aid stringlengths 9 15 | mid stringlengths 7 10 | abstract stringlengths 78 2.56k | related_work stringlengths 92 1.77k | ref_abstract dict |
|---|---|---|---|---|
1511.06062 | 2261271299 | 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. | Several other clustering methods have been considered for visual recognition. Leung and Malik used vector quantization in the Bag of Visual Words (BoVW) framework @cite_11 initially used for texture classification, but later adopted for other visual tasks. VLAD @cite_10 and Improved Fisher Vector @cite_16 encoding improved over hard vector quantization by including second order information in the descriptors. Fisher vector has been recently been used to achieved start-of-art performances on many data-sets @cite_29 . | {
"cite_N": [
"@cite_29",
"@cite_16",
"@cite_10",
"@cite_11"
],
"mid": [
"2950004209",
"1606858007",
"2012592962",
"1484228140"
],
"abstract": [
"Visual textures have played a key role in image understanding because they convey important semantics of images, and because texture representations that pool local image descriptors in an orderless manner have had a tremendous impact in diverse applications. In this paper we make several contributions to texture understanding. First, instead of focusing on texture instance and material category recognition, we propose a human-interpretable vocabulary of texture attributes to describe common texture patterns, complemented by a new describable texture dataset for benchmarking. Second, we look at the problem of recognizing materials and texture attributes in realistic imaging conditions, including when textures appear in clutter, developing corresponding benchmarks on top of the recently proposed OpenSurfaces dataset. Third, we revisit classic texture representations, including bag-of-visual-words and the Fisher vectors, in the context of deep learning and show that these have excellent efficiency and generalization properties if the convolutional layers of a deep model are used as filter banks. We obtain in this manner state-of-the-art performance in numerous datasets well beyond textures, an efficient method to apply deep features to image regions, as well as benefit in transferring features from one domain to another.",
"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.",
"We address the problem of image search on a very large scale, where three constraints have to be considered jointly: the accuracy of the search, its efficiency, and the memory usage of the representation. We first propose a simple yet efficient way of aggregating local image descriptors into a vector of limited dimension, which can be viewed as a simplification of the Fisher kernel representation. We then show how to jointly optimize the dimension reduction and the indexing algorithm, so that it best preserves the quality of vector comparison. The evaluation shows that our approach significantly outperforms the state of the art: the search accuracy is comparable to the bag-of-features approach for an image representation that fits in 20 bytes. Searching a 10 million image dataset takes about 50ms.",
"We study the recognition of surfaces made from different materials such as concrete, rug, marble, or leather on the basis of their textural appearance. Such natural textures arise from spatial variation of two surface attributes: (1) reflectance and (2) surface normal. In this paper, we provide a unified model to address both these aspects of natural texture. The main idea is to construct a vocabulary of prototype tiny surface patches with associated local geometric and photometric properties. We call these 3D textons. Examples might be ridges, grooves, spots or stripes or combinations thereof. Associated with each texton is an appearance vector, which characterizes the local irradiance distribution, represented as a set of linear Gaussian derivative filter outputs, under different lighting and viewing conditions. Given a large collection of images of different materials, a clustering approach is used to acquire a small (on the order of 100) 3D texton vocabulary. Given a few (1 to 4) images of any material, it can be characterized using these textons. We demonstrate the application of this representation for recognition of the material viewed under novel lighting and viewing conditions. We also illustrate how the 3D texton model can be used to predict the appearance of materials under novel conditions."
]
} |
1511.06062 | 2261271299 | 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. | Reducing the number of parameters in CNN is important for training large networks and for deployment (e.g. on embedded systems). Deep Fried Convnets @cite_35 aims to reduce the number of parameters in the fully connected layer, which usually accounts for 90 | {
"cite_N": [
"@cite_35"
],
"mid": [
"2949825606"
],
"abstract": [
"The fully connected layers of a deep convolutional neural network typically contain over 90 of the network parameters, and consume the majority of the memory required to store the network parameters. Reducing the number of parameters while preserving essentially the same predictive performance is critically important for operating deep neural networks in memory constrained environments such as GPUs or embedded devices. In this paper we show how kernel methods, in particular a single Fastfood layer, can be used to replace all fully connected layers in a deep convolutional neural network. This novel Fastfood layer is also end-to-end trainable in conjunction with convolutional layers, allowing us to combine them into a new architecture, named deep fried convolutional networks, which substantially reduces the memory footprint of convolutional networks trained on MNIST and ImageNet with no drop in predictive performance."
