaded-regression copse (R-CR-C) for robust facial landmark detection. Its key innovations include a new parallel cascade structure design, and an adaptive scheme for scale-invariant shape update and local feature extraction. Evaluation on two challenging benchmarks shows the superiority of the proposed algorithm to state-of-the-art methods. © 1994-2012 IEEE.",
"",
"We demonstrate a novel method of interpreting images using an Active Appearance Model (AAM). An AAM contains a statistical model of the shape and grey-level appearance of the object of interest which can generalise to almost any valid example. During a training phase we learn the relationship between model parameter displacements and the residual errors induced between a training image and a synthesised model example. To match to an image we measure the current residuals and use the model to predict changes to the current parameters, leading to a better fit. A good overall match is obtained in a few iterations, even from poor starting estimates. We describe the technique in detail and give results of quantitative performance tests. We anticipate that the AAM algorithm will be an important method for locating deformable objects in many applications.",
"Face alignment, which is the task of finding the locations of a set of facial landmark points in an image of a face, is useful in widespread application areas. Face alignment is particularly challenging when there are large variations in pose (in-plane and out-of-plane rotations) and facial expression. To address this issue, we propose a cascade in which each stage consists of a mixture of regression experts. Each expert learns a customized regression model that is specialized to a different subset of the joint space of pose and expressions. The system is invariant to a predefined class of transformations (e.g., affine), because the input is transformed to match each expert’s prototype shape before the regression is applied. We also present a method to include deformation constraints within the discriminative alignment framework, which makes our algorithm more robust. Our algorithm significantly outperforms previous methods on publicly available face alignment datasets.",
"We present a practical approach to address the problem of unconstrained face alignment for a single image. In our unconstrained problem, we need to deal with large shape and appearance variations under extreme head poses and rich shape deformation. To equip cascaded regressors with the capability to handle global shape variation and irregular appearance-shape relation in the unconstrained scenario, we partition the optimisation space into multiple domains of homogeneous descent, and predict a shape as a composition of estimations from multiple domain-specific regressors. With a specially formulated learning objective and a novel tree splitting function, our approach is capable of estimating a robust and meaningful composition. In addition to achieving state-of-the-art accuracy over existing approaches, our framework is also an efficient solution (350 FPS), thanks to the on-the-fly domain exclusion mechanism and the capability of leveraging the fast pixel feature."
]
} |
1611.05244 | 2953350812 | Person re-identification (Re-ID) poses a unique challenge to deep learning: how to learn a deep model with millions of parameters on a small training set of few or no labels. In this paper, a number of deep transfer learning models are proposed to address the data sparsity problem. First, a deep network architecture is designed which differs from existing deep Re-ID models in that (a) it is more suitable for transferring representations learned from large image classification datasets, and (b) classification loss and verification loss are combined, each of which adopts a different dropout strategy. Second, a two-stepped fine-tuning strategy is developed to transfer knowledge from auxiliary datasets. Third, given an unlabelled Re-ID dataset, a novel unsupervised deep transfer learning model is developed based on co-training. The proposed models outperform the state-of-the-art deep Re-ID models by large margins: we achieve Rank-1 accuracy of 85.4 , 83.7 and 56.3 on CUHK03, Market1501, and VIPeR respectively, whilst on VIPeR, our unsupervised model (45.1 ) beats most supervised models. | Dropout @cite_68 is a widely adopted technique in deep learning to counter overfitting, a problem that is particularly acute in Re-ID due to the small data size. Given the two losses, we propose to use different dropout strategies for each loss-associated layers. Specifically, the standard random dropout @cite_68 is deployed for the classification loss layers, whilst for the pairwise verification loss layers, we introduce pairwise-consistent dropout, that is, each pair of compared training data points share the same dropout mask. We show experimentally that such a modification can bring about 3 | {
"cite_N": [
"@cite_68"
],
"mid": [
"2095705004"
],
"abstract": [
"Deep neural nets with a large number of parameters are very powerful machine learning systems. However, overfitting is a serious problem in such networks. Large networks are also slow to use, making it difficult to deal with overfitting by combining the predictions of many different large neural nets at test time. Dropout is a technique for addressing this problem. The key idea is to randomly drop units (along with their connections) from the neural network during training. This prevents units from co-adapting too much. During training, dropout samples from an exponential number of different \"thinned\" networks. At test time, it is easy to approximate the effect of averaging the predictions of all these thinned networks by simply using a single unthinned network that has smaller weights. This significantly reduces overfitting and gives major improvements over other regularization methods. We show that dropout improves the performance of neural networks on supervised learning tasks in vision, speech recognition, document classification and computational biology, obtaining state-of-the-art results on many benchmark data sets."
]
} |
1611.05244 | 2953350812 | Person re-identification (Re-ID) poses a unique challenge to deep learning: how to learn a deep model with millions of parameters on a small training set of few or no labels. In this paper, a number of deep transfer learning models are proposed to address the data sparsity problem. First, a deep network architecture is designed which differs from existing deep Re-ID models in that (a) it is more suitable for transferring representations learned from large image classification datasets, and (b) classification loss and verification loss are combined, each of which adopts a different dropout strategy. Second, a two-stepped fine-tuning strategy is developed to transfer knowledge from auxiliary datasets. Third, given an unlabelled Re-ID dataset, a novel unsupervised deep transfer learning model is developed based on co-training. The proposed models outperform the state-of-the-art deep Re-ID models by large margins: we achieve Rank-1 accuracy of 85.4 , 83.7 and 56.3 on CUHK03, Market1501, and VIPeR respectively, whilst on VIPeR, our unsupervised model (45.1 ) beats most supervised models. | Transfer learning or domain adaptation is an extensively studied topic @cite_61 . Transfer learning is widely used for deep learning when a target task is short of labelled data. The most common deep transfer learning strategy is fine-tuning @cite_14 : first train a base network using a large source data and then copy the first @math layers to the corresponding layers of the target network, followed by randomly initialising the remaining layers and finally fine-tune only them or all layers. A systematic study is presented in @cite_14 which examines how transferable features of different layers are between the source and target domains. It concludes that the generalisation ability diminishes when the discrepancy between the source and target tasks increases. Note that the source and target tasks considered in @cite_14 were classifying different subsets of ImageNet, so the task domain discrepancy studied is nowhere near as big as in our ImageNet @math Re-ID transfer setting. As a result, the conventional one-stepped fine-tuning strategy becomes inadequate. To overcome the large task discrepancy between classification and verification, we propose a two-stepped fine-tuning strategy whereby the network is first fine-tuned with the classification loss only, followed by fine-tuning with both classification and verification losses. | {
"cite_N": [
"@cite_61",
"@cite_14"
],
"mid": [
"2165698076",
"2949667497"
],
"abstract": [
"A major assumption in many machine learning and data mining algorithms is that the training and future data must be in the same feature space and have the same distribution. However, in many real-world applications, this assumption may not hold. For example, we sometimes have a classification task in one domain of interest, but we only have sufficient training data in another domain of interest, where the latter data may be in a different feature space or follow a different data distribution. In such cases, knowledge transfer, if done successfully, would greatly improve the performance of learning by avoiding much expensive data-labeling efforts. In recent years, transfer learning has emerged as a new learning framework to address this problem. This survey focuses on categorizing and reviewing the current progress on transfer learning for classification, regression, and clustering problems. In this survey, we discuss the relationship between transfer learning and other related machine learning techniques such as domain adaptation, multitask learning and sample selection bias, as well as covariate shift. We also explore some potential future issues in transfer learning research.",
"Many deep neural networks trained on natural images exhibit a curious phenomenon in common: on the first layer they learn features similar to Gabor filters and color blobs. Such first-layer features appear not to be specific to a particular dataset or task, but general in that they are applicable to many datasets and tasks. Features must eventually transition from general to specific by the last layer of the network, but this transition has not been studied extensively. In this paper we experimentally quantify the generality versus specificity of neurons in each layer of a deep convolutional neural network and report a few surprising results. Transferability is negatively affected by two distinct issues: (1) the specialization of higher layer neurons to their original task at the expense of performance on the target task, which was expected, and (2) optimization difficulties related to splitting networks between co-adapted neurons, which was not expected. In an example network trained on ImageNet, we demonstrate that either of these two issues may dominate, depending on whether features are transferred from the bottom, middle, or top of the network. We also document that the transferability of features decreases as the distance between the base task and target task increases, but that transferring features even from distant tasks can be better than using random features. A final surprising result is that initializing a network with transferred features from almost any number of layers can produce a boost to generalization that lingers even after fine-tuning to the target dataset."
]
} |
1611.05244 | 2953350812 | Person re-identification (Re-ID) poses a unique challenge to deep learning: how to learn a deep model with millions of parameters on a small training set of few or no labels. In this paper, a number of deep transfer learning models are proposed to address the data sparsity problem. First, a deep network architecture is designed which differs from existing deep Re-ID models in that (a) it is more suitable for transferring representations learned from large image classification datasets, and (b) classification loss and verification loss are combined, each of which adopts a different dropout strategy. Second, a two-stepped fine-tuning strategy is developed to transfer knowledge from auxiliary datasets. Third, given an unlabelled Re-ID dataset, a novel unsupervised deep transfer learning model is developed based on co-training. The proposed models outperform the state-of-the-art deep Re-ID models by large margins: we achieve Rank-1 accuracy of 85.4 , 83.7 and 56.3 on CUHK03, Market1501, and VIPeR respectively, whilst on VIPeR, our unsupervised model (45.1 ) beats most supervised models. | Note that beyond fine-tuning, several recent works take a multi-task joint training approach @cite_25 @cite_36 @cite_65 @cite_63 @cite_30 @cite_64 , one of which is designed specifically for Re-ID @cite_64 . Most of them aim to minimise the discrepancy between the marginal @cite_25 @cite_36 @cite_65 or joint @cite_63 distributions of the source and target data, e.g., by introducing a cross-domain loss that is designed to blur the domain boundary @cite_36 . However, these works assume that the tasks are the same or similar in the two domains, e.g., classifying the same object classes shared by two datasets. They are thus unsuitable when the source and target domains have completely different tasks, in our case object categorisation in ImageNet and person matching in Re-ID -- aligning the data distributions of the two datasets would not make any sense. The joint learning + multi-task learning + fine-tuning based deep Re-ID model in @cite_64 is clearly not suitable for transferring from ImageNet to Re-ID with the different source and target tasks. | {
"cite_N": [
"@cite_30",
"@cite_64",
"@cite_36",
"@cite_65",
"@cite_63",
"@cite_25"
],
"mid": [
"2950361018",
"2342611082",
"1882958252",
"174360075",
"2408201877",
"2951670162"
],
"abstract": [
"The recent success of deep neural networks relies on massive amounts of labeled data. For a target task where labeled data is unavailable, domain adaptation can transfer a learner from a different source domain. In this paper, we propose a new approach to domain adaptation in deep networks that can simultaneously learn adaptive classifiers and transferable features from labeled data in the source domain and unlabeled data in the target domain. We relax a shared-classifier assumption made by previous methods and assume that the source classifier and target classifier differ by a residual function. We enable classifier adaptation by plugging several layers into the deep network to explicitly learn the residual function with reference to the target classifier. We embed features of multiple layers into reproducing kernel Hilbert spaces (RKHSs) and match feature distributions for feature adaptation. The adaptation behaviors can be achieved in most feed-forward models by extending them with new residual layers and loss functions, which can be trained efficiently using standard back-propagation. Empirical evidence exhibits that the approach outperforms state of art methods on standard domain adaptation datasets.",
"Learning generic and robust feature representations with data from multiple domains for the same problem is of great value, especially for the problems that have multiple datasets but none of them are large enough to provide abundant data variations. In this work, we present a pipeline for learning deep feature representations from multiple domains with Convolutional Neural Networks (CNNs). When training a CNN with data from all the domains, some neurons learn representations shared across several domains, while some others are effective only for a specific one. Based on this important observation, we propose a Domain Guided Dropout algorithm to improve the feature learning procedure. Experiments show the effectiveness of our pipeline and the proposed algorithm. Our methods on the person re-identification problem outperform stateof-the-art methods on multiple datasets by large margins.",
"Top-performing deep architectures are trained on massive amounts of labeled data. In the absence of labeled data for a certain task, domain adaptation often provides an attractive option given that labeled data of similar nature but from a different domain (e.g. synthetic images) are available. Here, we propose a new approach to domain adaptation in deep architectures that can be trained on large amount of labeled data from the source domain and large amount of unlabeled data from the target domain (no labeled target-domain data is necessary). As the training progresses, the approach promotes the emergence of \"deep\" features that are (i) discriminative for the main learning task on the source domain and (ii) invariant with respect to the shift between the domains. We show that this adaptation behaviour can be achieved in almost any feed-forward model by augmenting it with few standard layers and a simple new gradient reversal layer. The resulting augmented architecture can be trained using standard backpropagation. Overall, the approach can be implemented with little effort using any of the deep-learning packages. The method performs very well in a series of image classification experiments, achieving adaptation effect in the presence of big domain shifts and outperforming previous state-of-the-art on Office datasets.",
"Domain adaptation aims at training a classifier in one dataset and applying it to a related but not identical dataset. One successfully used framework of domain adaptation is to learn a transformation to match both the distribution of the features (marginal distribution), and the distribution of the labels given features (conditional distribution). In this paper, we propose a new domain adaptation framework named Deep Transfer Network (DTN), where the highly flexible deep neural networks are used to implement such a distribution matching process. This is achieved by two types of layers in DTN: the shared feature extraction layers which learn a shared feature subspace in which the marginal distributions of the source and the target samples are drawn close, and the discrimination layers which match conditional distributions by classifier transduction. We also show that DTN has a computation complexity linear to the number of training samples, making it suitable to large-scale problems. By combining the best paradigms in both worlds (deep neural networks in recognition, and matching marginal and conditional distributions in domain adaptation), we demonstrate by extensive experiments that DTN improves significantly over former methods in both execution time and classification accuracy.",
"Deep networks have been successfully applied to learn transferable features for adapting models from a source domain to a different target domain. In this paper, we present joint adaptation networks (JAN), which learn a transfer network by aligning the joint distributions of multiple domain-specific layers across domains based on a joint maximum mean discrepancy (JMMD) criterion. Adversarial training strategy is adopted to maximize JMMD such that the distributions of the source and target domains are made more distinguishable. Learning can be performed by stochastic gradient descent with the gradients computed by back-propagation in linear-time. Experiments testify that our model yields state of the art results on standard datasets.",
"Recent studies reveal that a deep neural network can learn transferable features which generalize well to novel tasks for domain adaptation. However, as deep features eventually transition from general to specific along the network, the feature transferability drops significantly in higher layers with increasing domain discrepancy. Hence, it is important to formally reduce the dataset bias and enhance the transferability in task-specific layers. In this paper, we propose a new Deep Adaptation Network (DAN) architecture, which generalizes deep convolutional neural network to the domain adaptation scenario. In DAN, hidden representations of all task-specific layers are embedded in a reproducing kernel Hilbert space where the mean embeddings of different domain distributions can be explicitly matched. The domain discrepancy is further reduced using an optimal multi-kernel selection method for mean embedding matching. DAN can learn transferable features with statistical guarantees, and can scale linearly by unbiased estimate of kernel embedding. Extensive empirical evidence shows that the proposed architecture yields state-of-the-art image classification error rates on standard domain adaptation benchmarks."
]
} |
1611.05244 | 2953350812 | Person re-identification (Re-ID) poses a unique challenge to deep learning: how to learn a deep model with millions of parameters on a small training set of few or no labels. In this paper, a number of deep transfer learning models are proposed to address the data sparsity problem. First, a deep network architecture is designed which differs from existing deep Re-ID models in that (a) it is more suitable for transferring representations learned from large image classification datasets, and (b) classification loss and verification loss are combined, each of which adopts a different dropout strategy. Second, a two-stepped fine-tuning strategy is developed to transfer knowledge from auxiliary datasets. Third, given an unlabelled Re-ID dataset, a novel unsupervised deep transfer learning model is developed based on co-training. The proposed models outperform the state-of-the-art deep Re-ID models by large margins: we achieve Rank-1 accuracy of 85.4 , 83.7 and 56.3 on CUHK03, Market1501, and VIPeR respectively, whilst on VIPeR, our unsupervised model (45.1 ) beats most supervised models. | In theory, any unsupervised deep learning methods can potentially be applied for domain adaptation when the first @math layers are pretrained on the source data. These include auto-encoder @cite_52 and dictionary learning @cite_4 which can be implemented as neural network layers and integrated as the later top layers of a CNN network @cite_20 . The main limitation of an unsupervised model is that it cannot learn discriminative features. Hence soft-label self-training based deep unsupervised learning has become popular recently @cite_29 . In this work, a novel co-training @cite_34 based unsupervised domain adaptation method is proposed to overcome the main drawback of self-training based methods, i.e., model drift @cite_44 . Combined with the proposed two-stepped fine-tuning method, this gives us a powerful deep unsupervised Re-ID model that outperforms not only alternative unsupervised models, but also most supervised models which use training labels. Recently a number of deep unsupervised transfer learning models are proposed @cite_36 @cite_65 @cite_30 . Nevertheless, the domain gap between different Re-ID datasets is significant and cannot be overcome by just aligning the data distributions, making them less effective than the proposed co-training based unsupervised learning method, as demonstrated in our experiments. | {
"cite_N": [
"@cite_30",
"@cite_4",
"@cite_36",
"@cite_29",
"@cite_65",
"@cite_52",
"@cite_44",
"@cite_34",
"@cite_20"
],
"mid": [
"2950361018",
"2142940228",
"1882958252",
"2519998487",
"174360075",
"2100495367",
"2020477327",
"2037603696",
""
],
"abstract": [
"The recent success of deep neural networks relies on massive amounts of labeled data. For a target task where labeled data is unavailable, domain adaptation can transfer a learner from a different source domain. In this paper, we propose a new approach to domain adaptation in deep networks that can simultaneously learn adaptive classifiers and transferable features from labeled data in the source domain and unlabeled data in the target domain. We relax a shared-classifier assumption made by previous methods and assume that the source classifier and target classifier differ by a residual function. We enable classifier adaptation by plugging several layers into the deep network to explicitly learn the residual function with reference to the target classifier. We embed features of multiple layers into reproducing kernel Hilbert spaces (RKHSs) and match feature distributions for feature adaptation. The adaptation behaviors can be achieved in most feed-forward models by extending them with new residual layers and loss functions, which can be trained efficiently using standard back-propagation. Empirical evidence exhibits that the approach outperforms state of art methods on standard domain adaptation datasets.",
"Algorithms for data-driven learning of domain-specific overcomplete dictionaries are developed to obtain maximum likelihood and maximum a posteriori dictionary estimates based on the use of Bayesian models with concave Schur-concave (CSC) negative log priors. Such priors are appropriate for obtaining sparse representations of environmental signals within an appropriately chosen (environmentally matched) dictionary. The elements of the dictionary can be interpreted as concepts, features, or words capable of succinct expression of events encountered in the environment (the source of the measured signals). This is a generalization of vector quantization in that one is interested in a description involving a few dictionary entries (the proverbial \"25 words or less\"), but not necessarily as succinct as one entry. To learn an environmentally adapted dictionary capable of concise expression of signals generated by the environment, we develop algorithms that iterate between a representative set of sparse representations found by variants of FOCUSS and an update of the dictionary using these sparse representations.Experiments were performed using synthetic data and natural images. For complete dictionaries, we demonstrate that our algorithms have improved performance over other independent component analysis (ICA) methods, measured in terms of signal-to-noise ratios of separated sources. In the overcomplete case, we show that the true underlying dictionary and sparse sources can be accurately recovered. In tests with natural images, learned overcomplete dictionaries are shown to have higher coding efficiency than complete dictionaries; that is, images encoded with an overcomplete dictionary have both higher compression (fewer bits per pixel) and higher accuracy (lower mean square error).",
"Top-performing deep architectures are trained on massive amounts of labeled data. In the absence of labeled data for a certain task, domain adaptation often provides an attractive option given that labeled data of similar nature but from a different domain (e.g. synthetic images) are available. Here, we propose a new approach to domain adaptation in deep architectures that can be trained on large amount of labeled data from the source domain and large amount of unlabeled data from the target domain (no labeled target-domain data is necessary). As the training progresses, the approach promotes the emergence of \"deep\" features that are (i) discriminative for the main learning task on the source domain and (ii) invariant with respect to the shift between the domains. We show that this adaptation behaviour can be achieved in almost any feed-forward model by augmenting it with few standard layers and a simple new gradient reversal layer. The resulting augmented architecture can be trained using standard backpropagation. Overall, the approach can be implemented with little effort using any of the deep-learning packages. The method performs very well in a series of image classification experiments, achieving adaptation effect in the presence of big domain shifts and outperforming previous state-of-the-art on Office datasets.",
"Attributes offer useful mid-level features to interpret visual data. While most attribute learning methods are supervised by costly human-generated labels, we introduce a simple yet powerful unsupervised approach to learn and predict visual attributes directly from data. Given a large unlabeled image collection as input, we train deep Convolutional Neural Networks (CNNs) to output a set of discriminative, binary attributes often with semantic meanings. Specifically, we first train a CNN coupled with unsupervised discriminative clustering, and then use the cluster membership as a soft supervision to discover shared attributes from the clusters while maximizing their separability. The learned attributes are shown to be capable of encoding rich imagery properties from both natural images and contour patches. The visual representations learned in this way are also transferrable to other tasks such as object detection. We show other convincing results on the related tasks of image retrieval and classification, and contour detection.",
"Domain adaptation aims at training a classifier in one dataset and applying it to a related but not identical dataset. One successfully used framework of domain adaptation is to learn a transformation to match both the distribution of the features (marginal distribution), and the distribution of the labels given features (conditional distribution). In this paper, we propose a new domain adaptation framework named Deep Transfer Network (DTN), where the highly flexible deep neural networks are used to implement such a distribution matching process. This is achieved by two types of layers in DTN: the shared feature extraction layers which learn a shared feature subspace in which the marginal distributions of the source and the target samples are drawn close, and the discrimination layers which match conditional distributions by classifier transduction. We also show that DTN has a computation complexity linear to the number of training samples, making it suitable to large-scale problems. By combining the best paradigms in both worlds (deep neural networks in recognition, and matching marginal and conditional distributions in domain adaptation), we demonstrate by extensive experiments that DTN improves significantly over former methods in both execution time and classification accuracy.",
"High-dimensional data can be converted to low-dimensional codes by training a multilayer neural network with a small central layer to reconstruct high-dimensional input vectors. Gradient descent can be used for fine-tuning the weights in such “autoencoder” networks, but this works well only if the initial weights are close to a good solution. We describe an effective way of initializing the weights that allows deep autoencoder networks to learn low-dimensional codes that work much better than principal components analysis as a tool to reduce the dimensionality of data.",
"A conventional approach to learning object detectors uses fully supervised learning techniques which assumes that a training image set with manual annotation of object bounding boxes are provided. The manual annotation of objects in large image sets is tedious and unreliable. Therefore, a weakly supervised learning approach is desirable, where the training set needs only binary labels regarding whether an image contains the target object class. In the weakly supervised approach a detector is used to iteratively annotate the training set and learn the object model. We present a novel weakly supervised learning framework for learning an object detector. Our framework incorporates a new initial annotation model to start the iterative learning of a detector and a model drift detection method that is able to detect and stop the iterative learning when the detector starts to drift away from the objects of interest. We demonstrate the effectiveness of our approach on the challenging PASCAL 2007 dataset.",
"",
""
]
} |
1611.05401 | 2552339304 | Several new methods have been proposed for performing valid inference after model selection. An older method is sampling splitting: use part of the data for model selection and part for inference. In this paper we revisit sample splitting combined with the bootstrap (or the Normal approximation). We show that this leads to a simple, assumption-free approach to inference and we establish results on the accuracy of the method. In fact, we find new bounds on the accuracy of the bootstrap and the Normal approximation for general nonlinear parameters with increasing dimension which we then use to assess the accuracy of regression inference. We show that an alternative, called the image bootstrap, has higher coverage accuracy at the cost of more computation. We define new parameters that measure variable importance and that can be inferred with greater accuracy than the usual regression coefficients. There is a inference-prediction tradeoff: splitting increases the accuracy and robustness of inference but can decrease the accuracy of the predictions. | There have been several additional approaches to this problem that don't fall in any of these broad categories. While this is a larger literature than can be addressed completely here, it includes early work on model selection @cite_1 and model averaging interpretations @cite_60 ; the impossibility results of @cite_15 , @cite_26 on random @math and model misspecification; methods based on resampling or sample splitting ; stability selection ; the conformal inference approach of @cite_22 ; goodness-of-fit tests of @cite_45 ; moment-constraint-based uniform confidence sets ; @cite_28 on inference about groups of variables under general designs; @cite_29 in the instrumental variable setting; @cite_41 on post-selection inference for @math -estimators, and the knockoffs approach of @cite_62 and later @cite_49 . Although they are not directed at linear models, @cite_39 and @cite_16 address similar problems for random forests. | {
"cite_N": [
"@cite_26",
"@cite_62",
"@cite_22",
"@cite_60",
"@cite_28",
"@cite_41",
"@cite_29",
"@cite_1",
"@cite_39",
"@cite_45",
"@cite_49",
"@cite_15",
"@cite_16"
],
"mid": [
"2769275910",
"",
"2340356402",
"",
"",
"",
"",
"2071168995",
"",
"",
"2116815574",
"2124142297",
""
],
"abstract": [
"Abstract. More than thirty years ago Halbert White inaugurated a “modelrobust” form of statistical inference based on the “sandwich estimator” of standard error. It is asymptotically correct even under “model misspecification,” that is, when models are approximations rather than generative truths. It is well-known to be “heteroskedasticity-consistent”, but it is less well-known to be “nonlinearity-consistent” as well. Nonlinearity, however, raises fundamental issues: When fitted models are approximations, conditioning on the regressor is no longer permitted because the ancillarity argument that justifies it breaks down. Two effects occur: (1) parameters become dependent on the regressor distribution; (2) the sampling variability of parameter estimates no longer derives from the conditional distribution of the response alone. Additional sampling variability arises when the nonlinearity conspires with the randomness of the regressors to generate a 1 √ N contribution to standard errors. Asymptotically, standard errors from “model-trusting” fixedregressor theories can deviate from those of “model-robust” randomregressor theories by arbitrary magnitudes. In the case of linear models, a test will be proposed for comparing the two types of standard errors.",
"",
"We develop a general framework for distribution-free predictive inference in regression, using conformal inference. The proposed methodology allows for the construction of a prediction band for the response variable using any estimator of the regression function. The resulting prediction band preserves the consistency properties of the original estimator under standard assumptions, while guaranteeing finite-sample marginal coverage even when these assumptions do not hold. We analyze and compare, both empirically and theoretically, the two major variants of our conformal framework: full conformal inference and split conformal inference, along with a related jackknife method. These methods offer different tradeoffs between statistical accuracy (length of resulting prediction intervals) and computational efficiency. As extensions, we develop a method for constructing valid in-sample prediction intervals called rank-one-out conformal inference, which has essentially the same computational efficiency as split conformal inference. We also describe an extension of our procedures for producing prediction bands with locally varying length, in order to adapt to heteroskedascity in the data. Finally, we propose a model-free notion of variable importance, called leave-one-covariate-out or LOCO inference. Accompanying this paper is an R package conformalInference that implements all of the proposals we have introduced. In the spirit of reproducibility, all of our empirical results can also be easily (re)generated using this package.",
"",
"",
"",
"",
"Let Z = (Y , X 1, ..., X i m), i = 1, ..., n, be i.i.d. random vectors, Z ∼ F, F ∈ F . It is desired to predict Y by βjXj , where (β1, ..., βm) ∈ B ⊆ R, under a prediction loss. Suppose that m = n, α > 1, i.e., there are many more explanatory variables than observations. We consider sets B restricted by the maximal number of non-zero coefficients of their members, or by their l1 radius. We study the following asymptotic question: How ‘large’ may the set B be, so that it is still possible to select empirically a predictor whose risk under F is close to that of the best predictor in the set. Sharp bounds for orders of magnitudes are given under various assumptions on F . Algorithmic complexity of the ensuing procedures is also studied. The main message of this paper and the implications of the above derived orders are that under various sparsity assumptions on the optimal predictor there is “asymptotically no harm” in introducing many more explanatory variables than observations. Furthermore, such practice can be beneficial in comparison with a procedure that screens in advance a small subset of explanatory variables. Another main result is that ’Lasso’-type procedures, i.e., optimization under l1 constraint, could be efficient in finding optimal sparse predictors in high dimensions. Running head: Persistence and Predictor Selection.",
"",
"",
"Summary. Stability selection was recently introduced by Meinshausen and Buhlmann as a very general technique designed to improve the performance of a variable selection algorithm. It is based on aggregating the results of applying a selection procedure to subsamples of the data. We introduce a variant, called complementary pairs stability selection, and derive bounds both on the expected number of variables included by complementary pairs stability selection that have low selection probability under the original procedure, and on the expected number of high selection probability variables that are excluded. These results require no (e.g. exchangeability) assumptions on the underlying model or on the quality of the original selection procedure. Under reasonable shape restrictions, the bounds can be further tightened, yielding improved error control, and therefore increasing the applicability of the methodology.",
"We consider the problem of estimating the conditional distribution of a post-model-selection estimator where the conditioning is on the selected model. The notion of a post-model-selection estimator here refers to the combined procedure resulting from first selecting a model (e.g., by a model selection criterion like AIC or by a hypothesis testing procedure) and second estimating the parameters in the selected model (e.g., by least-squares or maximum likelihood), all based on the same data set. We show that it is impossible to estimate this distribution with reasonable accuracy even asymptotically. In particular, we show that no estimator for this distribution can be uniformly consistent (not even locally). This follows as a corollary to (local) minimax lower bounds on the performance of estimators for this distribution. Similar impossibility results are also obtained for the conditional distribution of linear functions (e.g., predictors) of the post-model-selection estimator.",
""
]
} |
1611.05301 | 2552833967 | We propose and evaluate several triplet CNN architectures for measuring the similarity between sketches and photographs, within the context of the sketch based image retrieval (SBIR) task. In contrast to recent fine-grained SBIR work, we study the ability of our networks to generalise across diverse object categories from limited training data, and explore in detail strategies for weight sharing, pre-processing, data augmentation and dimensionality reduction. We exceed the performance of pre-existing techniques on both the Flickr15k category level SBIR benchmark by @math , and the TU-Berlin SBIR benchmark by @math , when trained on the 250 category TU-Berlin classification dataset augmented with 25k corresponding photographs harvested from the Internet. | Sketch based Image Retrieval (SBIR) began to gain momentum in the early nineties with colour-blob based query systems such as Flickner al's QBIC @cite_9 that matched coarse attributes of colour, shape and texture using region adjacency graphs. Several global image descriptors for matching blob based queries were subsequently proposed, using spectral signatures derived from Haar Wavelets @cite_26 and the Short-Time Fourier Transform @cite_10 . This early wave of SBIR systems was complemented in the late nineties by algorithms accepting line-art sketches, more closely resembling the free-hand sketches casually generated by lay users in the act of sketching a throw-away query @cite_22 . Such systems are characterised by their optimisation based matching approach; fitting the sketch under a deformable model to measure the support for sketched structure within each photograph in the database @cite_23 @cite_13 . Despite good accuracy, such approaches are slow and scale at best linearly. It was not until comparatively recently that global image descriptors were derived from line-art sketches, enabling more scalable indexing solutions. | {
"cite_N": [
"@cite_13",
"@cite_26",
"@cite_22",
"@cite_9",
"@cite_23",
"@cite_10"
],
"mid": [
"2119061765",
"2068272887",
"2144762834",
"2144679084",
"2110807182",
"2106795979"
],
"abstract": [
"We present a novel Content Based Video Retrieval (CBVR) system, driven by free-hand sketch queries depicting both objects and their movement (via dynamic cues; streak-lines and arrows). Our main contribution is a probabilistic model of video clips (based on Linear Dynamical Systems), leading to an algorithm for matching descriptions of sketched objects to video. We demonstrate our model fitting to clips under static and moving camera conditions, exhibiting linear and oscillatory motion. We evaluate retrieval on two real video data sets, and on a video data set exhibiting controlled variation in shape, color, motion and clutter.",
"We present a method for searching in an image database using a query image that is similar to the intended target. The query image may be a hand-drawn sketch or a (potentially low-quality) scan of the image to be retrieved. Our searching algorithm makes use of multiresolution wavelet decompositions of the query and database images. The coefficients of these decompositions are distilled into small “signatures” for each image. We introduce an “image querying metric” that operates on these signatures. This metric essentially compares how many significant wavelet coefficients the query has in common with potential targets. The metric includes parameters that can be tuned, using a statistical analysis, to accommodate the kinds of image distortions found in different types of image queries. The resulting algorithm is simple, requires very little storage overhead for the database of signatures, and is fast enough to be performed on a database of 20,000 images at interactive rates (on standard desktop machines) as a query is sketched. Our experiments with hundreds of queries in databases of 1000 and 20,000 images show dramatic improvement, in both speed and success rate, over using a conventional L1, L2, or color histogram norm. CR",
"We present an algorithm for extracting object descriptions from free-hand sketches of remembered scenes, drawn as video retrieval queries. Our sketches depict scene content, as well as indicators of motion. We report an exploratory study investigating how people sketch to depict recalled events. We incorporate several observations from this study into the design of a novel sketch parsing algorithm. We draw upon a temporal HMM classifier to recognise common pictograms, and graph-cut to identify more general objects.",
"Research on ways to extend and improve query methods for image databases is widespread. We have developed the QBIC (Query by Image Content) system to explore content-based retrieval methods. QBIC allows queries on large image and video databases based on example images, user-constructed sketches and drawings, selected color and texture patterns, camera and object motion, and other graphical information. Two key properties of QBIC are (1) its use of image and video content-computable properties of color, texture, shape and motion of images, videos and their objects-in the queries, and (2) its graphical query language, in which queries are posed by drawing, selecting and other graphical means. This article describes the QBIC system and demonstrates its query capabilities. QBIC technology is part of several IBM products. >",
"Effective image retrieval by content from database requires that visual image properties are used instead of textual labels to properly index and recover pictorial data. Retrieval by shape similarity, given a user-sketched template is particularly challenging, owing to the difficulty to derive a similarity measure that closely conforms to the common perception of similarity by humans. In this paper, we present a technique which is based on elastic matching of sketched templates over the shapes in the images to evaluate similarity ranks. The degree of matching achieved and the elastic deformation energy spent by the sketch to achieve such a match are used to derive a measure of similarity between the sketch and the images in the database and to rank images to be displayed. The elastic matching is integrated with arrangements to provide scale invariance and take into account spatial relationships between objects in multi-object queries. Examples from a prototype system are expounded with considerations about the effectiveness of the approach and comparative performance analysis.",
"This paper investigates the combined use of query by sketch and relevance feedback as techniques to ease user interaction and improve retrieval effectiveness in content-based image retrieval over the World Wide Web. To substantiate our ideas we implemented DrawSearch, a prototype image retrieval by content system that uses color, shape and texture to index and retrieve images. The system avails of Java applets for query by sketch and uses relevance feedback to allow users dynamically refine queries."
]
} |
1611.05301 | 2552833967 | We propose and evaluate several triplet CNN architectures for measuring the similarity between sketches and photographs, within the context of the sketch based image retrieval (SBIR) task. In contrast to recent fine-grained SBIR work, we study the ability of our networks to generalise across diverse object categories from limited training data, and explore in detail strategies for weight sharing, pre-processing, data augmentation and dimensionality reduction. We exceed the performance of pre-existing techniques on both the Flickr15k category level SBIR benchmark by @math , and the TU-Berlin SBIR benchmark by @math , when trained on the 250 category TU-Berlin classification dataset augmented with 25k corresponding photographs harvested from the Internet. | Mirroring the success of gradient domain features and dictionary learning methods to photo retrieval, both Hu al @cite_17 and Eitz al @cite_7 extended the Bag of Visual Words (BoVW) pipeline to SBIR, notably also proposing the Flickr15k dataset which became a defacto benchmark for category-level SBIR @cite_12 . Sparse features including the Structure Tensor @cite_30 , SHoG @cite_7 , Gradient Field Histogram of Oriented Gradients (GF-HOG) @cite_12 and its extended version @cite_0 are extracted from images pre-processed via Canny edge detection. Mid-level features were also explored through the HELO and key-shapes schemes of Saavedra and Bebustos @cite_21 @cite_27 , which although not indexable via BoVW could be matched via Hungarian algorithm. Mid-level structures were also explored in the Mindfinder system of Cao al @cite_14 who were the first to propose inverse index structure for scalable SBIR. Such systems score around 10-15 | {
"cite_N": [
"@cite_30",
"@cite_14",
"@cite_7",
"@cite_21",
"@cite_0",
"@cite_27",
"@cite_12",
"@cite_17"
],
"mid": [
"2136763397",
"2048546747",
"2153404544",
"2011459581",
"2237261560",
"2332729875",
"",
""
],
"abstract": [
"We address the problem of large scale sketch based image retrieval, searching in a database of over a million images. The search is based on a descriptor that elegantly addresses the asymmetry between the binary user sketch on the one hand and the full color image on the other hand. The proposed descriptor is constructed such that both the full color image and the sketch undergo exactly the same preprocessing steps. We also design an adapted version of the descriptor proposed for MPEG-7 and compare their performance on a database of 1.5 million images. Best matching images are clustered based on color histograms, to offset the lacking color in the query. Overall, the query results demonstrate that the system allows users an intuitive access to large image databases.",
"Retrieving images to match with a hand-drawn sketch query is a highly desired feature, especially with the popularity of devices with touch screens. Although query-by-sketch has been extensively studied since 1990s, it is still very challenging to build a real-time sketch-based image search engine on a large-scale database due to the lack of effective and efficient matching indexing solutions. The explosive growth of web images and the phenomenal success of search techniques have encouraged us to revisit this problem and target at solving the problem of web-scale sketch-based image retrieval. In this work, a novel index structure and the corresponding raw contour-based matching algorithm are proposed to calculate the similarity between a sketch query and natural images, and make sketch-based image retrieval scalable to millions of images. The proposed solution simultaneously considers storage cost, retrieval accuracy, and efficiency, based on which we have developed a real-time sketch-based image search engine by indexing more than 2 million images. Extensive experiments on various retrieval tasks (basic shape search, specific image search, and similar image search) show better accuracy and efficiency than state-of-the-art methods.",
"We introduce a benchmark for evaluating the performance of large-scale sketch-based image retrieval systems. The necessary data are acquired in a controlled user study where subjects rate how well given sketch image pairs match. We suggest how to use the data for evaluating the performance of sketch-based image retrieval systems. The benchmark data as well as the large image database are made publicly available for further studies of this type. Furthermore, we develop new descriptors based on the bag-of-features approach and use the benchmark to demonstrate that they significantly outperform other descriptors in the literature.",
"Although sketch based image retrieval (SBIR) is still a young research area, there are many applications capable of exploiting this retrieval paradigm, such as web searching and pattern detection. Moreover, nowadays drawing a simple sketch query turns very simple since touch screen based technology is being expanded. In this work, we propose a novel local approach for SBIR based on detecting simple shapes which are named keyshapes. Our method works as a local strategy, but instead of detecting keypoints, it detects keyshapes over which local descriptors are computed. Our proposal based on keyshapes allow us to represent the structure of the objects in an image which could be used to increase the effectiveness in the retrieval task. Indeed, our results show an improvement in the retrieval effectiveness with respect to the state of the art. Furthermore, we demonstrate that combining our keyshape approach with a Bag of Feature approach allows us to achieve significant improvement with respect to the effectiveness of the retrieval task.",
"We present a scalable system for sketch-based image retrieval (SBIR), extending the state of the art Gradient Field HoG (GF-HoG) retrieval framework through two technical contributions. First, we extend GF-HoG to enable colorshape retrieval and comprehensively evaluate several earlyand late-fusion approaches for integrating the modality of color, considering both the accuracy and speed of sketch retrieval. Second, we propose an efficient inverse-index representation for GF-HoG that delivers scalable search with interactive query times over millions of images. A mobile app demo accompanies this paper (Android).",
"",
"",
""
]
} |
1611.05377 | 2951657494 | Multi-task learning aims to improve generalization performance of multiple prediction tasks by appropriately sharing relevant information across them. In the context of deep neural networks, this idea is often realized by hand-designed network architectures with layers that are shared across tasks and branches that encode task-specific features. However, the space of possible multi-task deep architectures is combinatorially large and often the final architecture is arrived at by manual exploration of this space subject to designer's bias, which can be both error-prone and tedious. In this work, we propose a principled approach for designing compact multi-task deep learning architectures. Our approach starts with a thin network and dynamically widens it in a greedy manner during training using a novel criterion that promotes grouping of similar tasks together. Our Extensive evaluation on person attributes classification tasks involving facial and clothing attributes suggests that the models produced by the proposed method are fast, compact and can closely match or exceed the state-of-the-art accuracy from strong baselines by much more expensive models. | Multi-Task Learning. There is a long history of research in multi-task learning @cite_27 @cite_42 @cite_11 @cite_15 @cite_7 . Most proposed techniques assume that all tasks are related and appropriate for joint training. A few methods have addressed the problem of with whom'' each task should share features @cite_33 @cite_11 @cite_9 @cite_25 @cite_15 @cite_32 . These methods are generally designed for shallow classification models, while our work investigates feature sharing among tasks in hierarchical models such as deep neural networks. | {
"cite_N": [
"@cite_33",
"@cite_7",
"@cite_9",
"@cite_42",
"@cite_32",
"@cite_27",
"@cite_15",
"@cite_25",
"@cite_11"
],
"mid": [
"",
"",
"2118099552",
"99485931",
"2951389238",
"",
"",
"2186054958",
"2949664970"
],
"abstract": [
"",
"",
"Multi-task learning (MTL) learns multiple related tasks simultaneously to improve generalization performance. Alternating structure optimization (ASO) is a popular MTL method that learns a shared low-dimensional predictive structure on hypothesis spaces from multiple related tasks. It has been applied successfully in many real world applications. As an alternative MTL approach, clustered multi-task learning (CMTL) assumes that multiple tasks follow a clustered structure, i.e., tasks are partitioned into a set of groups where tasks in the same group are similar to each other, and that such a clustered structure is unknown a priori. The objectives in ASO and CMTL differ in how multiple tasks are related. Interestingly, we show in this paper the equivalence relationship between ASO and CMTL, providing significant new insights into ASO and CMTL as well as their inherent relationship. The CMTL formulation is non-convex, and we adopt a convex relaxation to the CMTL formulation. We further establish the equivalence relationship between the proposed convex relaxation of CMTL and an existing convex relaxation of ASO, and show that the proposed convex CMTL formulation is significantly more efficient especially for high-dimensional data. In addition, we present three algorithms for solving the convex CMTL formulation. We report experimental results on benchmark datasets to demonstrate the efficiency of the proposed algorithms.",
"Preface. Part I: Overview Articles. 1. Learning to Learn: Introduction and Overview S. Thrun, L. Pratt. 2. A Survey of Connectionist Network Reuse Through Transfer L. Pratt, B. Jennings. 3. Transfer in Cognition A. Robins. Part II: Prediction. 4. Theoretical Models of Learning to Learn J. Baxter. 5. Multitask Learning R. Caruana. 6. Making a Low-Dimensional Representation Suitable for Diverse Tasks N. Intrator, S. Edelman. 7. The Canonical Distortion Measure for Vector Quantization and Function Approximation J. Baxter. 8. Lifelong Learning Algorithms S. Thrun. Part III: Relatedness. 9. The Parallel Transfer of Task Knowledge Using Dynamic Learning Rates Based on a Measure of Relatedness D.L. Silver, R.E. Mercer. 10. Clustering Learning Tasks and the Selective Cross-Task Transfer of Knowledge S. Thrun, J. O'Sullivan. Part IV: Control. 11. CHILD: A First Step Towards Continual Learning M.B. Ring. 12. Reinforcement Learning with Self-Modifying Policies J. Schmidhuber, et al 13. Creating Advice-Taking Reinforcement Learners R. Maclin, J.W. Shavlik. Contributing Authors. Index.",
"Multitask learning algorithms are typically designed assuming some fixed, a priori known latent structure shared by all the tasks. However, it is usually unclear what type of latent task structure is the most appropriate for a given multitask learning problem. Ideally, the \"right\" latent task structure should be learned in a data-driven manner. We present a flexible, nonparametric Bayesian model that posits a mixture of factor analyzers structure on the tasks. The nonparametric aspect makes the model expressive enough to subsume many existing models of latent task structures (e.g, mean-regularized tasks, clustered tasks, low-rank or linear non-linear subspace assumption on tasks, etc.). Moreover, it can also learn more general task structures, addressing the shortcomings of such models. We present a variational inference algorithm for our model. Experimental results on synthetic and real-world datasets, on both regression and classification problems, demonstrate the effectiveness of the proposed method.",
"",
"",
"In multi-task learning (MTL), multiple tasks are learnt jointly. A major assumption for this paradigm is that all those tasks are indeed related so that the joint training is appropriate and beneficial. In this paper, we study the problem of multi-task learning of shared feature representations among tasks, while simultaneously determining \"with whom\" each task should share. We formulate the problem as a mixed integer programming and provide an alternating minimization technique to solve the optimization problem of jointly identifying grouping structures and parameters. The algorithm mono-tonically decreases the objective function and converges to a local optimum. Compared to the standard MTL paradigm where all tasks are in a single group, our algorithm improves its performance with statistical significance for three out of the four datasets we have studied. We also demonstrate its advantage over other task grouping techniques investigated in literature.",
"In multi-task learning several related tasks are considered simultaneously, with the hope that by an appropriate sharing of information across tasks, each task may benefit from the others. In the context of learning linear functions for supervised classification or regression, this can be achieved by including a priori information about the weight vectors associated with the tasks, and how they are expected to be related to each other. In this paper, we assume that tasks are clustered into groups, which are unknown beforehand, and that tasks within a group have similar weight vectors. We design a new spectral norm that encodes this a priori assumption, without the prior knowledge of the partition of tasks into groups, resulting in a new convex optimization formulation for multi-task learning. We show in simulations on synthetic examples and on the IEDB MHC-I binding dataset, that our approach outperforms well-known convex methods for multi-task learning, as well as related non convex methods dedicated to the same problem."