]
} |
1511.06201 | 2262882992 | Binary representation is desirable for its memory efficiency, computation speed and robustness. In this paper, we propose adjustable bounded rectifiers to learn binary representations for deep neural networks. While hard constraining representations across layers to be binary makes training unreasonably difficult, we softly encourage activations to diverge from real values to binary by approximating step functions. Our final representation is completely binary. We test our approach on MNIST, CIFAR10, and ILSVRC2012 dataset, and systematically study the training dynamics of the binarization process. Our approach can binarize the last layer representation without loss of performance and binarize all the layers with reasonably small degradations. The memory space that it saves may allow more sophisticated models to be deployed, thus compensating the loss. To the best of our knowledge, this is the first work to report results on current deep network architectures using complete binary middle representations. Given the learned representations, we find that the firing or inhibition of a binary neuron is usually associated with a meaningful interpretation across different classes. This suggests that the semantic structure of a neural network may be manifested through a guided binarization process. | Previous work does make some positive observations by directly binarizing the features after the regular training is done. As the ReLU activation function has been widely used, the feature patterns can be understood as either activated or inhibited. In the paper @cite_6 , they binarize the features as positives and zeros, and find that the performance drop is negligible. The experiment reveals that the local minimum of a binary representation for an individual layer exists. But it never discusses about how the binary representation could be further optimized and how binary representations for multiple layers coexist. Later on, @cite_4 @cite_7 merely optimize the representation of the last layer using a static sigmoid function, which is never truly binary. Our work enforces the representations to be truly binary and it can be applied throughout the whole network. | {
"cite_N": [
"@cite_4",
"@cite_7",
"@cite_6"
],
"mid": [
"1913628733",
"",
"2160921898"
],
"abstract": [
"Approximate nearest neighbor search is an efficient strategy for large-scale image retrieval. Encouraged by the recent advances in convolutional neural networks (CNNs), we propose an effective deep learning framework to generate binary hash codes for fast image retrieval. Our idea is that when the data labels are available, binary codes can be learned by employing a hidden layer for representing the latent concepts that dominate the class labels. The utilization of the CNN also allows for learning image representations. Unlike other supervised methods that require pair-wised inputs for binary code learning, our method learns hash codes and image representations in a point-wised manner, making it suitable for large-scale datasets. Experimental results show that our method outperforms several state-of-the-art hashing algorithms on the CIFAR-10 and MNIST datasets. We further demonstrate its scalability and efficacy on a large-scale dataset of 1 million clothing images.",
"",
"In the last two years, convolutional neural networks (CNNs) have achieved an impressive suite of results on standard recognition datasets and tasks. CNN-based features seem poised to quickly replace engineered representations, such as SIFT and HOG. However, compared to SIFT and HOG, we understand much less about the nature of the features learned by large CNNs. In this paper, we experimentally probe several aspects of CNN feature learning in an attempt to help practitioners gain useful, evidence-backed intuitions about how to apply CNNs to computer vision problems."
]
} |
1511.05497 | 2507936800 | 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. | There have been many works which look at performing compression of a neural network. Weight-pruning techniques were popularized by @cite_19 and @cite_5 . Recently, @cite_6 proposed a neuron pruning technique, which relied on neuronal similarity. Our work, on the other hand, performs neuron pruning based on learning, rather than hand-crafted rules. Our learning objective can thus be seen as performing pruning and learning together, unlike the work of Han @cite_7 , who perform both operations alternately. | {
"cite_N": [
"@cite_19",
"@cite_5",
"@cite_7",
"@cite_6"
],
"mid": [
"2114766824",
"2125389748",
"1845051632",
"992687842"
],
"abstract": [
"We have used information-theoretic ideas to derive a class of practical and nearly optimal schemes for adapting the size of a neural network. By removing unimportant weights from a network, several improvements can be expected: better generalization, fewer training examples required, and improved speed of learning and or classification. The basic idea is to use second-derivative information to make a tradeoff between network complexity and training set error. Experiments confirm the usefulness of the methods on a real-world application.",
"We investigate the use of information from all second order derivatives of the error function to perform network pruning (i.e., removing unimportant weights from a trained network) in order to improve generalization, simplify networks, reduce hardware or storage requirements, increase the speed of further training, and in some cases enable rule extraction. Our method, Optimal Brain Surgeon (OBS), is Significantly better than magnitude-based methods and Optimal Brain Damage [Le Cun, Denker and Solla, 1990], which often remove the wrong weights. OBS permits the pruning of more weights than other methods (for the same error on the training set), and thus yields better generalization on test data. Crucial to OBS is a recursion relation for calculating the inverse Hessian matrix H-1 from training data and structural information of the net. OBS permits a 90 , a 76 , and a 62 reduction in weights over backpropagation with weight decay on three benchmark MONK's problems [, 1991]. Of OBS, Optimal Brain Damage, and magnitude-based methods, only OBS deletes the correct weights from a trained XOR network in every case. Finally, whereas Sejnowski and Rosenberg [1987] used 18,000 weights in their NETtalk network, we used OBS to prune a network to just 1560 weights, yielding better generalization.",