]
} |
1611.05377 | 2951657494 | Multi-task learning aims to improve generalization performance of multiple prediction tasks by appropriately sharing relevant information across them. In the context of deep neural networks, this idea is often realized by hand-designed network architectures with layers that are shared across tasks and branches that encode task-specific features. However, the space of possible multi-task deep architectures is combinatorially large and often the final architecture is arrived at by manual exploration of this space subject to designer's bias, which can be both error-prone and tedious. In this work, we propose a principled approach for designing compact multi-task deep learning architectures. Our approach starts with a thin network and dynamically widens it in a greedy manner during training using a novel criterion that promotes grouping of similar tasks together. Our Extensive evaluation on person attributes classification tasks involving facial and clothing attributes suggests that the models produced by the proposed method are fast, compact and can closely match or exceed the state-of-the-art accuracy from strong baselines by much more expensive models. | Recently, several methods have been proposed for multi-task learning using deep neural networks. HyperFace @cite_24 simultaneously learns to perform face detection, landmarks localization, pose estimation and gender recognition. UberNet @cite_36 jointly learns low-, mid-, and high-level computer vision tasks using a compact network model. MultiNet @cite_5 exploits recurrent networks for transferring information across tasks. Cross-ResNet @cite_48 connects tasks through residual learning for knowledge transfer. However, all these methods rely on hand-designed network architectures composed of base layers that are shared across tasks and specialized branches that learn task-specific features. | {
"cite_N": [
"@cite_24",
"@cite_5",
"@cite_36",
"@cite_48"
],
"mid": [
"2290180618",
"2412879760",
"2510182739",
"2337753992"
],
"abstract": [
"We present an algorithm for simultaneous face detection, landmarks localization, pose estimation and gender recognition using deep convolutional neural networks (CNN). The proposed method called, HyperFace, fuses the intermediate layers of a deep CNN using a separate CNN followed by a multi-task learning algorithm that operates on the fused features. It exploits the synergy among the tasks which boosts up their individual performances. Additionally, we propose two variants of HyperFace: (1) HyperFace-ResNet that builds on the ResNet-101 model and achieves significant improvement in performance, and (2) Fast-HyperFace that uses a high recall fast face detector for generating region proposals to improve the speed of the algorithm. Extensive experiments show that the proposed models are able to capture both global and local information in faces and performs significantly better than many competitive algorithms for each of these four tasks.",
"Modern discriminative predictors have been shown to match natural intelligences in specific perceptual tasks in image classification, object and part detection, boundary extraction, etc. However, a major advantage that natural intelligences still have is that they work well for perceptual problems together, solving them efficiently and coherently in an . In order to capture some of these advantages in machine perception, we ask two questions: whether deep neural networks can learn universal image representations, useful not only for a single task but for all of them, and how the solutions to the different tasks can be integrated in this framework. We answer by proposing a new architecture, which we call , in which not only deep image features are shared between tasks, but where tasks can interact in a recurrent manner by encoding the results of their analysis in a common shared representation of the data. In this manner, we show that the performance of individual tasks in standard benchmarks can be improved first by sharing features between them and then, more significantly, by integrating their solutions in the common representation.",
"In this work we introduce a convolutional neural network (CNN) that jointly handles low-, mid-, and high-level vision tasks in a unified architecture that is trained end-to-end. Such a universal network can act like a swiss knife' for vision tasks; we call this architecture an UberNet to indicate its overarching nature. We address two main technical challenges that emerge when broadening up the range of tasks handled by a single CNN: (i) training a deep architecture while relying on diverse training sets and (ii) training many (potentially unlimited) tasks with a limited memory budget. Properly addressing these two problems allows us to train accurate predictors for a host of tasks, without compromising accuracy. Through these advances we train in an end-to-end manner a CNN that simultaneously addresses (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 of these tasks in 0.7 seconds per frame on a single GPU. A demonstration of this system can be found at this http URL",
"Residual learning has recently surfaced as an effective means of constructing very deep neural networks for object recognition. However, current incarnations of residual networks do not allow for the modeling and integration of complex relations between closely coupled recognition tasks or across domains. Such problems are often encountered in multimedia applications involving large-scale content recognition. We propose a novel extension of residual learning for deep networks that enables intuitive learning across multiple related tasks using cross-connections called cross-residuals. These cross-residuals connections can be viewed as a form of in-network regularization and enables greater network generalization. We show how cross-residual learning (CRL) can be integrated in multitask networks to jointly train and detect visual concepts across several tasks. We present a single multitask cross-residual network with >40 less parameters that is able to achieve competitive, or even better, detection performance on a visual sentiment concept detection problem normally requiring multiple specialized single-task networks. The resulting multitask cross-residual network also achieves better detection performance by about 10.4 over a standard multitask residual network without cross-residuals with even a small amount of cross-task weighting."
]
} |
1611.05377 | 2951657494 | Multi-task learning aims to improve generalization performance of multiple prediction tasks by appropriately sharing relevant information across them. In the context of deep neural networks, this idea is often realized by hand-designed network architectures with layers that are shared across tasks and branches that encode task-specific features. However, the space of possible multi-task deep architectures is combinatorially large and often the final architecture is arrived at by manual exploration of this space subject to designer's bias, which can be both error-prone and tedious. In this work, we propose a principled approach for designing compact multi-task deep learning architectures. Our approach starts with a thin network and dynamically widens it in a greedy manner during training using a novel criterion that promotes grouping of similar tasks together. Our Extensive evaluation on person attributes classification tasks involving facial and clothing attributes suggests that the models produced by the proposed method are fast, compact and can closely match or exceed the state-of-the-art accuracy from strong baselines by much more expensive models. | As network architectures become deeper, defining the right level of feature sharing across tasks through handcrafted network branches is impractical. Cross-stitching networks @cite_7 have been recently proposed to learn an optimal combination of shared and task-specific representations. Although cross-stitching units connecting task-specific sub-networks are designed to the feature sharing among tasks, the size of the network grows linearly with the number of tasks, causing scalability issues. We instead propose a novel algorithm that makes decisions about branching based on task relatedness, while optimizing for the efficiency of the model. We note that other techniques such as HD-CNN @cite_40 and Network of Experts @cite_34 also group related classes to perform hierarchical classification, but these methods are not applicable for the multi-label setting (where labels are not mutually exclusive). | {
"cite_N": [
"@cite_40",
"@cite_34",
"@cite_7"
],
"mid": [
"1937922215",
"2336829997",
""
],
"abstract": [
"In image classification, visual separability between different object categories is highly uneven, and some categories are more difficult to distinguish than others. Such difficult categories demand more dedicated classifiers. However, existing deep convolutional neural networks (CNN) are trained as flat N-way classifiers, and few efforts have been made to leverage the hierarchical structure of categories. In this paper, we introduce hierarchical deep CNNs (HD-CNNs) by embedding deep CNNs into a category hierarchy. An HD-CNN separates easy classes using a coarse category classifier while distinguishing difficult classes using fine category classifiers. During HD-CNN training, component-wise pretraining is followed by global finetuning with a multinomial logistic loss regularized by a coarse category consistency term. In addition, conditional executions of fine category classifiers and layer parameter compression make HD-CNNs scalable for large-scale visual recognition. We achieve state-of-the-art results on both CIFAR100 and large-scale ImageNet 1000-class benchmark datasets. In our experiments, we build up three different HD-CNNs and they lower the top-1 error of the standard CNNs by 2.65 , 3.1 and 1.1 , respectively.",
"We present a tree-structured network architecture for large-scale image classification. The trunk of the network contains convolutional layers optimized over all classes. At a given depth, the trunk splits into separate branches, each dedicated to discriminate a different subset of classes. Each branch acts as an expert classifying a set of categories that are difficult to tell apart, while the trunk provides common knowledge to all experts in the form of shared features. The training of our “network of experts” is completely end-to-end: the partition of categories into disjoint subsets is learned simultaneously with the parameters of the network trunk and the experts are trained jointly by minimizing a single learning objective over all classes. The proposed structure can be built from any existing convolutional neural network (CNN). We demonstrate its generality by adapting 4 popular CNNs for image categorization into the form of networks of experts. Our experiments on CIFAR100 and ImageNet show that in every case our method yields a substantial improvement in accuracy over the base CNN, and gives the best result achieved so far on CIFAR100. Finally, the improvement in accuracy comes at little additional cost: compared to the base network, the training time is only moderately increased and the number of parameters is comparable or in some cases even lower.",
""
]
} |
1611.05377 | 2951657494 | Multi-task learning aims to improve generalization performance of multiple prediction tasks by appropriately sharing relevant information across them. In the context of deep neural networks, this idea is often realized by hand-designed network architectures with layers that are shared across tasks and branches that encode task-specific features. However, the space of possible multi-task deep architectures is combinatorially large and often the final architecture is arrived at by manual exploration of this space subject to designer's bias, which can be both error-prone and tedious. In this work, we propose a principled approach for designing compact multi-task deep learning architectures. Our approach starts with a thin network and dynamically widens it in a greedy manner during training using a novel criterion that promotes grouping of similar tasks together. Our Extensive evaluation on person attributes classification tasks involving facial and clothing attributes suggests that the models produced by the proposed method are fast, compact and can closely match or exceed the state-of-the-art accuracy from strong baselines by much more expensive models. | Model Compression and Acceleration. Existing deep convolutional neural network models are computationally and memory intensive, hindering their deployment in devices with low memory resources or in applications with strict latency requirements. Methods for compressing and accelerating convolutional networks include knowledge distillation @cite_44 @cite_8 , low-rank-factorization @cite_2 @cite_3 @cite_12 , pruning and quantization @cite_50 @cite_16 , structured matrices @cite_43 @cite_35 @cite_30 , and dynamic capacity networks @cite_26 . These methods are task-agnostic and therefore most of them are complementary to our approach, which seeks to obtain a compact multi-task model by widening a low-capacity network based on task relatedness. Moreover, many of these state-of-the-art compression techniques can be used to further reduce the size of our learned multi-task architectures. | {
"cite_N": [
"@cite_30",
"@cite_35",
"@cite_26",
"@cite_8",
"@cite_3",
"@cite_44",
"@cite_43",
"@cite_50",
"@cite_2",
"@cite_16",
"@cite_12"
],
"mid": [
"2524379082",
"2949964376",
"2173038751",
"1690739335",
"2963225922",
"1821462560",
"2949560654",
"2964299589",
"",
"2123469553",
"2058641082"
],
"abstract": [
"We introduce Tamp, an open source C++ library for reducing the space and time costs of deep neural network models. In particular, Tamp implements several recent works which use structured matrices to replace unstructured matrices which are often bottlenecks in neural networks. Tamp is also designed to serve as a unified development platform with several supported optimization back-ends and abstracted data types. This paper introduces the design and API and also demonstrates the effectiveness with experiments on public datasets.",
"We consider the task of building compact deep learning pipelines suitable for deployment on storage and power constrained mobile devices. We propose a unified framework to learn a broad family of structured parameter matrices that are characterized by the notion of low displacement rank. Our structured transforms admit fast function and gradient evaluation, and span a rich range of parameter sharing configurations whose statistical modeling capacity can be explicitly tuned along a continuum from structured to unstructured. Experimental results show that these transforms can significantly accelerate inference and forward backward passes during training, and offer superior accuracy-compactness-speed tradeoffs in comparison to a number of existing techniques. In keyword spotting applications in mobile speech recognition, our methods are much more effective than standard linear low-rank bottleneck layers and nearly retain the performance of state of the art models, while providing more than 3.5-fold compression.",
"We introduce the Dynamic Capacity Network (DCN), a neural network that can adaptively assign its capacity across different portions of the input data. This is achieved by combining modules of two types: low-capacity sub-networks and high-capacity sub-networks. The low-capacity sub-networks are applied across most of the input, but also provide a guide to select a few portions of the input on which to apply the high-capacity sub-networks. The selection is made using a novel gradient-based attention mechanism, that efficiently identifies input regions for which the DCN's output is most sensitive and to which we should devote more capacity. We focus our empirical evaluation on the Cluttered MNIST and SVHN image datasets. Our findings indicate that DCNs are able to drastically reduce the number of computations, compared to traditional convolutional neural networks, while maintaining similar or even better performance.",
"While depth tends to improve network performances, it also makes gradient-based training more difficult since deeper networks tend to be more non-linear. The recently proposed knowledge distillation approach is aimed at obtaining small and fast-to-execute models, and it has shown that a student network could imitate the soft output of a larger teacher network or ensemble of networks. In this paper, we extend this idea to allow the training of a student that is deeper and thinner than the teacher, using not only the outputs but also the intermediate representations learned by the teacher as hints to improve the training process and final performance of the student. Because the student intermediate hidden layer will generally be smaller than the teacher's intermediate hidden layer, additional parameters are introduced to map the student hidden layer to the prediction of the teacher hidden layer. This allows one to train deeper students that can generalize better or run faster, a trade-off that is controlled by the chosen student capacity. For example, on CIFAR-10, a deep student network with almost 10.4 times less parameters outperforms a larger, state-of-the-art teacher network.",
"Abstract: Large CNNs have delivered impressive performance in various computer vision applications. But the storage and computation requirements make it problematic for deploying these models on mobile devices. Recently, tensor decompositions have been used for speeding up CNNs. In this paper, we further develop the tensor decomposition technique. We propose a new algorithm for computing the low-rank tensor decomposition for removing the redundancy in the convolution kernels. The algorithm finds the exact global optimizer of the decomposition and is more effective than iterative methods. Based on the decomposition, we further propose a new method for training low-rank constrained CNNs from scratch. Interestingly, while achieving a significant speedup, sometimes the low-rank constrained CNNs delivers significantly better performance than their non-constrained counterparts. On the CIFAR-10 dataset, the proposed low-rank NIN model achieves @math accuracy (without data augmentation), which also improves upon state-of-the-art result. We evaluated the proposed method on CIFAR-10 and ILSVRC12 datasets for a variety of modern CNNs, including AlexNet, NIN, VGG and GoogleNet with success. For example, the forward time of VGG-16 is reduced by half while the performance is still comparable. Empirical success suggests that low-rank tensor decompositions can be a very useful tool for speeding up large CNNs.",
"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.",
"We explore the redundancy of parameters in deep neural networks by replacing the conventional linear projection in fully-connected layers with the circulant projection. The circulant structure substantially reduces memory footprint and enables the use of the Fast Fourier Transform to speed up the computation. Considering a fully-connected neural network layer with d input nodes, and d output nodes, this method improves the time complexity from O(d^2) to O(dlogd) and space complexity from O(d^2) to O(d). The space savings are particularly important for modern deep convolutional neural network architectures, where fully-connected layers typically contain more than 90 of the network parameters. We further show that the gradient computation and optimization of the circulant projections can be performed very efficiently. Our experiments on three standard datasets show that the proposed approach achieves this significant gain in storage and efficiency with minimal increase in error rate compared to neural networks with unstructured projections.",
"Abstract: 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.",
"",
"A major challenge in biometrics is performing the test at the client side, where hardware resources are often limited. Deep learning approaches pose a unique challenge: while such architectures dominate the field of face recognition with regard to accuracy, they require elaborate, multi-stage computations. Recently, there has been some work on compressing networks for the purpose of reducing run time and network size. However, it is not clear that these compression methods would work in deep face nets, which are, generally speaking, less redundant than the object recognition networks, i.e., they are already relatively lean. We propose two novel methods for compression: one based on eliminating lowly active channels and the other on coupling pruning with repeated use of already computed elements. Pruning of entire channels is an appealing idea, since it leads to direct saving in run time in almost every reasonable architecture.",
"While Deep Neural Networks (DNNs) have achieved tremendous success for large vocabulary continuous speech recognition (LVCSR) tasks, training of these networks is slow. One reason is that DNNs are trained with a large number of training parameters (i.e., 10-50 million). Because networks are trained with a large number of output targets to achieve good performance, the majority of these parameters are in the final weight layer. In this paper, we propose a low-rank matrix factorization of the final weight layer. We apply this low-rank technique to DNNs for both acoustic modeling and language modeling. We show on three different LVCSR tasks ranging between 50-400 hrs, that a low-rank factorization reduces the number of parameters of the network by 30-50 . This results in roughly an equivalent reduction in training time, without a significant loss in final recognition accuracy, compared to a full-rank representation."