"Neural networks are both computationally intensive and memory intensive, making them difficult to deploy on embedded systems. Also, conventional networks fix the architecture before training starts; as a result, training cannot improve the architecture. To address these limitations, we describe a method to reduce the storage and computation required by neural networks by an order of magnitude without affecting their accuracy by learning only the important connections. Our method prunes redundant connections using a three-step method. First, we train the network to learn which connections are important. Next, we prune the unimportant connections. Finally, we retrain the network to fine tune the weights of the remaining connections. On the ImageNet dataset, our method reduced the number of parameters of AlexNet by a factor of 9x, from 61 million to 6.7 million, without incurring accuracy loss. Similar experiments with VGG-16 found that the number of parameters can be reduced by 13x, from 138 million to 10.3 million, again with no loss of accuracy.",
"Deep Neural nets (NNs) with millions of parameters are at the heart of many state-of-the-art computer vision systems today. However, recent works have shown that much smaller models can achieve similar levels of performance. In this work, we address the problem of pruning parameters in a trained NN model. Instead of removing individual weights one at a time as done in previous works, we remove one neuron at a time. We show how similar neurons are redundant, and propose a systematic way to remove them. Our experiments in pruning the densely connected layers show that we can remove upto 85 of the total parameters in an MNIST-trained network, and about 35 for AlexNet without significantly affecting performance. Our method can be applied on top of most networks with a fully connected layer to give a smaller network."
]
} |
1511.05497 | 2507936800 | 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. | Many methods have been proposed to train models that are deep, yet have a lower parameterisation than conventional networks. Collins and Kohli @cite_15 propose a sparsity inducing regulariser for backpropogation which promotes many weights to have zero magnitude. They achieve reduction in memory consumption when compared to traditionally trained models. In contrast, our method promotes to have a zero-magnitude. As a result, our overall objective function is much simpler to solve. Denil @cite_3 demonstrate that most of the parameters of a model can be given only a few parameters. At training time, they learn only a few parameters and predict the rest. Yang @cite_24 propose an , which is an efficient re-parametrization of fully-connected layer weights. This results in a reduction of complexity for weight storage and computation. | {
"cite_N": [
"@cite_24",
"@cite_15",
"@cite_3"
],
"mid": [
"2949825606",
"1570197553",
"2952899695"
],
"abstract": [
"The fully connected layers of a deep convolutional neural network typically contain over 90 of the network parameters, and consume the majority of the memory required to store the network parameters. Reducing the number of parameters while preserving essentially the same predictive performance is critically important for operating deep neural networks in memory constrained environments such as GPUs or embedded devices. In this paper we show how kernel methods, in particular a single Fastfood layer, can be used to replace all fully connected layers in a deep convolutional neural network. This novel Fastfood layer is also end-to-end trainable in conjunction with convolutional layers, allowing us to combine them into a new architecture, named deep fried convolutional networks, which substantially reduces the memory footprint of convolutional networks trained on MNIST and ImageNet with no drop in predictive performance.",
"In this work, we investigate the use of sparsity-inducing regularizers during training of Convolution Neural Networks (CNNs). These regularizers encourage that fewer connections in the convolution and fully connected layers take non-zero values and in effect result in sparse connectivity between hidden units in the deep network. This in turn reduces the memory and runtime cost involved in deploying the learned CNNs. We show that training with such regularization can still be performed using stochastic gradient descent implying that it can be used easily in existing codebases. Experimental evaluation of our approach on MNIST, CIFAR, and ImageNet datasets shows that our regularizers can result in dramatic reductions in memory requirements. For instance, when applied on AlexNet, our method can reduce the memory consumption by a factor of four with minimal loss in accuracy.",
"We demonstrate that there is significant redundancy in the parameterization of several deep learning models. Given only a few weight values for each feature it is possible to accurately predict the remaining values. Moreover, we show that not only can the parameter values be predicted, but many of them need not be learned at all. We train several different architectures by learning only a small number of weights and predicting the rest. In the best case we are able to predict more than 95 of the weights of a network without any drop in accuracy."