]
} |
1611.05377 | 2951657494 | Multi-task learning aims to improve generalization performance of multiple prediction tasks by appropriately sharing relevant information across them. In the context of deep neural networks, this idea is often realized by hand-designed network architectures with layers that are shared across tasks and branches that encode task-specific features. However, the space of possible multi-task deep architectures is combinatorially large and often the final architecture is arrived at by manual exploration of this space subject to designer's bias, which can be both error-prone and tedious. In this work, we propose a principled approach for designing compact multi-task deep learning architectures. Our approach starts with a thin network and dynamically widens it in a greedy manner during training using a novel criterion that promotes grouping of similar tasks together. Our Extensive evaluation on person attributes classification tasks involving facial and clothing attributes suggests that the models produced by the proposed method are fast, compact and can closely match or exceed the state-of-the-art accuracy from strong baselines by much more expensive models. | Person Attribute Classification. Methods for recognizing attributes of people, such as facial and clothing attributes, have received increased attention in the past few years. In the visual surveillance domain, person attributes serve as features for improving person re-identification @cite_47 and enable search of suspects based on their description @cite_39 @cite_4 . In e-commerce applications, these attributes have proven effective in improving clothing retrieval @cite_6 , and fashion recommendation @cite_17 . It has also been shown that facial attribute prediction is helpful as an auxiliary task for improving face detection @cite_20 and face alignment @cite_19 . | {
"cite_N": [
"@cite_4",
"@cite_6",
"@cite_39",
"@cite_19",
"@cite_47",
"@cite_20",
"@cite_17"
],
"mid": [
"2058102599",
"2950940417",
"2102497689",
"1795776638",
"",
"2950557924",
"1971375352"
],
"abstract": [
"We address the problem of attribute-based people search in real surveillance environments. The system we developed is capable of answering user queries such as \"show me all people with a beard and sunglasses, wearing a white hat and a patterned blue shirt, from all metro cameras in the downtown area, from 2pm to 4pm last Saturday\". In this paper, we describe the lessons we learned from practical deployments of our system, and how we made our algorithms achieve the accuracy and efficiency required by many police departments around the world. In particular, we show that a novel set of multimodal integral filters and proper normalization of attribute scores are critical to obtain good performance. We conduct a comprehensive experimental analysis on video footage captured from a large set of surveillance cameras monitoring metro chokepoints, in both crowded and normal activity periods. Moreover, we show impressive results using images from the recent Boston marathon bombing event, where our system can rapidly retrieve the two suspects based on their attributes from a database containing more than one thousand people present at the event.",
"We address the problem of cross-domain image retrieval, considering the following practical application: given a user photo depicting a clothing image, our goal is to retrieve the same or attribute-similar clothing items from online shopping stores. This is a challenging problem due to the large discrepancy between online shopping images, usually taken in ideal lighting pose background conditions, and user photos captured in uncontrolled conditions. To address this problem, we propose a Dual Attribute-aware Ranking Network (DARN) for retrieval feature learning. More specifically, DARN consists of two sub-networks, one for each domain, whose retrieval feature representations are driven by semantic attribute learning. We show that this attribute-guided learning is a key factor for retrieval accuracy improvement. In addition, to further align with the nature of the retrieval problem, we impose a triplet visual similarity constraint for learning to rank across the two sub-networks. Another contribution of our work is a large-scale dataset which makes the network learning feasible. We exploit customer review websites to crawl a large set of online shopping images and corresponding offline user photos with fine-grained clothing attributes, i.e., around 450,000 online shopping images and about 90,000 exact offline counterpart images of those online ones. All these images are collected from real-world consumer websites reflecting the diversity of the data modality, which makes this dataset unique and rare in the academic community. We extensively evaluate the retrieval performance of networks in different configurations. The top-20 retrieval accuracy is doubled when using the proposed DARN other than the current popular solution using pre-trained CNN features only (0.570 vs. 0.268).",
"We propose a novel framework for searching for people in surveillance environments. Rather than relying on face recognition technology, which is known to be sensitive to typical surveillance conditions such as lighting changes, face pose variation, and low-resolution imagery, we approach the problem in a different way: we search for people based on a parsing of human parts and their attributes, including facial hair, eyewear, clothing color, etc. These attributes can be extracted using detectors learned from large amounts of training data. A complete system that implements our framework is presented. At the interface, the user can specify a set of personal characteristics, and the system then retrieves events that match the provided description. For example, a possible query is “show me the bald people who entered a given building last Saturday wearing a red shirt and sunglasses.” This capability is useful in several applications, such as finding suspects or missing people. To evaluate the performance of our approach, we present extensive experiments on a set of images collected from the Internet, on infrared imagery, and on two-and-a-half months of video from a real surveillance environment. We are not aware of any similar surveillance system capable of automatically finding people in video based on their fine-grained body parts and attributes.",
"In this study, we show that landmark detection or face alignment task is not a single and independent problem. Instead, its robustness can be greatly improved with auxiliary information. Specifically, we jointly optimize landmark detection together with the recognition of heterogeneous but subtly correlated facial attributes, such as gender, expression, and appearance attributes. This is non-trivial since different attribute inference tasks have different learning difficulties and convergence rates. To address this problem, we formulate a novel tasks-constrained deep model, which not only learns the inter-task correlation but also employs dynamic task coefficients to facilitate the optimization convergence when learning multiple complex tasks. Extensive evaluations show that the proposed task-constrained learning (i) outperforms existing face alignment methods, especially in dealing with faces with severe occlusion and pose variation, and (ii) reduces model complexity drastically compared to the state-of-the-art methods based on cascaded deep model.",
"",
"In this paper, we propose a novel deep convolutional network (DCN) that achieves outstanding performance on FDDB, PASCAL Face, and AFW. Specifically, our method achieves a high recall rate of 90.99 on the challenging FDDB benchmark, outperforming the state-of-the-art method by a large margin of 2.91 . Importantly, we consider finding faces from a new perspective through scoring facial parts responses by their spatial structure and arrangement. The scoring mechanism is carefully formulated considering challenging cases where faces are only partially visible. This consideration allows our network to detect faces under severe occlusion and unconstrained pose variation, which are the main difficulty and bottleneck of most existing face detection approaches. We show that despite the use of DCN, our network can achieve practical runtime speed.",
"Beauty e-Experts, a fully automatic system for makeover recommendation and synthesis, is developed in this work. The makeover recommendation and synthesis system simultaneously considers many kinds of makeover items on hairstyle and makeup. Given a user-provided frontal face image with short bound hair and no light makeup, the Beauty e-Experts system not only recommends the most suitable hairdo and makeup, but also synthesizes the virtual hairdo and makeup effects. To acquire enough knowledge for beauty modeling, we built the Beauty e-Experts Database, which contains 1,505 female photos with a variety of attributes annotated with different discrete values. We organize these attributes into two different categories, beauty attributes and beauty-related attributes. Beauty attributes refer to those values that are changeable during the makeover process and thus need to be recommended by the system. Beauty-related attributes are those values that cannot be changed during the makeup process but can help the system to perform recommendation. Based on this Beauty e-Experts Dataset, two problems are addressed for the Beauty e-Experts system: what to recommend and how to wear it, which describes a similar process of selecting hairstyle and cosmetics in daily life. For the what-to-recommend problem, we propose a multiple tree-structured supergraph model to explore the complex relationships among high-level beauty attributes, mid-level beauty-related attributes, and low-level image features. Based on this model, the most compatible beauty attributes for a given facial image can be efficiently inferred. For the how-to-wear-it problem, an effective and efficient facial image synthesis module is designed to seamlessly synthesize the recommended makeovers into the user facial image. We have conducted extensive experiments on testing images of various conditions to evaluate and analyze the proposed system. The experimental results well demonstrate the effectiveness and efficiency of the proposed system."
]
} |
1611.05377 | 2951657494 | Multi-task learning aims to improve generalization performance of multiple prediction tasks by appropriately sharing relevant information across them. In the context of deep neural networks, this idea is often realized by hand-designed network architectures with layers that are shared across tasks and branches that encode task-specific features. However, the space of possible multi-task deep architectures is combinatorially large and often the final architecture is arrived at by manual exploration of this space subject to designer's bias, which can be both error-prone and tedious. In this work, we propose a principled approach for designing compact multi-task deep learning architectures. Our approach starts with a thin network and dynamically widens it in a greedy manner during training using a novel criterion that promotes grouping of similar tasks together. Our Extensive evaluation on person attributes classification tasks involving facial and clothing attributes suggests that the models produced by the proposed method are fast, compact and can closely match or exceed the state-of-the-art accuracy from strong baselines by much more expensive models. | State-of-the-art methods for person attribute prediction are based on deep convolutional neural networks @cite_28 @cite_14 @cite_49 @cite_37 . Most methods either train separate classifiers per attribute @cite_37 or perform joint learning with a fully shared network @cite_13 . Multi-task networks have been used with base layers that are shared across all attributes, and branches to encode task-specific features for each attribute category @cite_6 @cite_31 . However, in contrast to our work, the network branches are hand-designed and do not exploit the fact that some attributes are more related than others in order to determine the level of sharing among tasks in the network. Moreover, we show that our approach produces a single compact network that can predict both facial and clothing attributes simultaneously. | {
"cite_N": [
"@cite_37",
"@cite_14",
"@cite_28",
"@cite_6",
"@cite_49",
"@cite_31",
"@cite_13"
],
"mid": [
"2147414309",
"1834627138",
"2963721945",
"2950940417",
"1946323491",
"2286727787",
"2311038409"
],
"abstract": [
"We propose a method for inferring human attributes (such as gender, hair style, clothes style, expression, action) from images of people under large variation of viewpoint, pose, appearance, articulation and occlusion. Convolutional Neural Nets (CNN) have been shown to perform very well on large scale object recognition problems. In the context of attribute classification, however, the signal is often subtle and it may cover only a small part of the image, while the image is dominated by the effects of pose and viewpoint. Discounting for pose variation would require training on very large labeled datasets which are not presently available. Part-based models, such as poselets [4] and DPM [12] have been shown to perform well for this problem but they are limited by shallow low-level features. We propose a new method which combines part-based models and deep learning by training pose-normalized CNNs. We show substantial improvement vs. state-of-the-art methods on challenging attribute classification tasks in unconstrained settings. Experiments confirm that our method outperforms both the best part-based methods on this problem and conventional CNNs trained on the full bounding box of the person.",
"Predicting face attributes in the wild is challenging due to complex face variations. We propose a novel deep learning framework for attribute prediction in the wild. It cascades two CNNs, LNet and ANet, which are fine-tuned jointly with attribute tags, but pre-trained differently. LNet is pre-trained by massive general object categories for face localization, while ANet is pre-trained by massive face identities for attribute prediction. This framework not only outperforms the state-of-the-art with a large margin, but also reveals valuable facts on learning face representation. (1) It shows how the performances of face localization (LNet) and attribute prediction (ANet) can be improved by different pre-training strategies. (2) It reveals that although the filters of LNet are fine-tuned only with image-level attribute tags, their response maps over entire images have strong indication of face locations. This fact enables training LNet for face localization with only image-level annotations, but without face bounding boxes or landmarks, which are required by all attribute recognition works. (3) It also demonstrates that the high-level hidden neurons of ANet automatically discover semantic concepts after pre-training with massive face identities, and such concepts are significantly enriched after fine-tuning with attribute tags. Each attribute can be well explained with a sparse linear combination of these concepts.",
"The way people look in terms of facial attributes (ethnicity, hair color, facial hair, etc.) and the clothes or accessories they wear (sunglasses, hat, hoodies, etc.) is highly dependent on geo-location and weather condition, respectively. This work explores, for the first time, the use of this contextual information, as people with wearable cameras walk across different neighborhoods of a city, in order to learn a rich feature representation for facial attribute classification, without the costly manual annotation required by previous methods. By tracking the faces of casual walkers on more than 40 hours of egocentric video, we are able to cover tens of thousands of different identities and automatically extract nearly 5 million pairs of images connected by or from different face tracks, along with their weather and location context, under pose and lighting variations. These image pairs are then fed into a deep network that preserves similarity of images connected by the same track, in order to capture identity-related attribute features, and optimizes for location and weather prediction to capture additional facial attribute features. Finally, the network is fine-tuned with manually annotated samples. We perform an extensive experimental analysis on wearable data and two standard benchmark datasets based on web images (LFWA and CelebA). Our method outperforms by a large margin a network trained from scratch. Moreover, even without using manually annotated identity labels for pre-training as in previous methods, our approach achieves results that are better than the state of the art.",
"We address the problem of cross-domain image retrieval, considering the following practical application: given a user photo depicting a clothing image, our goal is to retrieve the same or attribute-similar clothing items from online shopping stores. This is a challenging problem due to the large discrepancy between online shopping images, usually taken in ideal lighting pose background conditions, and user photos captured in uncontrolled conditions. To address this problem, we propose a Dual Attribute-aware Ranking Network (DARN) for retrieval feature learning. More specifically, DARN consists of two sub-networks, one for each domain, whose retrieval feature representations are driven by semantic attribute learning. We show that this attribute-guided learning is a key factor for retrieval accuracy improvement. In addition, to further align with the nature of the retrieval problem, we impose a triplet visual similarity constraint for learning to rank across the two sub-networks. Another contribution of our work is a large-scale dataset which makes the network learning feasible. We exploit customer review websites to crawl a large set of online shopping images and corresponding offline user photos with fine-grained clothing attributes, i.e., around 450,000 online shopping images and about 90,000 exact offline counterpart images of those online ones. All these images are collected from real-world consumer websites reflecting the diversity of the data modality, which makes this dataset unique and rare in the academic community. We extensively evaluate the retrieval performance of networks in different configurations. The top-20 retrieval accuracy is doubled when using the proposed DARN other than the current popular solution using pre-trained CNN features only (0.570 vs. 0.268).",
"We address the problem of describing people based on fine-grained clothing attributes. This is an important problem for many practical applications, such as identifying target suspects or finding missing people based on detailed clothing descriptions in surveillance videos or consumer photos. We approach this problem by first mining clothing images with fine-grained attribute labels from online shopping stores. A large-scale dataset is built with about one million images and fine-detailed attribute sub-categories, such as various shades of color (e.g., watermelon red, rosy red, purplish red), clothing types (e.g., down jacket, denim jacket), and patterns (e.g., thin horizontal stripes, houndstooth). As these images are taken in ideal pose lighting background conditions, it is unreliable to directly use them as training data for attribute prediction in the domain of unconstrained images captured, for example, by mobile phones or surveillance cameras. In order to bridge this gap, we propose a novel double-path deep domain adaptation network to model the data from the two domains jointly. Several alignment cost layers placed inbetween the two columns ensure the consistency of the two domain features and the feasibility to predict unseen attribute categories in one of the domains. Finally, to achieve a working system with automatic human body alignment, we trained an enhanced RCNN-based detector to localize human bodies in images. Our extensive experimental evaluation demonstrates the effectiveness of the proposed approach for describing people based on fine-grained clothing attributes.",
"This paper addresses the problem of human visual attribute recognition, i.e., the prediction of a fixed set of semantic attributes given an image of a person. Previous work often considered the different attributes independently from each other, without taking advantage of possible dependencies between them. In contrast, we propose a method to jointly train a CNN model for all attributes that can take advantage of those dependencies, considering as input only the image without additional external pose, part or context information. We report detailed experiments examining the contribution of individual aspects, which yields beneficial insights for other researchers. Our holistic CNN achieves superior performance on two publicly available attribute datasets improving on methods that additionally rely on pose-alignment or context. To support further evaluations, we present a novel dataset, based on realistic outdoor video sequences, that contains more than 27,000 pedestrians annotated with 10 attributes. Finally, we explore design options to embrace the N A labels inherently present in this task.",
"Attribute recognition, particularly facial, extracts many labels for each image. While some multi-task vision problems can be decomposed into separate tasks and stages, e.g., training independent models for each task, for a growing set of problems joint optimization across all tasks has been shown to improve performance. We show that for deep convolutional neural network (DCNN) facial attribute extraction, multi-task optimization is better. Unfortunately, it can be difficult to apply joint optimization to DCNNs when training data is imbalanced, and re-balancing multi-label data directly is structurally infeasible, since adding removing data to balance one label will change the sampling of the other labels. This paper addresses the multi-label imbalance problem by introducing a novel mixed objective optimization network (MOON) with a loss function that mixes multiple task objectives with domain adaptive re-weighting of propagated loss. Experiments demonstrate that not only does MOON advance the state of the art in facial attribute recognition, but it also outperforms independently trained DCNNs using the same data. When using facial attributes for the LFW face recognition task, we show that our balanced (domain adapted) network outperforms the unbalanced trained network."
]
} |
1611.05520 | 2555945617 | Action recognition and anticipation are key to the success of many computer vision applications. Existing methods can roughly be grouped into those that extract global, context-aware representations of the entire image or sequence, and those that aim at focusing on the regions where the action occurs. While the former may suffer from the fact that context is not always reliable, the latter completely ignore this source of information, which can nonetheless be helpful in many situations. In this paper, we aim at making the best of both worlds by developing an approach that leverages both context-aware and action-aware features. At the core of our method lies a novel multi-stage recurrent architecture that allows us to effectively combine these two sources of information throughout a video. This architecture first exploits the global, context-aware features, and merges the resulting representation with the localized, action-aware ones. Our experiments on standard datasets evidence the benefits of our approach over methods that use each information type separately. We outperform the state-of-the-art methods that, as us, rely only on RGB frames as input for both action recognition and anticipation. | Over the years, great progress has been made in activity recognition @cite_23 @cite_6 @cite_36 @cite_32 @cite_24 @cite_20 @cite_40 . Unsurprisingly, while earlier approaches relied on handcrafted features @cite_6 @cite_23 , recent ones have turned towards deep learning. Below, we focus on these approaches, which are most related to our work. | {
"cite_N": [
"@cite_36",
"@cite_32",
"@cite_6",
"@cite_24",
"@cite_40",
"@cite_23",
"@cite_20"
],
"mid": [
"2951183276",
"2471775118",
"2105101328",
"",
"",
"2020163092",
"2507009361"
],
"abstract": [
"Models based on deep convolutional networks have dominated recent image interpretation tasks; we investigate whether models which are also recurrent, or \"temporally deep\", are effective for tasks involving sequences, visual and otherwise. We develop a novel recurrent convolutional architecture suitable for large-scale visual learning which is end-to-end trainable, and demonstrate the value of these models on benchmark video recognition tasks, image description and retrieval problems, and video narration challenges. In contrast to current models which assume a fixed spatio-temporal receptive field or simple temporal averaging for sequential processing, recurrent convolutional models are \"doubly deep\"' in that they can be compositional in spatial and temporal \"layers\". Such models may have advantages when target concepts are complex and or training data are limited. Learning long-term dependencies is possible when nonlinearities are incorporated into the network state updates. Long-term RNN models are appealing in that they directly can map variable-length inputs (e.g., video frames) to variable length outputs (e.g., natural language text) and can model complex temporal dynamics; yet they can be optimized with backpropagation. Our recurrent long-term models are directly connected to modern visual convnet models and can be jointly trained to simultaneously learn temporal dynamics and convolutional perceptual representations. Our results show such models have distinct advantages over state-of-the-art models for recognition or generation which are separately defined and or optimized.",
"We present hierarchical rank pooling, a video sequence encoding method for activity recognition. It consists of a network of rank pooling functions which captures the dynamics of rich convolutional neural network features within a video sequence. By stacking non-linear feature functions and rank pooling over one another, we obtain a high capacity dynamic encoding mechanism, which is used for action recognition. We present a method for jointly learning the video representation and activity classifier parameters. Our method obtains state-of-the art results on three important activity recognition benchmarks: 76.7 on Hollywood2, 66.9 on HMDB51 and, 91.4 on UCF101.",
"Recently dense trajectories were shown to be an efficient video representation for action recognition and achieved state-of-the-art results on a variety of datasets. This paper improves their performance by taking into account camera motion to correct them. To estimate camera motion, we match feature points between frames using SURF descriptors and dense optical flow, which are shown to be complementary. These matches are, then, used to robustly estimate a homography with RANSAC. Human motion is in general different from camera motion and generates inconsistent matches. To improve the estimation, a human detector is employed to remove these matches. Given the estimated camera motion, we remove trajectories consistent with it. We also use this estimation to cancel out camera motion from the optical flow. This significantly improves motion-based descriptors, such as HOF and MBH. Experimental results on four challenging action datasets (i.e., Hollywood2, HMDB51, Olympic Sports and UCF50) significantly outperform the current state of the art.",
"",
"",
"Local image features or interest points provide compact and abstract representations of patterns in an image. In this paper, we extend the notion of spatial interest points into the spatio-temporal domain and show how the resulting features often reflect interesting events that can be used for a compact representation of video data as well as for interpretation of spatio-temporal events. To detect spatio-temporal events, we build on the idea of the Harris and Forstner interest point operators and detect local structures in space-time where the image values have significant local variations in both space and time. We estimate the spatio-temporal extents of the detected events by maximizing a normalized spatio-temporal Laplacian operator over spatial and temporal scales. To represent the detected events, we then compute local, spatio-temporal, scale-invariant N-jets and classify each event with respect to its jet descriptor. For the problem of human motion analysis, we illustrate how a video representation in terms of local space-time features allows for detection of walking people in scenes with occlusions and dynamic cluttered backgrounds.",
"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)."
]
} |
1611.05520 | 2555945617 | Action recognition and anticipation are key to the success of many computer vision applications. Existing methods can roughly be grouped into those that extract global, context-aware representations of the entire image or sequence, and those that aim at focusing on the regions where the action occurs. While the former may suffer from the fact that context is not always reliable, the latter completely ignore this source of information, which can nonetheless be helpful in many situations. In this paper, we aim at making the best of both worlds by developing an approach that leverages both context-aware and action-aware features. At the core of our method lies a novel multi-stage recurrent architecture that allows us to effectively combine these two sources of information throughout a video. This architecture first exploits the global, context-aware features, and merges the resulting representation with the localized, action-aware ones. Our experiments on standard datasets evidence the benefits of our approach over methods that use each information type separately. We outperform the state-of-the-art methods that, as us, rely only on RGB frames as input for both action recognition and anticipation. | In this deep learning context, many methods rely on CNNs @cite_35 @cite_34 @cite_45 @cite_10 @cite_17 to extract a global representation of images. These CNN-based methods, however, typically have small temporal support, and thus fail to capture long-range dynamics. For instance, the two-stream networks @cite_20 @cite_45 @cite_39 act on single images in conjunction with optical flow information to model the temporal information. While 3D convolutional filters have been proposed @cite_35 , they are typically limited to acting on small sets of stacked video frames, 10 to 20 in practice. | {
"cite_N": [
"@cite_35",
"@cite_39",
"@cite_45",
"@cite_34",
"@cite_10",
"@cite_20",
"@cite_17"
],
"mid": [
"2952633803",
"2952186347",
"2342662179",
"2308045930",
"2160253527",
"2507009361",
""
],
"abstract": [
"We propose a simple, yet effective approach for spatiotemporal feature learning using deep 3-dimensional convolutional networks (3D ConvNets) trained on a large scale supervised video dataset. Our findings are three-fold: 1) 3D ConvNets are more suitable for spatiotemporal feature learning compared to 2D ConvNets; 2) A homogeneous architecture with small 3x3x3 convolution kernels in all layers is among the best performing architectures for 3D ConvNets; and 3) Our learned features, namely C3D (Convolutional 3D), with a simple linear classifier outperform state-of-the-art methods on 4 different benchmarks and are comparable with current best methods on the other 2 benchmarks. In addition, the features are compact: achieving 52.8 accuracy on UCF101 dataset with only 10 dimensions and also very efficient to compute due to the fast inference of ConvNets. Finally, they are conceptually very simple and easy to train and use.",
"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.",
"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.",
"",
"A descriptor for describing action in terms of shape and motion is proposed.The descriptor is employed in the online action recognition scenario.Two runtime optimizations are introduced to enable applicability in real time.The method is simple, fast and accurate enough to be of practical value. We propose a novel approach for online action recognition. The action is represented in a low dimensional (15D) space using a covariance descriptor of shape and motion features - spatio-temporal coordinates and optical flow of pixels belonging to extracted silhouettes. We analyze the applicability of the descriptor for online scenarios where action classification is performed based on incomplete spatio-temporal volumes. In order to enable our online action classification algorithm to be applied in real time, we introduce two modifications, namely the incremental covariance update and the on demand nearest neighbor classification. In our experiments we use quality measures, such as latency, especially designed for the online scenario to report the algorithm's performance. We evaluate the performance of our descriptor on standard, publicly available datasets for gesture recognition, namely the Cambridge-Gestures dataset and the ChaLearn One-Shot-Learning dataset and show that its performance is comparable to the state-of-the-art despite its relative simplicity. The evaluation on the UCF-101 action recognition dataset demonstrates that the descriptor is applicable in challenging unconstrained environments.",
"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).",
""
]
} |
1611.05520 | 2555945617 | Action recognition and anticipation are key to the success of many computer vision applications. Existing methods can roughly be grouped into those that extract global, context-aware representations of the entire image or sequence, and those that aim at focusing on the regions where the action occurs. While the former may suffer from the fact that context is not always reliable, the latter completely ignore this source of information, which can nonetheless be helpful in many situations. In this paper, we aim at making the best of both worlds by developing an approach that leverages both context-aware and action-aware features. At the core of our method lies a novel multi-stage recurrent architecture that allows us to effectively combine these two sources of information throughout a video. This architecture first exploits the global, context-aware features, and merges the resulting representation with the localized, action-aware ones. Our experiments on standard datasets evidence the benefits of our approach over methods that use each information type separately. We outperform the state-of-the-art methods that, as us, rely only on RGB frames as input for both action recognition and anticipation. | By contrast, recurrent architectures, such as the popular Long-Short Term Memory networks @cite_1 , can, in principle, learn complex, long-range dynamics, and have therefore recently been investigated for action recognition @cite_36 @cite_15 @cite_19 @cite_40 @cite_26 @cite_19 . For instance, in @cite_36 , an LSTM was employed to model the dynamics of CNN activations; in @cite_40 , a bi-directional LSTM was combined with a multi-stream CNN to encode the long-term dynamics within and between activities in videos. Other works, such as @cite_15 , have proposed to exploit additional annotations, in the form of 3D skeletons, into an LSTM-based model. Such annotations, however, are not always available in practice, thus limiting the applicability of these methods. | {
"cite_N": [
"@cite_26",
"@cite_36",
"@cite_1",
"@cite_19",
"@cite_40",
"@cite_15"
],
"mid": [
"2510185399",
"2951183276",
"",
"2952453038",
"",
"2462496837"
],
"abstract": [
"3D action recognition – analysis of human actions based on 3D skeleton data – becomes popular recently due to its succinctness, robustness, and view-invariant representation. Recent attempts on this problem suggested to develop RNN-based learning methods to model the contextual dependency in the temporal domain. In this paper, we extend this idea to spatio-temporal domains to analyze the hidden sources of action-related information within the input data over both domains concurrently. Inspired by the graphical structure of the human skeleton, we further propose a more powerful tree-structure based traversal method. To handle the noise and occlusion in 3D skeleton data, we introduce new gating mechanism within LSTM to learn the reliability of the sequential input data and accordingly adjust its effect on updating the long-term context information stored in the memory cell. Our method achieves state-of-the-art performance on 4 challenging benchmark datasets for 3D human action analysis.",
"Models based on deep convolutional networks have dominated recent image interpretation tasks; we investigate whether models which are also recurrent, or \"temporally deep\", are effective for tasks involving sequences, visual and otherwise. We develop a novel recurrent convolutional architecture suitable for large-scale visual learning which is end-to-end trainable, and demonstrate the value of these models on benchmark video recognition tasks, image description and retrieval problems, and video narration challenges. In contrast to current models which assume a fixed spatio-temporal receptive field or simple temporal averaging for sequential processing, recurrent convolutional models are \"doubly deep\"' in that they can be compositional in spatial and temporal \"layers\". Such models may have advantages when target concepts are complex and or training data are limited. Learning long-term dependencies is possible when nonlinearities are incorporated into the network state updates. Long-term RNN models are appealing in that they directly can map variable-length inputs (e.g., video frames) to variable length outputs (e.g., natural language text) and can model complex temporal dynamics; yet they can be optimized with backpropagation. Our recurrent long-term models are directly connected to modern visual convnet models and can be jointly trained to simultaneously learn temporal dynamics and convolutional perceptual representations. Our results show such models have distinct advantages over state-of-the-art models for recognition or generation which are separately defined and or optimized.",
"",
"We use multilayer Long Short Term Memory (LSTM) networks to learn representations of video sequences. Our model uses an encoder LSTM to map an input sequence into a fixed length representation. This representation is decoded using single or multiple decoder LSTMs to perform different tasks, such as reconstructing the input sequence, or predicting the future sequence. We experiment with two kinds of input sequences - patches of image pixels and high-level representations (\"percepts\") of video frames extracted using a pretrained convolutional net. We explore different design choices such as whether the decoder LSTMs should condition on the generated output. We analyze the outputs of the model qualitatively to see how well the model can extrapolate the learned video representation into the future and into the past. We try to visualize and interpret the learned features. We stress test the model by running it on longer time scales and on out-of-domain data. We further evaluate the representations by finetuning them for a supervised learning problem - human action recognition on the UCF-101 and HMDB-51 datasets. We show that the representations help improve classification accuracy, especially when there are only a few training examples. Even models pretrained on unrelated datasets (300 hours of YouTube videos) can help action recognition performance.",
"",
"This paper argues that large-scale action recognition in video can be greatly improved by providing an additional modality in training data – namely, 3D human-skeleton sequences – aimed at complementing poorly represented or missing features of human actions in the training videos. For recognition, we use Long Short Term Memory (LSTM) grounded via a deep Convolutional Neural Network (CNN) onto the video. Training of LSTM is regularized using the output of another encoder LSTM (eLSTM) grounded on 3D human-skeleton training data. For such regularized training of LSTM, we modify the standard backpropagation through time (BPTT) in order to address the wellknown issues with gradient descent in constraint optimization. Our evaluation on three benchmark datasets – Sports-1M, HMDB-51, and UCF101 – shows accuracy improvements from 1.7 up to 14.8 relative to the state of the art."
]
} |
1611.05520 | 2555945617 | Action recognition and anticipation are key to the success of many computer vision applications. Existing methods can roughly be grouped into those that extract global, context-aware representations of the entire image or sequence, and those that aim at focusing on the regions where the action occurs. While the former may suffer from the fact that context is not always reliable, the latter completely ignore this source of information, which can nonetheless be helpful in many situations. In this paper, we aim at making the best of both worlds by developing an approach that leverages both context-aware and action-aware features. At the core of our method lies a novel multi-stage recurrent architecture that allows us to effectively combine these two sources of information throughout a video. This architecture first exploits the global, context-aware features, and merges the resulting representation with the localized, action-aware ones. Our experiments on standard datasets evidence the benefits of our approach over methods that use each information type separately. We outperform the state-of-the-art methods that, as us, rely only on RGB frames as input for both action recognition and anticipation. | Beyond recurrent models, rank pooling has also proven effective to model activities in videos @cite_37 @cite_33 @cite_32 . In this context, @cite_37 computes a representation encoding the dynamics of the video, and @cite_33 introduces the concept of Dynamic Images to summarize the gist of a sequence. | {
"cite_N": [
"@cite_37",
"@cite_32",
"@cite_33"
],
"mid": [
"2193384753",
"2471775118",
"2462996230"
],
"abstract": [
"We propose a function-based temporal pooling method that captures the latent structure of the video sequence data - e.g., how frame-level features evolve over time in a video. We show how the parameters of a function that has been fit to the video data can serve as a robust new video representation. As a specific example, we learn a pooling function via ranking machines. By learning to rank the frame-level features of a video in chronological order, we obtain a new representation that captures the video-wide temporal dynamics of a video, suitable for action recognition. Other than ranking functions, we explore different parametric models that could also explain the temporal changes in videos. The proposed functional pooling methods, and rank pooling in particular, is easy to interpret and implement, fast to compute and effective in recognizing a wide variety of actions. We evaluate our method on various benchmarks for generic action, fine-grained action and gesture recognition. Results show that rank pooling brings an absolute improvement of 7-10 average pooling baseline. At the same time, rank pooling is compatible with and complementary to several appearance and local motion based methods and features, such as improved trajectories and deep learning features.",
"We present hierarchical rank pooling, a video sequence encoding method for activity recognition. It consists of a network of rank pooling functions which captures the dynamics of rich convolutional neural network features within a video sequence. By stacking non-linear feature functions and rank pooling over one another, we obtain a high capacity dynamic encoding mechanism, which is used for action recognition. We present a method for jointly learning the video representation and activity classifier parameters. Our method obtains state-of-the art results on three important activity recognition benchmarks: 76.7 on Hollywood2, 66.9 on HMDB51 and, 91.4 on UCF101.",
"We introduce the concept of dynamic image, a novel compact representation of videos useful for video analysis especially when convolutional neural networks (CNNs) are used. The dynamic image is based on the rank pooling concept and is obtained through the parameters of a ranking machine that encodes the temporal evolution of the frames of the video. Dynamic images are obtained by directly applying rank pooling on the raw image pixels of a video producing a single RGB image per video. This idea is simple but powerful as it enables the use of existing CNN models directly on video data with fine-tuning. We present an efficient and effective approximate rank pooling operator, speeding it up orders of magnitude compared to rank pooling. Our new approximate rank pooling CNN layer allows us to generalize dynamic images to dynamic feature maps and we demonstrate the power of our new representations on standard benchmarks in action recognition achieving state-of-the-art performance."
]
} |
1611.05520 | 2555945617 | Action recognition and anticipation are key to the success of many computer vision applications. Existing methods can roughly be grouped into those that extract global, context-aware representations of the entire image or sequence, and those that aim at focusing on the regions where the action occurs. While the former may suffer from the fact that context is not always reliable, the latter completely ignore this source of information, which can nonetheless be helpful in many situations. In this paper, we aim at making the best of both worlds by developing an approach that leverages both context-aware and action-aware features. At the core of our method lies a novel multi-stage recurrent architecture that allows us to effectively combine these two sources of information throughout a video. This architecture first exploits the global, context-aware features, and merges the resulting representation with the localized, action-aware ones. Our experiments on standard datasets evidence the benefits of our approach over methods that use each information type separately. We outperform the state-of-the-art methods that, as us, rely only on RGB frames as input for both action recognition and anticipation. | This, in essence, is the goal of methods based on the notion of actionness @cite_27 @cite_9 @cite_14 @cite_46 @cite_2 @cite_28 . Inspired by the concept of objectness @cite_41 @cite_31 , commonly used in object detection, actionness aims at localizing the regions in a video where an action of interest occurs. @cite_27 , this was achieved by exploiting appearance (RGB) and motion (optical flow) in a two-stream architecture. @cite_27 , the resulting actionness map was then employed to generate action bounding box proposals via an action detection framework based on @cite_3 , and classifying these proposals. The ActionTube approach of @cite_21 follows a similar framework, but relies on @cite_42 instead of @cite_3 . More importantly, by focusing on the actions themselves, these methods throw away all the information about context. However, in many scenarios, such as to recognize different sports, context provides helpful information about the observed actions. Note that for extracting actionness in @cite_27 @cite_21 , bounding box annotation are used as an extra supervision during the training process, while our approach requires no additional annotations. | {
"cite_N": [
"@cite_31",
"@cite_14",
"@cite_28",
"@cite_41",
"@cite_9",
"@cite_21",
"@cite_42",
"@cite_3",
"@cite_27",
"@cite_2",
"@cite_46"
],
"mid": [
"",
"2018068650",
"",
"2066624635",
"",
"2484328966",
"",
"2102605133",
"2410118306",
"",
"1945129080"
],
"abstract": [
"",
"This paper considers the problem of action localization, where the objective is to determine when and where certain actions appear. We introduce a sampling strategy to produce 2D+t sequences of bounding boxes, called tubelets. Compared to state-of-the-art alternatives, this drastically reduces the number of hypotheses that are likely to include the action of interest. Our method is inspired by a recent technique introduced in the context of image localization. Beyond considering this technique for the first time for videos, we revisit this strategy for 2D+t sequences obtained from super-voxels. Our sampling strategy advantageously exploits a criterion that reflects how action related motion deviates from background motion. We demonstrate the interest of our approach by extensive experiments on two public datasets: UCF Sports and MSR-II. Our approach significantly outperforms the state-of-the-art on both datasets, while restricting the search of actions to a fraction of possible bounding box sequences.",
"",
"We present a generic objectness measure, quantifying how likely it is for an image window to contain an object of any class. We explicitly train it to distinguish objects with a well-defined boundary in space, such as cows and telephones, from amorphous background elements, such as grass and road. The measure combines in a Bayesian framework several image cues measuring characteristics of objects, such as appearing different from their surroundings and having a closed boundary. These include an innovative cue to measure the closed boundary characteristic. In experiments on the challenging PASCAL VOC 07 dataset, we show this new cue to outperform a state-of-the-art saliency measure, and the combined objectness measure to perform better than any cue alone. We also compare to interest point operators, a HOG detector, and three recent works aiming at automatic object segmentation. Finally, we present two applications of objectness. In the first, we sample a small numberof windows according to their objectness probability and give an algorithm to employ them as location priors for modern class-specific object detectors. As we show experimentally, this greatly reduces the number of windows evaluated by the expensive class-specific model. In the second application, we use objectness as a complementary score in addition to the class-specific model, which leads to fewer false positives. As shown in several recent papers, objectness can act as a valuable focus of attention mechanism in many other applications operating on image windows, including weakly supervised learning of object categories, unsupervised pixelwise segmentation, and object tracking in video. Computing objectness is very efficient and takes only about 4 sec. per image.",
"",
"In this work, we propose an approach to the spatiotemporal localisation (detection) and classification of multiple concurrent actions within temporally untrimmed videos. Our framework is composed of three stages. In stage 1, appearance and motion detection networks are employed to localise and score actions from colour images and optical flow. In stage 2, the appearance network detections are boosted by combining them with the motion detection scores, in proportion to their respective spatial overlap. In stage 3, sequences of detection boxes most likely to be associated with a single action instance, called action tubes, are constructed by solving two energy maximisation problems via dynamic programming. While in the first pass, action paths spanning the whole video are built by linking detection boxes over time using their class-specific scores and their spatial overlap, in the second pass, temporal trimming is performed by ensuring label consistency for all constituting detection boxes. We demonstrate the performance of our algorithm on the challenging UCF101, J-HMDB-21 and LIRIS-HARL datasets, achieving new state-of-the-art results across the board and significantly increasing detection speed at test time. We achieve a huge leap forward in action detection performance and report a 20 and 11 gain in mAP (mean average precision) on UCF-101 and J-HMDB-21 datasets respectively when compared to the state-of-the-art.",
"",
"Object detection performance, as measured on the canonical PASCAL VOC dataset, has plateaued in the last few years. The best-performing methods are complex ensemble systems that typically combine multiple low-level image features with high-level context. In this paper, we propose a simple and scalable detection algorithm that improves mean average precision (mAP) by more than 30 relative to the previous best result on VOC 2012 -- achieving a mAP of 53.3 . Our approach combines two key insights: (1) one can apply high-capacity convolutional neural networks (CNNs) to bottom-up region proposals in order to localize and segment objects and (2) when labeled training data is scarce, supervised pre-training for an auxiliary task, followed by domain-specific fine-tuning, yields a significant performance boost. Since we combine region proposals with CNNs, we call our method R-CNN: Regions with CNN features. We also present experiments that provide insight into what the network learns, revealing a rich hierarchy of image features. Source code for the complete system is available at http: www.cs.berkeley.edu rbg rcnn.",
"Actionness was introduced to quantify the likelihood of containing a generic action instance at a specific location. Accurate and efficient estimation of actionness is important in video analysis and may benefit other relevant tasks such as action recognition and action detection. This paper presents a new deep architecture for actionness estimation, called hybrid fully convolutional network (H-FCN), which is composed of appearance FCN (A-FCN) and motion FCN (M-FCN). These two FCNs leverage the strong capacity of deep models to estimate actionness maps from the perspectives of static appearance and dynamic motion, respectively. In addition, the fully convolutional nature of H-FCN allows it to efficiently process videos with arbitrary sizes. Experiments are conducted on the challenging datasets of Stanford40, UCF Sports, and JHMDB to verify the effectiveness of H-FCN on actionness estimation, which demonstrate that our method achieves superior performance to previous ones. Moreover, we apply the estimated actionness maps on action proposal generation and action detection. Our actionness maps advance the current state-of-the-art performance of these tasks substantially.",
"",
"In this paper we target at generating generic action proposals in unconstrained videos. Each action proposal corresponds to a temporal series of spatial bounding boxes, i.e., a spatio-temporal video tube, which has a good potential to locate one human action. Assuming each action is performed by a human with meaningful motion, both appearance and motion cues are utilized to measure the actionness of the video tubes. After picking those spatiotemporal paths of high actionness scores, our action proposal generation is formulated as a maximum set coverage problem, where greedy search is performed to select a set of action proposals that can maximize the overall actionness score. Compared with existing action proposal approaches, our action proposals do not rely on video segmentation and can be generated in nearly real-time. Experimental results on two challenging datasets, MSRII and UCF 101, validate the superior performance of our action proposals as well as competitive results on action detection and search."