]
} |
1511.05497 | 2507936800 | 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. | Some recent works have also focussed on using approximations of weight matrices to perform compression. Jaderberg @cite_17 and Denton @cite_20 use SVD-based low rank approximations of the weight matrix. Gong @cite_23 use a clustering-based product quantization approach to build an indexing scheme that reduces the space occupied by the matrix on disk. | {
"cite_N": [
"@cite_23",
"@cite_20",
"@cite_17"
],
"mid": [
"1724438581",
"2167215970",
"2950967261"
],
"abstract": [
"Deep convolutional neural networks (CNN) has become the most promising method for object recognition, repeatedly demonstrating record breaking results for image classification and object detection in recent years. However, a very deep CNN generally involves many layers with millions of parameters, making the storage of the network model to be extremely large. This prohibits the usage of deep CNNs on resource limited hardware, especially cell phones or other embedded devices. In this paper, we tackle this model storage issue by investigating information theoretical vector quantization methods for compressing the parameters of CNNs. In particular, we have found in terms of compressing the most storage demanding dense connected layers, vector quantization methods have a clear gain over existing matrix factorization methods. Simply applying k-means clustering to the weights or conducting product quantization can lead to a very good balance between model size and recognition accuracy. For the 1000-category classification task in the ImageNet challenge, we are able to achieve 16-24 times compression of the network with only 1 loss of classification accuracy using the state-of-the-art CNN.",
"We present techniques for speeding up the test-time evaluation of large convolutional networks, designed for object recognition tasks. These models deliver impressive accuracy, but each image evaluation requires millions of floating point operations, making their deployment on smartphones and Internet-scale clusters problematic. The computation is dominated by the convolution operations in the lower layers of the model. We exploit the redundancy present within the convolutional filters to derive approximations that significantly reduce the required computation. Using large state-of-the-art models, we demonstrate speedups of convolutional layers on both CPU and GPU by a factor of 2 x, while keeping the accuracy within 1 of the original model.",
"The focus of this paper is speeding up the evaluation of convolutional neural networks. While delivering impressive results across a range of computer vision and machine learning tasks, these networks are computationally demanding, limiting their deployability. Convolutional layers generally consume the bulk of the processing time, and so in this work we present two simple schemes for drastically speeding up these layers. This is achieved by exploiting cross-channel or filter redundancy to construct a low rank basis of filters that are rank-1 in the spatial domain. Our methods are architecture agnostic, and can be easily applied to existing CPU and GPU convolutional frameworks for tuneable speedup performance. We demonstrate this with a real world network designed for scene text character recognition, showing a possible 2.5x speedup with no loss in accuracy, and 4.5x speedup with less than 1 drop in accuracy, still achieving state-of-the-art on standard benchmarks."
]
} |
1511.05662 | 2951235355 | Plan recognition aims to discover target plans (i.e., sequences of actions) behind observed actions, with history plan libraries or domain models in hand. Previous approaches either discover plans by maximally "matching" observed actions to plan libraries, assuming target plans are from plan libraries, or infer plans by executing domain models to best explain the observed actions, assuming complete domain models are available. In real world applications, however, target plans are often not from plan libraries and complete domain models are often not available, since building complete sets of plans and complete domain models are often difficult or expensive. In this paper we view plan libraries as corpora and learn vector representations of actions using the corpora; we then discover target plans based on the vector representations. Our approach is capable of discovering underlying plans that are not from plan libraries, without requiring domain models provided. We empirically demonstrate the effectiveness of our approach by comparing its performance to traditional plan recognition approaches in three planning domains. | Instead of using a library of plans, Ramirez and Geffner @cite_28 proposed an approach to solving the plan recognition problem using slightly modified planning algorithms, assuming the action models were given as input. Except previous work @cite_38 @cite_9 @cite_3 @cite_28 on the plan recognition problem presented in the introduction section, Note that action models can be created by experts or learnt by previous systems, such as ARMS @cite_5 and LAMP @cite_25 . Saria and Mahadevan presented a hierarchical multi-agent markov processes as a framework for hierarchical probabilistic plan recognition in cooperative multi-agent systems @cite_14 . Singla and Mooney proposed an approach to abductive reasoning using a first-order probabilistic logic to recognize plans @cite_11 . Amir and Gal addressed a plan recognition approach to recognizing student behaviors using virtual science laboratories @cite_4 . Ramirez and Geffner exploited off-the-shelf classical planners to recognize probabilistic plans @cite_32 . | {
"cite_N": [
"@cite_38",
"@cite_14",
"@cite_4",
"@cite_28",
"@cite_9",
"@cite_32",
"@cite_3",
"@cite_5",
"@cite_25",
"@cite_11"
],
"mid": [
"94200858",
"",
"2175204249",
"1575558843",
"1735309676",
"1572384134",
"1981637451",
"2024113144",
"",
"2206942178"
],
"abstract": [
"This paper outlines a new theory of plan recognition that is significantly more powerful than previous approaches. Concurrent actions, shared steps between actions, and disjunctive information are all handled. The theory allows one to draw conclusions based on the class of possible plans being performed, rather than having to prematurely commit to a single interpretation. The theory employs circumscription to transform a first-order theory of action into an action taxonomy, which can be used to logically deduce the complex action(s) an agent is performing.",
"",
"This paper presents a plan recognition algorithm for inferring student behavior using virtual science laboratories. The algorithm extends existing plan recognition technology and was integrated with an existing educational application for chemistry. Automatic recognition of students' activities in virtual laboratories can provide important information to teachers as well as serve as the basis for intelligent tutoring. Student use of virtual laboratories presents several challenges: Students may repeat activities indefinitely, interleave between activities, and engage in exploratory behavior using trial-and-error. The plan recognition algorithm uses a recursive grammar that heuristically generates plans on the fly, taking into account chemical reactions and effects to determine students' intended high-level actions. The algorithm was evaluated empirically on data obtained from college students using virtual laboratory software for teaching chemistry. Results show that the algorithm was able to (1) infer the plans used by students to construct their models; (2) recognize such key processes as titration and dilution when they occurred in students' work; (3) identify partial solutions; (4) isolate sequences of actions that were part of a single error.",
"In this work we aim to narrow the gap between plan recognition and planning by exploiting the power and generality of recent planning algorithms for recognizing the set G* of goals G that explain a sequence of observations given a domain theory. After providing a crisp definition of this set, we show by means of a suitable problem transformation that a goal G belongs to G* if there is an action sequence π that is an optimal plan for both the goal G and the goal G extended with extra goals representing the observations. Exploiting this result, we show how the set G* can be computed exactly and approximately by minor modifications of existing optimal and suboptimal planning algorithms, and existing polynomial heuristics. Experiments over several domains show that the suboptimal planning algorithms and the polynomial heuristics provide good approximations of the optimal goal set G* while scaling up as well as state-of-the-art planning algorithms and heuristics.",
"We present a new general framework for online probabilistic plan recognition called the Abstract Hidden Markov Memory Model (AHMEM). The new model is an extension of the existing Abstract Hidden Markov Model to allow the policy to have internal memory which can be updated in a Markov fashion. We show that the A H M E M can repre sent a richer class of probabilistic plans, and at the same time derive an efficient algorithm for plan recognition in the A H M E M based on the RaoBlackwellised Particle Filter approximate inference method.",
"Plan recognition is the problem of inferring the goals and plans of an agent after observing its behavior. Recently, it has been shown that this problem can be solved efficiently, without the need of a plan library, using slightly modified planning algorithms. In this work, we extend this approach to the more general problem of probabilistic plan recognition where a probability distribution over the set of goals is sought under the assumptions that actions have deterministic effects and both agent and observer have complete information about the initial state. We show that this problem can be solved efficiently using classical planners provided that the probability of a partially observed execution given a goal is defined in terms of the cost difference of achieving the goal under two conditions: complying with the observations, and not complying with them. This cost, and hence the posterior goal probabilities, are computed by means of two calls to a classical planner that no longer has to be modified in any way. A number of examples is considered to illustrate the quality, flexibility, and scalability of the approach.",
"We present the PHATT algorithm for plan recognition. Unlike previous approaches to plan recognition, PHATT is based on a model of plan execution. We show that this clarifies several difficult issues in plan recognition including the execution of multiple interleaved root goals, partially ordered plans, and failing to observe actions. We present the PHATT algorithm's theoretical basis, and an implementation based on tree structures. We also investigate the algorithm's complexity, both analytically and empirically. Finally, we present PHATT's integrated constraint reasoning for parametrized actions and temporal constraints.",
"AI planning requires the definition of action models using a formal action and plan description language, such as the standard Planning Domain Definition Language (PDDL), as input. However, building action models from scratch is a difficult and time-consuming task, even for experts. In this paper, we develop an algorithm called ARMS (action-relation modelling system) for automatically discovering action models from a set of successful observed plans. Unlike the previous work in action-model learning, we do not assume complete knowledge of states in the middle of observed plans. In fact, our approach works when no or partial intermediate states are given. These example plans are obtained by an observation agent who does not know the logical encoding of the actions and the full state information between the actions. In a real world application, the cost is prohibitively high in labelling the training examples by manually annotating every state in a plan example from snapshots of an environment. To learn action models, ARMS gathers knowledge on the statistical distribution of frequent sets of actions in the example plans. It then builds a weighted propositional satisfiability (weighted MAX-SAT) problem and solves it using a MAX-SAT solver. We lay the theoretical foundations of the learning problem and evaluate the effectiveness of ARMS empirically.",
"",
"Plan recognition is a form of abductive reasoning that involves inferring plans that best explain sets of observed actions. Most existing approaches to plan recognition and other abductive tasks employ either purely logical methods that do not handle uncertainty, or purely probabilistic methods that do not handle structured representations. To overcome these limitations, this paper introduces an approach to abductive reasoning using a first-order probabilistic logic, specifically Markov Logic Networks (MLNs). It introduces several novel techniques for making MLNs efficient and effective for abduction. Experiments on three plan recognition datasets show the benefit of our approach over existing methods."