]
} |
1611.05402 | 2598713166 | Recently there has been significant interest in training machine-learning models at low precision: by reducing precision, one can reduce computation and communication by one order of magnitude. We examine training at reduced precision, both from a theoretical and practical perspective, and ask: is it possible to train models at end-to-end low precision with provable guarantees? Can this lead to consistent order-of-magnitude speedups? We present a framework called ZipML to answer these questions. For linear models, the answer is yes. We develop a simple framework based on one simple but novel strategy called double sampling. Our framework is able to execute training at low precision with no bias, guaranteeing convergence, whereas naive quantization would introduce significant bias. We validate our framework across a range of applications, and show that it enables an FPGA prototype that is up to 6.5x faster than an implementation using full 32-bit precision. We further develop a variance-optimal stochastic quantization strategy and show that it can make a significant difference in a variety of settings. When applied to linear models together with double sampling, we save up to another 1.7x in data movement compared with uniform quantization. When training deep networks with quantized models, we achieve higher accuracy than the state-of-the-art XNOR-Net. Finally, we extend our framework through approximation to non-linear models, such as SVM. We show that, although using low-precision data induces bias, we can appropriately bound and control the bias. We find in practice 8-bit precision is often sufficient to converge to the correct solution. Interestingly, however, in practice we notice that our framework does not always outperform the naive rounding approach. We discuss this negative result in detail. | Low-precision training of deep neural networks has been studied intensively and many heuristics work well for a subset of networks. OneBit SGD @cite_4 provides a gradient compression heuristic developed in the context of deep neural networks for speech recognition. There are successful applications of end-to-end quantization to training neural networks that result in little to no quality loss @cite_17 @cite_13 @cite_20 @cite_9 @cite_28 @cite_15 . They quantize weights, activations, and gradients to low precision (e.g., 1-bit) and revise the backpropagation algorithm to be aware of the quantization function. The empirical success of this work inspired this paper, in which we try to provide a theoretical understanding of end-to-end low-precision training for machine learning models. Another line of research concerns inference and model compression of a pre-trained model @cite_11 @cite_24 @cite_1 @cite_25 @cite_12 @cite_22 @cite_14 . In this paper, we focus on training and leave the study of inference for future work. | {
"cite_N": [
"@cite_11",
"@cite_4",
"@cite_22",
"@cite_14",
"@cite_28",
"@cite_9",
"@cite_1",
"@cite_24",
"@cite_12",
"@cite_15",
"@cite_13",
"@cite_25",
"@cite_20",
"@cite_17"
],
"mid": [
"587794757",
"2407022425",
"2177847924",
"2233116163",
"2405920868",
"2291160084",
"2119144962",
"1724438581",
"2266701264",
"1841592590",
"",
"2952936791",
"2469490737",
"2950894517"
],
"abstract": [
"Recent advances in deep learning have made the use of large, deep neural networks with tens of millions of parameters suitable for a number of applications that require real-time processing. The sheer size of these networks can represent a challenging computational burden, even for modern CPUs. For this reason, GPUs are routinely used instead to train and run such networks. This paper is a tutorial for students and researchers on some of the techniques that can be used to reduce this computational cost considerably on modern x86 CPUs. We emphasize data layout, batching of the computation, the use of SSE2 instructions, and particularly leverage SSSE3 and SSE4 fixed-point instructions which provide a 3× improvement over an optimized floating-point baseline. We use speech recognition as an example task, and show that a real-time hybrid hidden Markov model neural network (HMM NN) large vocabulary system can be built with a 10× speedup over an unoptimized baseline and a 4× speedup over an aggressively optimized floating-point baseline at no cost in accuracy. The techniques described extend readily to neural network training and provide an effective alternative to the use of specialized hardware.",
"",
"Although the latest high-end smartphone has powerful CPU and GPU, running deeper convolutional neural networks (CNNs) for complex tasks such as ImageNet classification on mobile devices is challenging. To deploy deep CNNs on mobile devices, we present a simple and effective scheme to compress the entire CNN, which we call one-shot whole network compression. The proposed scheme consists of three steps: (1) rank selection with variational Bayesian matrix factorization, (2) Tucker decomposition on kernel tensor, and (3) fine-tuning to recover accumulated loss of accuracy, and each step can be easily implemented using publicly available tools. We demonstrate the effectiveness of the proposed scheme by testing the performance of various compressed CNNs (AlexNet, VGGS, GoogLeNet, and VGG-16) on the smartphone. Significant reductions in model size, runtime, and energy consumption are obtained, at the cost of small loss in accuracy. In addition, we address the important implementation level issue on 1?1 convolution, which is a key operation of inception module of GoogLeNet as well as CNNs compressed by our proposed scheme.",
"Recently, convolutional neural networks (CNN) have demonstrated impressive performance in various computer vision tasks. However, high performance hardware is typically indispensable for the application of CNN models due to the high computation complexity, which prohibits their further extensions. In this paper, we propose an efficient framework, namely Quantized CNN, to simultaneously speed-up the computation and reduce the storage and memory overhead of CNN models. Both filter kernels in convolutional layers and weighting matrices in fully-connected layers are quantized, aiming at minimizing the estimation error of each layer's response. Extensive experiments on the ILSVRC-12 benchmark demonstrate 4 6× speed-up and 15 20× compression with merely one percentage loss of classification accuracy. With our quantized CNN model, even mobile devices can accurately classify images within one second.",
"We introduce ternary weight networks (TWNs) - neural networks with weights constrained to +1, 0 and -1. The Euclidian distance between full (float or double) precision weights and the ternary weights along with a scaling factor is minimized. Besides, a threshold-based ternary function is optimized to get an approximated solution which can be fast and easily computed. TWNs have stronger expressive abilities than the recently proposed binary precision counterparts and are thus more effective than the latter. Meanwhile, TWNs achieve up to 16 @math or 32 @math model compression rate and need fewer multiplications compared with the full precision counterparts. Benchmarks on MNIST, CIFAR-10, and large scale ImageNet datasets show that the performance of TWNs is only slightly worse than the full precision counterparts but outperforms the analogous binary precision counterparts a lot.",
"Recent advances in convolutional neural networks have considered model complexity and hardware efficiency to enable deployment onto embedded systems and mobile devices. For example, it is now well-known that the arithmetic operations of deep networks can be encoded down to 8-bit fixed-point without significant deterioration in performance. However, further reduction in precision down to as low as 3-bit fixed-point results in significant losses in performance. In this paper we propose a new data representation that enables state-of-the-art networks to be encoded to 3 bits with negligible loss in classification performance. To perform this, we take advantage of the fact that the weights and activations in a trained network naturally have non-uniform distributions. Using non-uniform, base-2 logarithmic representation to encode weights, communicate activations, and perform dot-products enables networks to 1) achieve higher classification accuracies than fixed-point at the same resolution and 2) eliminate bulky digital multipliers. Finally, we propose an end-to-end training procedure that uses log representation at 5-bits, which achieves higher final test accuracy than linear at 5-bits.",
"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 convolutional neural networks (CNN) has become the most promising method for object recognition, repeatedly demonstrating record breaking results for image classification and object detection in recent years. However, a very deep CNN generally involves many layers with millions of parameters, making the storage of the network model to be extremely large. This prohibits the usage of deep CNNs on resource limited hardware, especially cell phones or other embedded devices. In this paper, we tackle this model storage issue by investigating information theoretical vector quantization methods for compressing the parameters of CNNs. In particular, we have found in terms of compressing the most storage demanding dense connected layers, vector quantization methods have a clear gain over existing matrix factorization methods. Simply applying k-means clustering to the weights or conducting product quantization can lead to a very good balance between model size and recognition accuracy. For the 1000-category classification task in the ImageNet challenge, we are able to achieve 16-24 times compression of the network with only 1 loss of classification accuracy using the state-of-the-art CNN.",
"Based on the assumption that there exists a neural network that efficiently represents a set of Boolean functions between all binary inputs and outputs, we propose a process for developing and deploying neural networks whose weight parameters, bias terms, input, and intermediate hidden layer output signals, are all binary-valued, and require only basic bit logic for the feedforward pass. The proposed Bitwise Neural Network (BNN) is especially suitable for resource-constrained environments, since it replaces either floating or fixed-point arithmetic with significantly more efficient bitwise operations. Hence, the BNN requires for less spatial complexity, less memory bandwidth, and less power consumption in hardware. In order to design such networks, we propose to add a few training schemes, such as weight compression and noisy backpropagation, which result in a bitwise network that performs almost as well as its corresponding real-valued network. We test the proposed network on the MNIST dataset, represented using binary features, and show that BNNs result in competitive performance while offering dramatic computational savings.",
"Training of large-scale deep neural networks is often constrained by the available computational resources. We study the effect of limited precision data representation and computation on neural network training. Within the context of low-precision fixed-point computations, we observe the rounding scheme to play a crucial role in determining the network's behavior during training. Our results show that deep networks can be trained using only 16-bit wide fixed-point number representation when using stochastic rounding, and incur little to no degradation in the classification accuracy. We also demonstrate an energy-efficient hardware accelerator that implements low-precision fixed-point arithmetic with stochastic rounding.",
"",
"In recent years increasingly complex architectures for deep convolution networks (DCNs) have been proposed to boost the performance on image recognition tasks. However, the gains in performance have come at a cost of substantial increase in computation and model storage resources. Fixed point implementation of DCNs has the potential to alleviate some of these complexities and facilitate potential deployment on embedded hardware. In this paper, we propose a quantizer design for fixed point implementation of DCNs. We formulate and solve an optimization problem to identify optimal fixed point bit-width allocation across DCN layers. Our experiments show that in comparison to equal bit-width settings, the fixed point DCNs with optimized bit width allocation offer >20 reduction in the model size without any loss in accuracy on CIFAR-10 benchmark. We also demonstrate that fine-tuning can further enhance the accuracy of fixed point DCNs beyond that of the original floating point model. In doing so, we report a new state-of-the-art fixed point performance of 6.78 error-rate on CIFAR-10 benchmark.",
"We propose DoReFa-Net, a method to train convolutional neural networks that have low bitwidth weights and activations using low bitwidth parameter gradients. In particular, during backward pass, parameter gradients are stochastically quantized to low bitwidth numbers before being propagated to convolutional layers. As convolutions during forward backward passes can now operate on low bitwidth weights and activations gradients respectively, DoReFa-Net can use bit convolution kernels to accelerate both training and inference. Moreover, as bit convolutions can be efficiently implemented on CPU, FPGA, ASIC and GPU, DoReFa-Net opens the way to accelerate training of low bitwidth neural network on these hardware. Our experiments on SVHN and ImageNet datasets prove that DoReFa-Net can achieve comparable prediction accuracy as 32-bit counterparts. For example, a DoReFa-Net derived from AlexNet that has 1-bit weights, 2-bit activations, can be trained from scratch using 6-bit gradients to get 46.1 top-1 accuracy on ImageNet validation set. The DoReFa-Net AlexNet model is released publicly.",
"We introduce a method to train Quantized Neural Networks (QNNs) --- neural networks with extremely low precision (e.g., 1-bit) weights and activations, at run-time. At train-time the quantized weights and activations are used for computing the parameter gradients. During the forward pass, QNNs drastically reduce memory size and accesses, and replace most arithmetic operations with bit-wise operations. As a result, power consumption is expected to be drastically reduced. We trained QNNs over the MNIST, CIFAR-10, SVHN and ImageNet datasets. The resulting QNNs achieve prediction accuracy comparable to their 32-bit counterparts. For example, our quantized version of AlexNet with 1-bit weights and 2-bit activations achieves @math top-1 accuracy. Moreover, we quantize the parameter gradients to 6-bits as well which enables gradients computation using only bit-wise operation. Quantized recurrent neural networks were tested over the Penn Treebank dataset, and achieved comparable accuracy as their 32-bit counterparts using only 4-bits. Last but not least, we programmed a binary matrix multiplication GPU kernel with which it is possible to run our MNIST QNN 7 times faster than with an unoptimized GPU kernel, without suffering any loss in classification accuracy. The QNN code is available online."
]
} |
1611.05402 | 2598713166 | Recently there has been significant interest in training machine-learning models at low precision: by reducing precision, one can reduce computation and communication by one order of magnitude. We examine training at reduced precision, both from a theoretical and practical perspective, and ask: is it possible to train models at end-to-end low precision with provable guarantees? Can this lead to consistent order-of-magnitude speedups? We present a framework called ZipML to answer these questions. For linear models, the answer is yes. We develop a simple framework based on one simple but novel strategy called double sampling. Our framework is able to execute training at low precision with no bias, guaranteeing convergence, whereas naive quantization would introduce significant bias. We validate our framework across a range of applications, and show that it enables an FPGA prototype that is up to 6.5x faster than an implementation using full 32-bit precision. We further develop a variance-optimal stochastic quantization strategy and show that it can make a significant difference in a variety of settings. When applied to linear models together with double sampling, we save up to another 1.7x in data movement compared with uniform quantization. When training deep networks with quantized models, we achieve higher accuracy than the state-of-the-art XNOR-Net. Finally, we extend our framework through approximation to non-linear models, such as SVM. We show that, although using low-precision data induces bias, we can appropriately bound and control the bias. We find in practice 8-bit precision is often sufficient to converge to the correct solution. Interestingly, however, in practice we notice that our framework does not always outperform the naive rounding approach. We discuss this negative result in detail. | Precision of data representation is a key design decision for configurable hardwares such as FPGA. There have been attempts to lower the precision when training on such hardware @cite_3 . These results are mostly empirical; we aim at providing a theoretical understanding, which enables new algorithms. | {
"cite_N": [
"@cite_3"
],
"mid": [
"2162369476"
],
"abstract": [
"X-ray computed tomography (CT) images could be improved using iterative image reconstruction if the 3D cone-beam forward- and back-projection computations can be accelerated significantly. We investigated the feasibility of a field-programmable gate array (FPGA) implementation of the separable footprint (SF) forward projector. A 16-bit fixed-point quantization introduces negligible numerical errors without affecting the perceptual image quality. The SF-based 3D cone-beam projector can be efficiently parallelized and its memory bandwidth reduced by exploiting projection geometry and data locality. We demonstrate a fully pipelined, 75-way parallel hardware architecture of the SF forward projector on a Xilinx Virtex-5 FPGA that can complete one forward projection of a 320×320×61 object over 3,625 views in 6.3 seconds."
]
} |
1611.05362 | 2575352857 | Software-defined networking is considered a promising new paradigm, enabling more reliable and formally verifiable communication networks. However, this paper shows that the separation of the control plane from the data plane, which lies at the heart of Software-Defined Networks (SDNs), introduces a new vulnerability which we call teleportation. An attacker (e.g., a malicious switch in the data plane or a host connected to the network) can use teleportation to transmit information via the control plane and bypass critical network functions in the data plane (e.g., a firewall), and to violate security policies as well as logical and even physical separations. This paper characterizes the design space for teleportation attacks theoretically, and then identifies four different teleportation techniques. We demonstrate and discuss how these techniques can be exploited for different attacks (e.g., exfiltrating confidential data at high rates), and also initiate the discussion of possible countermeasures. Generally, and given today's trend toward more intent-based networking, we believe that our findings are relevant beyond the use cases considered in this paper. | While researchers have already pointed out several interesting novel challenges in providing a correct operation of networks with separate data and control planes @cite_24 @cite_34 @cite_16 , it is generally believed that SDN has the potential to render computer networking more verifiable @cite_28 @cite_0 and even secure @cite_35 @cite_18 @cite_38 @cite_39 @cite_23 . | {
"cite_N": [
"@cite_35",
"@cite_18",
"@cite_38",
"@cite_28",
"@cite_16",
"@cite_39",
"@cite_24",
"@cite_0",
"@cite_23",
"@cite_34"
],
"mid": [
"1583113493",
"2119246371",
"2137845741",
"1882012874",
"1917945289",
"2155624544",
"2030814077",
"1675033504",
"2160824842",
"2141988289"
],
"abstract": [
"Despite their exponential growth, home and small office home office networks continue to be poorly managed. Consequently, security of hosts in most home networks is easily compromised and these hosts are in turn used for largescale malicious activities without the home users' knowledge. We argue that the advent of Software Defined Networking (SDN) provides a unique opportunity to effectively detect and contain network security problems in home and home office networks. We show how four prominent traffic anomaly detection algorithms can be implemented in an SDN context using Openflow compliant switches and NOX as a controller. Our experiments indicate that these algorithms are significantly more accurate in identifying malicious activities in the home networks as compared to the ISP. Furthermore, the efficiency analysis of our SDN implementations on a programmable home network router indicates that the anomaly detectors can operate at line rates without introducing any performance penalties for the home network traffic.",
"Providers of \"Infrastructure-as-a-Service\" need datacenter networks that support multi-tenancy, scale, and ease of operation, at low cost. Most existing network architectures cannot meet all of these needs simultaneously. In this paper we present NetLord, a novel multi-tenant network architecture. NetLord provides tenants with simple and flexible network abstractions, by fully and efficiently virtualizing the address space at both L2 and L3. NetLord can exploit inexpensive commodity equipment to scale the network to several thousands of tenants and millions of virtual machines. NetLord requires only a small amount of offline, one-time configuration. We implemented NetLord on a testbed, and demonstrated its scalability, while achieving order-of-magnitude goodput improvements over previous approaches.",
"Software-defined networks facilitate rapid and open innovation at the network control layer by providing a programmable network infrastructure for computing flow policies on demand. However, the dynamism of programmable networks also introduces new security challenges that demand innovative solutions. A critical challenge is efficient detection and reconciliation of potentially conflicting flow rules imposed by dynamic OpenFlow (OF) applications. To that end, we introduce FortNOX, a software extension that provides role-based authorization and security constraint enforcement for the NOX OpenFlow controller. FortNOX enables NOX to check flow rule contradictions in real time, and implements a novel analysis algorithm that is robust even in cases where an adversarial OF application attempts to strategically insert flow rules that would otherwise circumvent flow rules imposed by OF security applications. We demonstrate the utility of FortNOX through a prototype implementation and use it to examine performance and efficiency aspects of the proposed framework.",
"Today's networks typically carry or deploy dozens of protocols and mechanisms simultaneously such as MPLS, NAT, ACLs and route redistribution. Even when individual protocols function correctly, failures can arise from the complex interactions of their aggregate, requiring network administrators to be masters of detail. Our goal is to automatically find an important class of failures, regardless of the protocols running, for both operational and experimental networks. To this end we developed a general and protocol-agnostic framework, called Header Space Analysis (HSA). Our formalism allows us to statically check network specifications and configurations to identify an important class of failures such as Reachability Failures, Forwarding Loops and Traffic Isolation and Leakage problems. In HSA, protocol header fields are not first class entities; instead we look at the entire packet header as a concatenation of bits without any associated meaning. Each packet is a point in the 0,1 L space where L is the maximum length of a packet header, and networking boxes transform packets from one point in the space to another point or set of points (multicast). We created a library of tools, called Hassel, to implement our framework, and used it to analyze a variety of networks and protocols. Hassel was used to analyze the Stanford University backbone network, and found all the forwarding loops in less than 10 minutes, and verified reachability constraints between two subnets in 13 seconds. It also found a large and complex loop in an experimental loose source routing protocol in 4 minutes.",
"It is critical to ensure that network policy remains consistent during state transitions. However, existing techniques impose a high cost in update delay, and or FIB space. We propose the Customizable Consistency Generator (CCG), a fast and generic framework to support customizable consistency policies during network updates. CCG effectively reduces the task of synthesizing an update plan under the constraint of a given consistency policy to a verification problem, by checking whether an update can safely be installed in the network at a particular time, and greedily processing network state transitions to heuristically minimize transition delay. We show a large class of consistency policies are guaranteed by this greedy heuristic alone; in addition, CCG makes judicious use of existing heavier-weight network update mechanisms to provide guarantees when necessary. As such, CCG nearly achieves the \"best of both worlds\": the efficiency of simply passing through updates in most cases, with the consistency guarantees of more heavyweight techniques. Mininet and physical testbed evaluations demonstrate CCG's capability to achieve various types of consistency, such as path and bandwidth properties, with zero switch memory overhead and up to a 3× delay reduction compared to previous solutions.",
"OpenFlow is an open standard that has gained tremendous interest in the last few years within the network community. It is an embodiment of the software-defined networking paradigm, in which higher-level flow routing decisions are derived from a control layer that, unlike classic network switch implementations, is separated from the data handling layer. The central attraction to this paradigm is that by decoupling the control logic from the closed and proprietary implementations of traditional network switch infrastructure, researchers can more easily design and distribute innovative flow handling and network control algorithms. Indeed, we also believe that OpenFlow can, in time, prove to be one of the more impactful technologies to drive a variety of innovations in network security. OpenFlow could offer a dramatic simplification to the way we design and integrate complex network security applications into large networks. However, to date there remains a stark paucity of compelling OpenFlow security applications. In this paper, we introduce FRESCO, an OpenFlow security application development framework designed to facilitate the rapid design, and modular composition of OF-enabled detection and mitigation modules. FRESCO, which is itself an OpenFlow application, offers a Click-inspired [19] programming framework that enables security researchers to implement, share, and compose together, many different security detection and mitigation modules. We demonstrate the utility of FRESCO through the implementation of several well-known security defenses as OpenFlow security services, and use them to examine various performance and efficiency aspects of our proposed framework.",
"Software-defined networking (SDN) is a new paradigm for operating and managing computer networks. SDN enables logically-centralized control over network devices through a \"controller\" --- software that operates independently of the network hardware. Network operators can run both in-house and third-party SDN programs on top of the controller, e.g., to specify routing and access control policies. In practice, having the controller handle events limits the network scalability. Therefore, the feasibility of SDN depends on the ability to efficiently decentralize network event-handling by installing forwarding rules on the switches. However, installing a rule too early or too late may lead to incorrect behavior, e.g., (1) packets may be forwarded to the wrong destination or incorrectly dropped; (2) packets handled by the switch may hide vital information from the controller, leading to incorrect forwarding behavior. The second issue is subtle and sometimes missed even by experienced programmers. The contributions of this paper are two fold. First, we formalize the correctness and optimality requirements for decentralizing network policies. Second, we identify a useful class of network policies which permits automatic synthesis of a controller which performs optimal forwarding rule installation.",
"Networks are complex and prone to bugs. Existing tools that check network configuration files and the data-plane state operate offline at timescales of seconds to hours, and cannot detect or prevent bugs as they arise. Is it possible to check network-wide invariants in real time, as the network state evolves? The key challenge here is to achieve extremely low latency during the checks so that network performance is not affected. In this paper, we present a design, VeriFlow, which achieves this goal. VeriFlow is a layer between a software-defined networking controller and network devices that checks for network-wide invariant violations dynamically as each forwarding rule is inserted, modified or deleted. VeriFlow supports analysis over multiple header fields, and an API for checking custom invariants. Based on a prototype implementation integrated with the NOX OpenFlow controller, and driven by a Mininet OpenFlow network and Route Views trace data, we find that VeriFlow can perform rigorous checking within hundreds of microseconds per rule insertion or deletion.",
"Among the leading reference implementations of the Software Defined Networking (SDN) paradigm is the OpenFlow framework, which decouples the control plane into a centralized application. In this paper, we consider two aspects of OpenFlow that pose security challenges, and we propose two solutions that could address these concerns. The first challenge is the inherent communication bottleneck that arises between the data plane and the control plane, which an adversary could exploit by mounting a \"control plane saturation attack\" that disrupts network operations. Indeed, even well-mined adversarial models, such as scanning or denial-of-service (DoS) activity, can produce more potent impacts on OpenFlow networks than traditional networks. To address this challenge, we introduce an extension to the OpenFlow data plane called \"connection migration\", which dramatically reduces the amount of data-to-control-plane interactions that arise during such attacks. The second challenge is that of enabling the control plane to expedite both detection of, and responses to, the changing flow dynamics within the data plane. For this, we introduce \"actuating triggers\" over the data plane's existing statistics collection services. These triggers are inserted by control layer applications to both register for asynchronous call backs, and insert conditional flow rules that are only activated when a trigger condition is detected within the data plane's statistics module. We present Avant-Guard, an implementation of our two data plane extensions, evaluate the performance impact, and examine its use for developing more scalable and resilient SDN security services.",
"Configuration changes are a common source of instability in networks, leading to broken connectivity, forwarding loops, and access control violations. Even when the initial and final states of the network are correct, the update process often steps through intermediate states with incorrect behaviors. These problems have been recognized in the context of specific protocols, leading to a number of point solutions. However, a piecemeal attack on this fundamental problem, while pragmatic in the short term, is unlikely to lead to significant long-term progress. Software-Defined Networking (SDN) provides an exciting opportunity to do better. Because SDN is a clean-slate platform, we can build general, reusable abstractions for network updates that come with strong semantic guarantees. We believe SDN desperately needs such abstractions to make programs simpler to design, more reliable, and easier to validate using automated tools. Moreover, we believe these abstractions should be provided by a runtime system, shielding the programmer from these concerns. We propose two simple, canonical, and effective update abstractions, and present implementation mechanisms. We also show how to integrate them with a network programming language, and discuss potential applications to program verification."
]
} |
1611.05362 | 2575352857 | Software-defined networking is considered a promising new paradigm, enabling more reliable and formally verifiable communication networks. However, this paper shows that the separation of the control plane from the data plane, which lies at the heart of Software-Defined Networks (SDNs), introduces a new vulnerability which we call teleportation. An attacker (e.g., a malicious switch in the data plane or a host connected to the network) can use teleportation to transmit information via the control plane and bypass critical network functions in the data plane (e.g., a firewall), and to violate security policies as well as logical and even physical separations. This paper characterizes the design space for teleportation attacks theoretically, and then identifies four different teleportation techniques. We demonstrate and discuss how these techniques can be exploited for different attacks (e.g., exfiltrating confidential data at high rates), and also initiate the discussion of possible countermeasures. Generally, and given today's trend toward more intent-based networking, we believe that our findings are relevant beyond the use cases considered in this paper. | Only recently researchers have started discovering security threats in SDN. Kl "o @cite_7 report on a STRIDE threat analysis of OpenFlow, and demonstrate data plane resource consumption attacks. @cite_12 survey several threat vectors that may enable the exploitation of SDN vulnerabilities. @cite_21 analyze vulnerabilities in OpenFlow. In particular they point out the lack of TLS adoption implementation in OpenFlow switches and controllers. In addition, they correctly identify the possibility of dos attacks on the centralized control plane. Another key challenge arising from the separation of the control and data planes, is the potential loss of network visibility. It has been shown that the network view of the controller may even be poisoned @cite_4 @cite_40 . @cite_45 , point out that threat models for the virtualized data plane need to account for a malicious compromised data plane in SDNs, and cloud operating systems such as OpenStack. | {
"cite_N": [
"@cite_4",
"@cite_7",
"@cite_21",
"@cite_40",
"@cite_45",
"@cite_12"
],
"mid": [
"2084915558",
"2026356331",
"2064266822",
"1984451560",
"2546300025",
"2067551786"
],
"abstract": [
"Software-defined networks (SDNs) allow greater control over network entities by centralizing the control plane, but place great burden on the administrator to manually ensure security and correct functioning of the entire network. We list several attacks on SDN controllers that violate network topology and data plane forwarding, and can be mounted by compromised network entities, such as end hosts and soft switches. We further demonstrate their feasibility on four popular SDN controllers. We propose SPHINX to detect both known and potentially unknown attacks on network topology and data plane forwarding originating within an SDN. SPHINX leverages the novel abstraction of flow graphs, which closely approximate the actual network operations, to enable incremental validation of all network updates and constraints. SPHINX dynamically learns new network behavior and raises alerts when it detects suspicious changes to existing network control plane behavior. Our evaluation shows that SPHINX is capable of detecting attacks in SDNs in realtime with low performance overheads, and requires no changes to the controllers for deployment.",
"Software Defined Networking (SDN) has been proposed as a drastic shift in the networking paradigm, by decoupling network control from the data plane and making the switching infrastructure truly programmable. The key enabler of SDN, OpenFlow, has seen widespread deployment on production networks and its adoption is constantly increasing. Although openness and programmability are primary features of OpenFlow, security is of core importance for real-world deployment. In this work, we perform a security analysis of OpenFlow using STRIDE and attack tree modeling methods, and we evaluate our approach on an emulated network testbed. The evaluation assumes an attacker model with access to the network data plane. Finally, we propose appropriate counter-measures that can potentially mitigate the security issues associated with OpenFlow networks. Our analysis and evaluation approach are not exhaustive, but are intended to be adaptable and extensible to new versions and deployment contexts of OpenFlow.",
"We provide a brief overview of the vulnerabilities present in the OpenFlow protocol as it is currently deployed by hardware and software vendors. We identify a widespread failure to adopt TLS for the OpenFlow control channel by both controller and switch vendors, leaving OpenFlow vulnerable to man-in-the-middle attacks. We also highlight the classes of vulnerabilities that emerge from the separation and centralization of the control plane in OpenFlow network designs. Finally, we offer suggestions for future work to address these vulnerabilities in a systematic fashion.",
"",
"Virtual switches have become popular among cloud operating systems to interconnect virtual machines in a more flexible manner. However, this paper demonstrates that virtual switches introduce new attack surfaces in cloud setups, whose effects can be disastrous. Our analysis shows that these vulnerabilities are caused by: (1) inappropriate security assumptions (privileged virtual switch execution in kernel and user space), (2) the logical centralization of such networks (e.g., OpenStack or SDN), (3) the presence of bi-directional communication channels between data plane systems and the centralized controller, and (4) non-standard protocol parsers. Our work highlights the need to accommodate the data plane(s) in our threat models. In particular, it forces us to revisit today's assumption that the data plane can only be compromised by a sophisticated attacker: we show that compromising the data plane of modern computer networks can actually be performed by a very simple attacker with limited resources only and at low cost (i.e., at the cost of renting a virtual machine in the Cloud). As a case study, we fuzzed only 2 of the code-base of a production quality virtual switch's packet processor (namely OvS), identifying serious vulnerabilities leading to unauthenticated remote code execution. In particular, we present the \"rein worm\" which allows us to fully compromise test-setups in less than 100 seconds. We also evaluate the performance overhead of existing mitigations such as ASLR, PIEs, and unconditional stack canaries on OvS. We find that while applying these countermeasures in kernel-space incurs a significant overhead, in user-space the performance overhead is negligible.",
"Software-defined networking empowers network operators with more flexibility to program their networks. With SDN, network management moves from codifying functionality in terms of low-level device configurations to building software that facilitates network management and debugging. By separating the complexity of state distribution from network specification, SDN provides new ways to solve long-standing problems in networking --- routing, for instance --- while simultaneously allowing the use of security and dependability techniques, such as access control or multi-path. However, the security and dependability of the SDN itself is still an open issue. In this position paper we argue for the need to build secure and dependable SDNs by design. As a first step in this direction we describe several threat vectors that may enable the exploit of SDN vulnerabilities. We then sketch the design of a secure and dependable SDN control platform as a materialization of the concept here advocated. We hope that this paper will trigger discussions in the SDN community around these issues and serve as a catalyser to join efforts from the networking and security & dependability communities in the ultimate goal of building resilient control planes."
]
} |
1611.05362 | 2575352857 | Software-defined networking is considered a promising new paradigm, enabling more reliable and formally verifiable communication networks. However, this paper shows that the separation of the control plane from the data plane, which lies at the heart of Software-Defined Networks (SDNs), introduces a new vulnerability which we call teleportation. An attacker (e.g., a malicious switch in the data plane or a host connected to the network) can use teleportation to transmit information via the control plane and bypass critical network functions in the data plane (e.g., a firewall), and to violate security policies as well as logical and even physical separations. This paper characterizes the design space for teleportation attacks theoretically, and then identifies four different teleportation techniques. We demonstrate and discuss how these techniques can be exploited for different attacks (e.g., exfiltrating confidential data at high rates), and also initiate the discussion of possible countermeasures. Generally, and given today's trend toward more intent-based networking, we believe that our findings are relevant beyond the use cases considered in this paper. | While much research went into designing more robust and secure control planes @cite_27 @cite_36 , less published work exists on the issue of malicious switches. A notable exception is the work by @cite_13 , who consider the possibility of a malicious relay node for a man-in-the-middle attack. Interestingly, in our paper, we have shown that the relay node can be the benign controller itself. | {
"cite_N": [
"@cite_36",
"@cite_27",
"@cite_13"
],
"mid": [
"1697123834",
"1714884520",
"106863888"
],
"abstract": [
"The emergence of OpenFlow-capable switches enables exciting new network functionality, at the risk of programming errors that make communication less reliable. The centralized programming model, where a single controller program manages the network, seems to reduce the likelihood of bugs. However, the system is inherently distributed and asynchronous, with events happening at different switches and end hosts, and inevitable delays affecting communication with the controller. In this paper, we present efficient, systematic techniques for testing unmodified controller programs. Our NICE tool applies model checking to explore the state space of the entire system--the controller, the switches, and the hosts. Scalability is the main challenge, given the diversity of data packets, the large system state, and the many possible event orderings. To address this, we propose a novel way to augment model checking with symbolic execution of event handlers (to identify representative packets that exercise code paths on the controller). We also present a simplified OpenFlow switch model (to reduce the state space), and effective strategies for generating event interleavings likely to uncover bugs. Our prototype tests Python applications on the popular NOX platform. In testing three real applications--a MAC-learning switch, in-network server load balancing, and energy-efficient traffic engineering--we uncover eleven bugs.",
"Software-defined networking (SDN) is a novel paradigm that outsources the control of programmable network switches to a set of software controllers. The most fundamental task of these controllers is the correct implementation of the network policy, i.e., the intended network behavior. In essence, such a policy specifies the rules by which packets must be forwarded across the network. This paper studies a distributed SDN control plane that enables concurrent and robust policy implementation. We introduce a formal model describing the interaction between the data plane and a distributed control plane (consisting of a collection of fault-prone controllers). Then we formulate the problem of consistent composition of concurrent network policy updates (termed the CPC Problem). To anticipate scenarios in which some conflicting policy updates must be rejected, we enable the composition via a natural transactional interface with all-or-nothing semantics. We show that the ability of an f-resilient distributed control plane to process concurrent policy updates depends on the tag complexity, i.e., the number of policy labels (a.k.a. tags) available to the controllers, and describe a CPC protocol with optimal tag complexity f + 2.",
"Software defined networking (SDN) and OpenFlow as one of its key technologies have received a lot of attention from the networking community. While SDN enables complex network applications and easier network management, the paradigm change comes along with new security threats. In this paper, we analyze attacks against a software-defined network in a scenario where the attacker has been able to compromise one or more OpenFlow-capable switches. We find out that such attacker can in suitable environments perform a wide range of attacks, including man-in-the-middle attacks against control-plane traffic, by using only the standard OpenFlow functionality of the switch. Furthermore, we show that in certain scenarios it is nearly impossible to detect that some switch has been compromised. We conclude that while the existing security mechanisms, such as TLS, give protection against many of the presented attacks, the threats should not be overlooked when moving to SDN and OpenFlow."
]
} |
1611.05362 | 2575352857 | Software-defined networking is considered a promising new paradigm, enabling more reliable and formally verifiable communication networks. However, this paper shows that the separation of the control plane from the data plane, which lies at the heart of Software-Defined Networks (SDNs), introduces a new vulnerability which we call teleportation. An attacker (e.g., a malicious switch in the data plane or a host connected to the network) can use teleportation to transmit information via the control plane and bypass critical network functions in the data plane (e.g., a firewall), and to violate security policies as well as logical and even physical separations. This paper characterizes the design space for teleportation attacks theoretically, and then identifies four different teleportation techniques. We demonstrate and discuss how these techniques can be exploited for different attacks (e.g., exfiltrating confidential data at high rates), and also initiate the discussion of possible countermeasures. Generally, and given today's trend toward more intent-based networking, we believe that our findings are relevant beyond the use cases considered in this paper. | Sphinx @cite_4 can alert on the path update teleportation. However, it cannot detect the path reset as the flow graph remains the same. Additionally, Sphinx assumes that switches cannot use the same DPIDs, therefore, we believe that our switch identification teleportation will not be detected by Sphinx. Also, our out-of-band forwarding relies on and messages, while s are not considered by Sphinx Unfortunately, the source code of Sphinx is not available. . Therefore the suggested out-of-band forwarding teleportation can evade Sphinx, until topology altering flows are installed. | {
"cite_N": [
"@cite_4"
],
"mid": [
"2084915558"
],
"abstract": [
"Software-defined networks (SDNs) allow greater control over network entities by centralizing the control plane, but place great burden on the administrator to manually ensure security and correct functioning of the entire network. We list several attacks on SDN controllers that violate network topology and data plane forwarding, and can be mounted by compromised network entities, such as end hosts and soft switches. We further demonstrate their feasibility on four popular SDN controllers. We propose SPHINX to detect both known and potentially unknown attacks on network topology and data plane forwarding originating within an SDN. SPHINX leverages the novel abstraction of flow graphs, which closely approximate the actual network operations, to enable incremental validation of all network updates and constraints. SPHINX dynamically learns new network behavior and raises alerts when it detects suspicious changes to existing network control plane behavior. Our evaluation shows that SPHINX is capable of detecting attacks in SDNs in realtime with low performance overheads, and requires no changes to the controllers for deployment."
]
} |
1611.05362 | 2575352857 | Software-defined networking is considered a promising new paradigm, enabling more reliable and formally verifiable communication networks. However, this paper shows that the separation of the control plane from the data plane, which lies at the heart of Software-Defined Networks (SDNs), introduces a new vulnerability which we call teleportation. An attacker (e.g., a malicious switch in the data plane or a host connected to the network) can use teleportation to transmit information via the control plane and bypass critical network functions in the data plane (e.g., a firewall), and to violate security policies as well as logical and even physical separations. This paper characterizes the design space for teleportation attacks theoretically, and then identifies four different teleportation techniques. We demonstrate and discuss how these techniques can be exploited for different attacks (e.g., exfiltrating confidential data at high rates), and also initiate the discussion of possible countermeasures. Generally, and given today's trend toward more intent-based networking, we believe that our findings are relevant beyond the use cases considered in this paper. | SDN Hypervisors such as CoVisor @cite_33 , Flowvisor @cite_29 , FortNOX @cite_38 depend on policies maintained in the hypervisor. Therefore, we believe that all our teleportation mechanisms hold unless a specific policy blocks it. | {
"cite_N": [
"@cite_38",
"@cite_29",
"@cite_33"
],
"mid": [
"2137845741",
"2186961980",
"1792894036"
],
"abstract": [
"Software-defined networks facilitate rapid and open innovation at the network control layer by providing a programmable network infrastructure for computing flow policies on demand. However, the dynamism of programmable networks also introduces new security challenges that demand innovative solutions. A critical challenge is efficient detection and reconciliation of potentially conflicting flow rules imposed by dynamic OpenFlow (OF) applications. To that end, we introduce FortNOX, a software extension that provides role-based authorization and security constraint enforcement for the NOX OpenFlow controller. FortNOX enables NOX to check flow rule contradictions in real time, and implements a novel analysis algorithm that is robust even in cases where an adversarial OF application attempts to strategically insert flow rules that would otherwise circumvent flow rules imposed by OF security applications. We demonstrate the utility of FortNOX through a prototype implementation and use it to examine performance and efficiency aspects of the proposed framework.",
"Network virtualization has long been a goal of of the network research community. With it, multiple isolated logical networks each with potentially different addressing and forwarding mechanisms can share the same physical infrastructure. Typically this is achieved by taking advantage of the flexibility of software (e.g. [20, 23]) or by duplicating components in (often specialized) hardware[19]. In this paper we present a new approach to switch virtualization in which the same hardware forwarding plane can be shared among multiple logical networks, each with distinct forwarding logic. We use this switch-level virtualization to build a research platform which allows multiple network experiments to run side-by-side with production traffic while still providing isolation and hardware forwarding speeds. We also show that this approach is compatible with commodity switching chipsets and does not require the use of programmable hardware such as FPGAs or network processors. We build and deploy this virtualization platform on our own production network and demonstrate its use in practice by running five experiments simultaneously within a campus network. Further, we quantify the overhead of our approach and evaluate the completeness of the isolation between virtual slices.",
"We present CoVisor, a new kind of network hypervisor that enables, in a single network, the deployment of multiple control applications written in different programming languages and operating on different controller platforms. Unlike past hypervisors, which focused on slicing the network into disjoint parts for separate control by separate entities, CoVisor allows multiple controllers to cooperate on managing the same shared traffic. Consequently, network administrators can use CoVisor to assemble a collection of independently-developed \"best of breed\" applications--a firewall, a load balancer, a gateway, a router, a traffic monitor--and can apply those applications in combination, or separately, to the desired traffic. CoVisor also abstracts concrete topologies, providing custom virtual topologies in their place, and allows administrators to specify access controls that regulate the packets a given controller may see, modify, monitor, or reroute. The central technical contribution of the work is a new set of efficient algorithms for composing controller policies, for compiling virtual networks into concrete OpenFlow rules, and for efficiently processing controller rule updates. We have built a CoVisor prototype, and shown that it is several orders of magnitude faster than a naive implementation."