]
} |
1511.05662 | 2951235355 | Plan recognition aims to discover target plans (i.e., sequences of actions) behind observed actions, with history plan libraries or domain models in hand. Previous approaches either discover plans by maximally "matching" observed actions to plan libraries, assuming target plans are from plan libraries, or infer plans by executing domain models to best explain the observed actions, assuming complete domain models are available. In real world applications, however, target plans are often not from plan libraries and complete domain models are often not available, since building complete sets of plans and complete domain models are often difficult or expensive. In this paper we view plan libraries as corpora and learn vector representations of actions using the corpora; we then discover target plans based on the vector representations. Our approach is capable of discovering underlying plans that are not from plan libraries, without requiring domain models provided. We empirically demonstrate the effectiveness of our approach by comparing its performance to traditional plan recognition approaches in three planning domains. | Early work on human-in-the-loop planning scenarios in automated planning went under the name of mixed-initiative planning'' (e.g. @cite_6 ). An important limitation of that work was that the humans in the loop were helping the automated planner (with a complete action model) navigate its search space of plans more efficiently. In contrast, we are interested in planning technology that helping humans develop plans, even in the absence of complete formal models of the planning domain. While some work in web-service composition (c.f. @cite_2 ) did focus on this type of planning support, they were hobbled by being limited to simple input output type comparison. In contrast, we believe that DUP learns and uses a model that captures more of the structure of the planning domain (while still not insisting on complete action models). | {
"cite_N": [
"@cite_6",
"@cite_2"
],
"mid": [
"1604986327",
"2171104838"
],
"abstract": [
"We have been examining mixed-initiative planning systems in the context of command and control or logistical overview situations. In such environments, the human and the computer must work together in a very tightly coupled way to solve problems that neither alone could manage. In this paper, we describe our implementation of a prototype version of such a system, TRAINS-95, which helps a manager solve routing problems in a simple transportation domain. Interestingly perhaps, traditional planning technology does not play a major role in the system, and in fact it is difficult to see how such components might fit into a mixed-initiative system. We describe some of these issues, and present our agenda for future research into mixed-initiative plan reasoning. At this writing, the TRAINS-95 system has been used by more than 100 people to solve simple problems at various conferences and workshops, and in our experiments.",
"Web services are loosely coupled software components, published, located, and invoked across the web. The growing number of web services available within an organization and on the Web raises a new and challenging search problem: locating desired web services. Traditional keyword search is insufficient in this context: the specific types of queries users require are not captured, the very small text fragments in web services are unsuitable for keyword search, and the underlying structure and semantics of the web services are not exploited. We describe the algorithms underlying the Woogle search engine for web services. Woogle supports similarity search for web services, such as finding similar web-service operations and finding operations that compose with a given one. We describe novel techniques to support these types of searches, and an experimental study on a collection of over 1500 web-service operations that shows the high recall and precision of our algorithms."