]
} |
1611.05592 | 2950553082 | Video captioning which automatically translates video clips into natural language sentences is a very important task in computer vision. By virtue of recent deep learning technologies, e.g., convolutional neural networks (CNNs) and recurrent neural networks (RNNs), video captioning has made great progress. However, learning an effective mapping from visual sequence space to language space is still a challenging problem. In this paper, we propose a Multimodal Memory Model (M3) to describe videos, which builds a visual and textual shared memory to model the long-term visual-textual dependency and further guide global visual attention on described targets. Specifically, the proposed M3 attaches an external memory to store and retrieve both visual and textual contents by interacting with video and sentence with multiple read and write operations. First, text representation in the Long Short-Term Memory (LSTM) based text decoder is written into the memory, and the memory contents will be read out to guide an attention to select related visual targets. Then, the selected visual information is written into the memory, which will be further read out to the text decoder. To evaluate the proposed model, we perform experiments on two publicly benchmark datasets: MSVD and MSR-VTT. The experimental results demonstrate that our method outperforms the state-of-theart methods in terms of BLEU and METEOR. | Memory Modelling 3mm To extend the memory ability of traditional neural networks, @cite_37 propose a Neural Turing Machine (NTM) which holds an external memory to interact with the internal state of neural networks by attention mechanism. NTM has shown the potential of storage and access of information over long time periods which has always been problematic for RNNs, e.g., copying, sorting and associative recall. Besides memory matrix in NTM, memory is also modelled as continuous and differentiable doubly-linked lists and stacks @cite_36 , queues and deques @cite_18 . Different from exploring various forms of dynamic storages, @cite_42 model large long-term static memory. The internal information stored in the static memory is not modified by external controllers, which is specially used for reading comprehension. These memory networks have been successfully applied to the tasks which need long-term dependency modelling, e.g., textual question answering @cite_15 @cite_29 , visual question answering @cite_32 and dialog systems @cite_5 . As we know, few memory models have been proposed for video captioning. In this paper, we will propose an external multimodal memory to interact with video and sentence simultaneously. | {
"cite_N": [
"@cite_37",
"@cite_18",
"@cite_36",
"@cite_29",
"@cite_42",
"@cite_32",
"@cite_5",
"@cite_15"
],
"mid": [
"2167839676",
"1602017060",
"2949626814",
"",
"",
"2293453011",
"2175256910",
"580074167"
],
"abstract": [
"We extend the capabilities of neural networks by coupling them to external memory resources, which they can interact with by attentional processes. The combined system is analogous to a Turing Machine or Von Neumann architecture but is differentiable end-toend, allowing it to be efficiently trained with gradient descent. Preliminary results demonstrate that Neural Turing Machines can infer simple algorithms such as copying, sorting, and associative recall from input and output examples.",
"Recently, strong results have been demonstrated by Deep Recurrent Neural Networks on natural language transduction problems. In this paper we explore the representational power of these models using synthetic grammars designed to exhibit phenomena similar to those found in real transduction problems such as machine translation. These experiments lead us to propose new memory-based recurrent networks that implement continuously differentiable analogues of traditional data structures such as Stacks, Queues, and DeQues. We show that these architectures exhibit superior generalisation performance to Deep RNNs and are often able to learn the underlying generating algorithms in our transduction experiments.",
"Despite the recent achievements in machine learning, we are still very far from achieving real artificial intelligence. In this paper, we discuss the limitations of standard deep learning approaches and show that some of these limitations can be overcome by learning how to grow the complexity of a model in a structured way. Specifically, we study the simplest sequence prediction problems that are beyond the scope of what is learnable with standard recurrent networks, algorithmically generated sequences which can only be learned by models which have the capacity to count and to memorize sequences. We show that some basic algorithms can be learned from sequential data using a recurrent network associated with a trainable memory.",
"",
"",
"Neural network architectures with memory and attention mechanisms exhibit certain reasoning capabilities required for question answering. One such architecture, the dynamic memory network (DMN), obtained high accuracy on a variety of language tasks. However, it was not shown whether the architecture achieves strong results for question answering when supporting facts are not marked during training or whether it could be applied to other modalities such as images. Based on an analysis of the DMN, we propose several improvements to its memory and input modules. Together with these changes we introduce a novel input module for images in order to be able to answer visual questions. Our new DMN+ model improves the state of the art on both the Visual Question Answering dataset and the -10k text question-answering dataset without supporting fact supervision.",
"A long-term goal of machine learning is to build intelligent conversational agents. One recent popular approach is to train end-to-end models on a large amount of real dialog transcripts between humans (, 2015; Vinyals & Le, 2015; , 2015). However, this approach leaves many questions unanswered as an understanding of the precise successes and shortcomings of each model is hard to assess. A contrasting recent proposal are the bAbI tasks (, 2015b) which are synthetic data that measure the ability of learning machines at various reasoning tasks over toy language. Unfortunately, those tests are very small and hence may encourage methods that do not scale. In this work, we propose a suite of new tasks of a much larger scale that attempt to bridge the gap between the two regimes. Choosing the domain of movies, we provide tasks that test the ability of models to answer factual questions (utilizing OMDB), provide personalization (utilizing MovieLens), carry short conversations about the two, and finally to perform on natural dialogs from Reddit. We provide a dataset covering 75k movie entities and with 3.5M training examples. We present results of various models on these tasks, and evaluate their performance.",
"Training large-scale question answering systems is complicated because training sources usually cover a small portion of the range of possible questions. This paper studies the impact of multitask and transfer learning for simple question answering; a setting for which the reasoning required to answer is quite easy, as long as one can retrieve the correct evidence given a question, which can be difficult in large-scale conditions. To this end, we introduce a new dataset of 100k questions that we use in conjunction with existing benchmarks. We conduct our study within the framework of Memory Networks (, 2015) because this perspective allows us to eventually scale up to more complex reasoning, and show that Memory Networks can be successfully trained to achieve excellent performance."
]
} |
1611.05365 | 2953135340 | This paper presents a method to assess a basketball player's performance from his her first-person video. A key challenge lies in the fact that the evaluation metric is highly subjective and specific to a particular evaluator. We leverage the first-person camera to address this challenge. The spatiotemporal visual semantics provided by a first-person view allows us to reason about the camera wearer's actions while he she is participating in an unscripted basketball game. Our method takes a player's first-person video and provides a player's performance measure that is specific to an evaluator's preference. To achieve this goal, we first use a convolutional LSTM network to detect atomic basketball events from first-person videos. Our network's ability to zoom-in to the salient regions addresses the issue of a severe camera wearer's head movement in first-person videos. The detected atomic events are then passed through the Gaussian mixtures to construct a highly non-linear visual spatiotemporal basketball assessment feature. Finally, we use this feature to learn a basketball assessment model from pairs of labeled first-person basketball videos, for which a basketball expert indicates, which of the two players is better. We demonstrate that despite not knowing the basketball evaluator's criterion, our model learns to accurately assess the players in real-world games. Furthermore, our model can also discover basketball events that contribute positively and negatively to a player's performance. | These kinematic data are often obtained by multiple third-person videos @cite_16 @cite_31 @cite_11 where the players and ball are detected using recognition algorithms combined with multiple view geometry @cite_37 . Tracking and data association is a key issue where the role of the players provides a strong cue to disambiguate appearance based tracking @cite_23 . Events such as ball movement, can be also recognized using a spatiotemporal analysis @cite_19 . As players behave strategically and collectively, their group movement can be predicted @cite_21 and the ball can be localized without detection. Various computational models have been used for such tasks, e.g., Dynamic Bayesian Network @cite_4 , hierarchical LSTM @cite_5 , attention based LSTM @cite_25 learned from a large collection of third-person videos. | {
"cite_N": [
"@cite_37",
"@cite_4",
"@cite_21",
"@cite_19",
"@cite_23",
"@cite_5",
"@cite_31",
"@cite_16",
"@cite_25",
"@cite_11"
],
"mid": [
"2033819227",
"1991412158",
"2056387739",
"2953334368",
"2170245723",
"",
"",
"",
"2206427987",
""
],
"abstract": [
"From the Publisher: A basic problem in computer vision is to understand the structure of a real world scene given several images of it. Recent major developments in the theory and practice of scene reconstruction are described in detail in a unified framework. The book covers the geometric principles and how to represent objects algebraically so they can be computed and applied. The authors provide comprehensive background material and explain how to apply the methods and implement the algorithms directly.",
"Modeling interactions of multiple co-occurring objects in a complex activity is becoming increasingly popular in the video domain. The Dynamic Bayesian Network (DBN) has been applied to this problem in the past due to its natural ability to statistically capture complex temporal dependencies. However, standard DBN structure learning algorithms are generatively learned, require manual structure definitions, and or are computationally complex or restrictive. We propose a novel structure learning solution that fuses the Granger Causality statistic, a direct measure of temporal dependence, with the Adaboost feature selection algorithm to automatically constrain the temporal links of a DBN in a discriminative manner. This approach enables us to completely define the DBN structure prior to parameter learning, which reduces computational complexity in addition to providing a more descriptive structure. We refer to this modeling approach as the Granger Constraints DBN (GCDBN). Our experiments show how the GCDBN outperforms two of the most relevant state-of-the-art graphical models in complex activity classification on handball video data, surveillance data, and synthetic data.",
"We present a method to detect the regions of interests in moving camera views of dynamic scenes with multiple moving objects. We start by extracting a global motion tendency that reflects the scene context by tracking movements of objects in the scene. We then use Gaussian process regression to represent the extracted motion tendency as a stochastic vector field. The generated stochastic field is robust to noise and can handle a video from an uncalibrated moving camera. We use the stochastic field for predicting important future regions of interest as the scene evolves dynamically. We evaluate our approach on a variety of videos of team sports and compare the detected regions of interest to the camera motion generated by actual camera operators. Our experimental results demonstrate that our approach is computationally efficient and provides better predictions than previously proposed RBF-based approaches.",
"Tracking the ball is critical for video-based analysis of team sports. However, it is difficult, especially in low-resolution images, due to the small size of the ball, its speed that creates motion blur, and its often being occluded by players. In this paper, we propose a generic and principled approach to modeling the interaction between the ball and the players while also imposing appropriate physical constraints on the ball's trajectory. We show that our approach, formulated in terms of a Mixed Integer Program, is more robust and more accurate than several state-of-the-art approaches on real-life volleyball, basketball, and soccer sequences.",
"In this paper, we describe a method to represent and discover adversarial group behavior in a continuous domain. In comparison to other types of behavior, adversarial behavior is heavily structured as the location of a player (or agent) is dependent both on their teammates and adversaries, in addition to the tactics or strategies of the team. We present a method which can exploit this relationship through the use of a spatiotemporal basis model. As players constantly change roles during a match, we show that employing a \"role-based\" representation instead of one based on player \"identity\" can best exploit the playing structure. As vision-based systems currently do not provide perfect detection tracking (e.g. missed or false detections), we show that our compact representation can effectively \"denoise\" erroneous detections as well as enabling temporal analysis, which was previously prohibitive due to the dimensionality of the signal. To evaluate our approach, we used a fully instrumented field-hockey pitch with 8 fixed high-definition (HD) cameras and evaluated our approach on approximately 200,000 frames of data from a state-of-the-art real-time player detector and compare it to manually labelled data.",
"",
"",
"",
"Multi-person event recognition is a challenging task, often with many people active in the scene but only a small subset contributing to an actual event. In this paper, we propose a model which learns to detect events in such videos while automatically \"attending\" to the people responsible for the event. Our model does not use explicit annotations regarding who or where those people are during training and testing. In particular, we track people in videos and use a recurrent neural network (RNN) to represent the track features. We learn time-varying attention weights to combine these features at each time-instant. The attended features are then processed using another RNN for event detection classification. Since most video datasets with multiple people are restricted to a small number of videos, we also collected a new basketball dataset comprising 257 basketball games with 14K event annotations corresponding to 11 event classes. Our model outperforms state-of-the-art methods for both event classification and detection on this new dataset. Additionally, we show that the attention mechanism is able to consistently localize the relevant players.",
""
]
} |
1611.05365 | 2953135340 | This paper presents a method to assess a basketball player's performance from his her first-person video. A key challenge lies in the fact that the evaluation metric is highly subjective and specific to a particular evaluator. We leverage the first-person camera to address this challenge. The spatiotemporal visual semantics provided by a first-person view allows us to reason about the camera wearer's actions while he she is participating in an unscripted basketball game. Our method takes a player's first-person video and provides a player's performance measure that is specific to an evaluator's preference. To achieve this goal, we first use a convolutional LSTM network to detect atomic basketball events from first-person videos. Our network's ability to zoom-in to the salient regions addresses the issue of a severe camera wearer's head movement in first-person videos. The detected atomic events are then passed through the Gaussian mixtures to construct a highly non-linear visual spatiotemporal basketball assessment feature. Finally, we use this feature to learn a basketball assessment model from pairs of labeled first-person basketball videos, for which a basketball expert indicates, which of the two players is better. We demonstrate that despite not knowing the basketball evaluator's criterion, our model learns to accurately assess the players in real-world games. Furthermore, our model can also discover basketball events that contribute positively and negatively to a player's performance. | Unlike third-person videos, first-person cameras closely capture what the players see. Such property is beneficial to understand activities highly correlated with visual attention, e.g., object manipulation and social communications. Important objects to the camera wearer are detected and segmented @cite_38 @cite_14 @cite_35 @cite_29 @cite_32 , which can be used to compress life-log videos @cite_38 @cite_39 . As visual attention is also related with the intent of the camera wearer, her his future movement can be predicted @cite_36 . Beyond individual behaviors, joint attention is a primary indicator of social interactions, which can be directly computed from first-person videos @cite_34 @cite_33 , and further used for human-robot interactions @cite_40 @cite_30 . | {
"cite_N": [
"@cite_30",
"@cite_38",
"@cite_35",
"@cite_14",
"@cite_33",
"@cite_36",
"@cite_29",
"@cite_32",
"@cite_39",
"@cite_40",
"@cite_34"
],
"mid": [
"602397586",
"2071711566",
"",
"",
"",
"1947050545",
"2031688197",
"2296893412",
"2120645068",
"",
""
],
"abstract": [
"The objective of this work is human pose estimation in videos, where multiple frames are available. We investigate a ConvNet architecture that is able to benefit from temporal context by combining information across the multiple frames using optical flow. To this end we propose a network architecture with the following novelties: (i) a deeper network than previously investigated for regressing heatmaps, (ii) spatial fusion layers that learn an implicit spatial model, (iii) optical flow is used to align heatmap predictions from neighbouring frames, and (iv) a final parametric pooling layer which learns to combine the aligned heatmaps into a pooled confidence map. We show that this architecture outperforms a number of others, including one that uses optical flow solely at the input layers, one that regresses joint coordinates directly, and one that predicts heatmaps without spatial fusion. The new architecture outperforms the state of the art by a large margin on three video pose estimation datasets, including the very challenging Poses in the Wild dataset, and outperforms other deep methods that don't use a graphical model on the single-image FLIC benchmark (and also [5, 35] in the high precision region).",
"We present a video summarization approach for egocentric or \"wearable\" camera data. Given hours of video, the proposed method produces a compact storyboard summary of the camera wearer's day. In contrast to traditional keyframe selection techniques, the resulting summary focuses on the most important objects and people with which the camera wearer interacts. To accomplish this, we develop region cues indicative of high-level saliency in egocentric video--such as the nearness to hands, gaze, and frequency of occurrence--and learn a regressor to predict the relative importance of any new region based on these cues. Using these predictions and a simple form of temporal event detection, our method selects frames for the storyboard that reflect the key object-driven happenings. We adjust the compactness of the final summary given either an importance selection criterion or a length budget; for the latter, we design an efficient dynamic programming solution that accounts for importance, visual uniqueness, and temporal displacement. Critically, the approach is neither camera-wearer-specific nor object-specific; that means the learned importance metric need not be trained for a given user or context, and it can predict the importance of objects and people that have never been seen previously. Our results on two egocentric video datasets show the method's promise relative to existing techniques for saliency and summarization.",
"",
"",
"",
"We address the challenging problem of recognizing the camera wearer's actions from videos captured by an egocentric camera. Egocentric videos encode a rich set of signals regarding the camera wearer, including head movement, hand pose and gaze information. We propose to utilize these mid-level egocentric cues for egocentric action recognition. We present a novel set of egocentric features and show how they can be combined with motion and object features. The result is a compact representation with superior performance. In addition, we provide the first systematic evaluation of motion, object and egocentric cues in egocentric action recognition. Our benchmark leads to several surprising findings. These findings uncover the best practices for egocentric actions, with a significant performance boost over all previous state-of-the-art methods on three publicly available datasets.",
"This paper addresses the problem of learning object models from egocentric video of household activities, using extremely weak supervision. For each activity sequence, we know only the names of the objects which are present within it, and have no other knowledge regarding the appearance or location of objects. The key to our approach is a robust, unsupervised bottom up segmentation method, which exploits the structure of the egocentric domain to partition each frame into hand, object, and background categories. By using Multiple Instance Learning to match object instances across sequences, we discover and localize object occurrences. Object representations are refined through transduction and object-level classifiers are trained. We demonstrate encouraging results in detecting novel object instances using models produced by weakly-supervised learning.",
"Unlike traditional third-person cameras mounted on robots, a first-person camera, captures a person's visual sensorimotor object interactions from up close. In this paper, we study the tight interplay between our momentary visual attention and motor action with objects from a first-person camera. We propose a concept of action-objects---the objects that capture person's conscious visual (watching a TV) or tactile (taking a cup) interactions. Action-objects may be task-dependent but since many tasks share common person-object spatial configurations, action-objects exhibit a characteristic 3D spatial distance and orientation with respect to the person. We design a predictive model that detects action-objects using EgoNet, a joint two-stream network that holistically integrates visual appearance (RGB) and 3D spatial layout (depth and height) cues to predict per-pixel likelihood of action-objects. Our network also incorporates a first-person coordinate embedding, which is designed to learn a spatial distribution of the action-objects in the first-person data. We demonstrate EgoNet's predictive power, by showing that it consistently outperforms previous baseline approaches. Furthermore, EgoNet also exhibits a strong generalization ability, i.e., it predicts semantically meaningful objects in novel first-person datasets. Our method's ability to effectively detect action-objects could be used to improve robots' understanding of human-object interactions.",
"We present a video summarization approach that discovers the story of an egocentric video. Given a long input video, our method selects a short chain of video sub shots depicting the essential events. Inspired by work in text analysis that links news articles over time, we define a random-walk based metric of influence between sub shots that reflects how visual objects contribute to the progression of events. Using this influence metric, we define an objective for the optimal k-subs hot summary. Whereas traditional methods optimize a summary's diversity or representative ness, ours explicitly accounts for how one sub-event \"leads to\" another-which, critically, captures event connectivity beyond simple object co-occurrence. As a result, our summaries provide a better sense of story. We apply our approach to over 12 hours of daily activity video taken from 23 unique camera wearers, and systematically evaluate its quality compared to multiple baselines with 34 human subjects.",
"",
""
]
} |
1611.05537 | 2951449363 | We study the tandem duplication distance between binary sequences and their roots. In other words, the quantity of interest is the number of tandem duplication operations of the form @math , where @math and @math are sequences and @math , @math , and @math are their substrings, needed to generate a binary sequence of length @math starting from a square-free sequence from the set @math . This problem is a restricted case of finding the duplication deduplication distance between two sequences, defined as the minimum number of duplication and deduplication operations required to transform one sequence to the other. We consider both exact and approximate tandem duplications. For exact duplication, denoting the maximum distance to the root of a sequence of length @math by @math , we prove that @math . For the case of approximate duplication, where a @math -fraction of symbols may be duplicated incorrectly, we show that the maximum distance has a sharp transition from linear in @math to logarithmic at @math . We also study the duplication distance to the root for sequences with a given root and for special classes of sequences, namely, the de Bruijn sequences, the Thue-Morse sequence, and the Fibbonaci words. The problem is motivated by genomic tandem duplication mutations and the smallest number of tandem duplication events required to generate a given biological sequence. | Another aspect, the study of the ability of duplication mutations to generate diversity, has been recently investigated from an information-theoretic point of view @cite_0 @cite_6 . In particular, @cite_0 models sequences generated from a starting seed'' through different types of duplications as sequence systems and studies their and . The notion of capacity quantifies the ability of the systems to generate diverse families of sequences, and expressiveness is concerned with determining whether every sequence can be generated as a substring of another sequence, if not independently. The results in @cite_0 @cite_6 include lower bounds on the capacity of tandem duplications and establishing that certain systems have nonzero capacity. The aforementioned works focus on the possibility of generating sequences and do not consider the number of duplication steps it takes to do so for any given sequence, which is the subject of the current paper. | {
"cite_N": [
"@cite_0",
"@cite_6"
],
"mid": [
"2065247540",
"2155603715"
],
"abstract": [
"It is known that the majority of the human genome consists of duplicated sequences. Furthermore, it is believed that a significant part of the rest of the genome also originated from duplicated sequences and has mutated to its current form. In this paper, we investigate the possibility of constructing an exponentially large number of sequences from a short initial sequence using simple duplication rules, including those resembling genomic-duplication processes. In other words, our goal is to find the capacity, or the expressive power, of these string-duplication systems. Our results include exact capacities, and bounds on the capacities, of four fundamental string-duplication systems. The study of these fundamental biologically inspired systems is an important step toward modeling and analyzing more complex biological processes.",
"The majority of the human genome consists of repeated sequences. An important type of repeats common in the human genome are tandem repeats, where identical copies appear next to each other. For example, in the sequence AGTCTGTGC, TGTG is a tandem repeat, namely, generated from AGTCTGC by a tandem duplication of length 2. In this work, we investigate the possibility of generating a large number of sequences from a small initial string (called the seed) by tandem duplications of bounded length. Our results include exact capacity values for certain tandem duplication string systems with alphabet sizes 2; 3; and 4. In addition, motivated by the role of DNA sequences in expressing proteins via RNA and the genetic code, we define the notion of the expressiveness of a tandem duplication system, as the feasibility of expressing arbitrary substrings. We then completely characterize the expressiveness of tandem duplication systems for general alphabet sizes and duplication lengths. Noticing that a system with capacity = 1 is expressive, we prove that for an alphabet size ≥ 4, the capacity is strictly smaller than 1, independent of the seed and the duplication lengths. The proof of this limit on the capacity (note that the genomic alphabet size is 4), is related to an interesting result by Axel Thue from 1906 which states that there exist arbitrary length sequences with no tandem repeats (square-free) for alphabet size ≥ 3. Finally, our results illustrate that duplication lengths play a more significant role than the seed in generating a large number of sequences for these systems."
]
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.