]
} |
1511.05662 | 2951235355 | Plan recognition aims to discover target plans (i.e., sequences of actions) behind observed actions, with history plan libraries or domain models in hand. Previous approaches either discover plans by maximally "matching" observed actions to plan libraries, assuming target plans are from plan libraries, or infer plans by executing domain models to best explain the observed actions, assuming complete domain models are available. In real world applications, however, target plans are often not from plan libraries and complete domain models are often not available, since building complete sets of plans and complete domain models are often difficult or expensive. In this paper we view plan libraries as corpora and learn vector representations of actions using the corpora; we then discover target plans based on the vector representations. Our approach is capable of discovering underlying plans that are not from plan libraries, without requiring domain models provided. We empirically demonstrate the effectiveness of our approach by comparing its performance to traditional plan recognition approaches in three planning domains. | While DUP focuses on learning models from plan corpora, some recent work looked at using crowdsourcing to acquire domain models. For example, @cite_0 introduce Legion:AR, which combines the benefits of automatic and human activity labeling for robust and deployable activity recognition. The system exploits an active learning approach @cite_12 in which automatic activity recognition is augmented with on-demand activity labels from the crowd when an observed activity cannot be confidently classified. By engaging a group of people, Legion:AR is able to label activities as they occur more reliably than a single person can, especially in complex domains with multiple actors performing activities quickly. @cite_26 built a crowdsourcing based system called ARchitect, using the crowd to capture the dependency structure of the actions that make up activities. Such crowd-sourcing methods can complement the plan-corpus based approach proposed in DUP. | {
"cite_N": [
"@cite_0",
"@cite_26",
"@cite_12"
],
"mid": [
"2088664555",
"2080951892",
"2188373600"
],
"abstract": [
"Systems that automatically recognize human activities offer the potential of timely, task-relevant information and support. For example, prompting systems can help keep people with cognitive disabilities on track and surveillance systems can warn of activities of concern. Current automatic systems are difficult to deploy because they cannot identify novel activities, and, instead, must be trained in advance to recognize important activities. Identifying and labeling these events is time consuming and thus not suitable for real-time support of already-deployed activity recognition systems. In this paper, we introduce Legion:AR, a system that provides robust, deployable activity recognition by supplementing existing recognition systems with on-demand, real-time activity identification using input from the crowd. Legion:AR uses activity labels collected from crowd workers to train an automatic activity recognition system online to automatically recognize future occurrences. To enable the crowd to keep up with real-time activities, Legion:AR intelligently merges input from multiple workers into a single ordered label set. We validate Legion:AR across multiple domains and crowds and discuss features that allow appropriate privacy and accuracy tradeoffs.",
"Activity recognition can provide computers with the context underlying user inputs, enabling more relevant responses and more fluid interaction. However, training these systems is difficult because it requires observing every possible sequence of actions that comprise a given activity. Prior work has enabled the crowd to provide labels in real-time to train automated systems on-the-fly, but numerous examples are still needed before the system can recognize an activity on its own. To reduce the need to collect this data by observing users, we introduce ARchitect, a system that uses the crowd to capture the dependency structure of the actions that make up activities. Our tests show that over seven times as many examples can be collected using our approach versus relying on direct observation alone, demonstrating that by leveraging the understanding of the crowd, it is possible to more easily train automated systems.",
"Recognizing human activities from wearable sensor data is an important problem, particularly for health and eldercare applications. However, collecting sufficient labeled training data is challenging, especially since interpreting IMU traces is difficult for human annotators. Recently, crowdsourcing through services such as Amazon's Mechanical Turk has emerged as a promising alternative for annotating such data, with active learning (Cohn, Ghahramani, and Jordan 1996) serving as a natural method for affordably selecting an appropriate subset of instances to label. Unfortunately, since most active learning strategies are greedy methods that select the most uncertain sample, they are very sensitive to annotation errors (which corrupt a significant fraction of crowdsourced labels). This paper proposes methods for robust active learning under these conditions. Specifically, we make three contributions: 1) we obtain better initial labels by asking labelers to solve a related task; 2) we propose a new principled method for selecting instances in active learning that is more robust to annotation noise; 3) we estimate confidence scores for labels acquired from MTurk and ask workers to relabel samples that receive low scores under this metric. The proposed method is shown to significantly outperform existing techniques both under controlled noise conditions and in real active learning scenarios. The resulting method trains classifiers that are close in accuracy to those trained using ground-truth data."
]
} |
1511.05688 | 2260429471 | Proposed methods for prediction interval estimation so far focus on cases where input variables are numerical. In datasets with solely nominal input variables, we observe records with the exact same input @math , but different real valued outputs due to the inherent noise in the system. Existing prediction interval estimation methods do not use representations that can accurately model such inherent noise in the case of nominal inputs. We propose a new prediction interval estimation method tailored for this type of data, which is prevalent in biology and medicine. We call this method Distribution Adaptive Prediction Interval Estimation given Nominal inputs (DAPIEN) and has four main phases. First, we select a distribution function that can best represent the inherent noise of the system for all unique inputs. Then we infer the parameters @math (e.g. @math ) of the selected distribution function for all unique input vectors @math and generate a new corresponding training set using pairs of @math . III). Then, we train a model to predict @math given a new @math . Finally, we calculate the prediction interval for a new sample using the inverse of the cumulative distribution function once the parameters @math is predicted by the trained model. We compared DAPIEN to the commonly used Bootstrap method on three synthetic datasets. Our results show that DAPIEN provides tighter prediction intervals while preserving the requested coverage when compared to Bootstrap. This work can facilitate broader usage of regression methods in medicine and biology where it is necessary to provide tight prediction intervals while preserving coverage when input variables are nominal. | An extensive review for neural network-based prediction intervals is provided in @cite_8 . Next we provide a concise summary of PI assessment measures and also discuss the latest developments in PI prediction methods, some of which have not been reviewed together before. | {
"cite_N": [
"@cite_8"
],
"mid": [
"2171666055"
],
"abstract": [
"This paper evaluates the four leading techniques proposed in the literature for construction of prediction intervals (PIs) for neural network point forecasts. The delta, Bayesian, bootstrap, and mean-variance estimation (MVE) methods are reviewed and their performance for generating high-quality PIs is compared. PI-based measures are proposed and applied for the objective and quantitative assessment of each method's performance. A selection of 12 synthetic and real-world case studies is used to examine each method's performance for PI construction. The comparison is performed on the basis of the quality of generated PIs, the repeatability of the results, the computational requirements and the PIs variability with regard to the data uncertainty. The obtained results in this paper indicate that: 1) the delta and Bayesian methods are the best in terms of quality and repeatability, and 2) the MVE and bootstrap methods are the best in terms of low computational load and the width variability of PIs. This paper also introduces the concept of combinations of PIs, and proposes a new method for generating combined PIs using the traditional PIs. Genetic algorithm is applied for adjusting the combiner parameters through minimization of a PI-based cost function subject to two sets of restrictions. It is shown that the quality of PIs produced by the combiners is dramatically better than the quality of PIs obtained from each individual method."
]
} |
1511.05552 | 2177436562 | Recurrent Neural Networks (RNNs) have the ability to retain memory and learn data sequences. Due to the recurrent nature of RNNs, it is sometimes hard to parallelize all its computations on conventional hardware. CPUs do not currently offer large parallelism, while GPUs offer limited parallelism due to sequential components of RNN models. In this paper we present a hardware implementation of Long-Short Term Memory (LSTM) recurrent network on the programmable logic Zynq 7020 FPGA from Xilinx. We implemented a RNN with @math layers and @math hidden units in hardware and it has been tested using a character level language model. The implementation is more than @math faster than the ARM CPU embedded on the Zynq 7020 FPGA. This work can potentially evolve to a RNN co-processor for future mobile devices. | This work presents a different approach of implementing RNN in FPGA, focusing the LSTM architecture. It is different than @cite_12 , in the sense that it uses a single module that consumes input @math and previous output @math simultaneously. | {
"cite_N": [
"@cite_12"
],
"mid": [
"1492347181"
],
"abstract": [
"Recurrent neural network (RNN) based language model (RNNLM) is a biologically inspired model for natural language processing. It records the historical information through additional recurrent connections and therefore is very effective in capturing semantics of sentences. However, the use of RNNLM has been greatly hindered for the high computation cost in training. This work presents an FPGA implementation framework for RNNLM training acceleration. At architectural level, we improve the parallelism of RNN training scheme and reduce the computing resource requirement for computation efficiency enhancement. The hardware implementation primarily targets at reducing data communication load. A multi-thread based computation engine is utilized which can successfully mask the long memory latency and reuse frequent accessed data. The evaluation based on the Microsoft Research Sentence Completion Challenge shows that the proposed FPGA implementation outperforms traditional class-based modest-size recurrent networks and obtains 46.2 in training accuracy. Moreover, experiments at different network sizes demonstrate a great scalability of the proposed framework."
]
} |
1511.05618 | 2272403233 | User behaviour analysis based on traffic log in wireless networks can be beneficial to many fields in real life: not only for commercial purposes, but also for improving network service quality and social management. We cluster users into groups marked by the most frequently visited websites to find their preferences. In this paper, we propose a user behaviour model based on Topic Model from document classification problems. We use the logarithmic TF-IDF (term frequency - inverse document frequency) weighing to form a high-dimensional sparse feature matrix. Then we apply LSA (Latent semantic analysis) to deduce the latent topic distribution and generate a low-dimensional dense feature matrix. K-means++, which is a classic clustering algorithm, is then applied to the dense feature matrix and several interpretable user clusters are found. Moreover, by combining the clustering results with additional demographical information, including age, gender, and financial information, we are able to uncover more realistic implications from the clustering results. | With the rapid development of wireless networks, the potential of user behaviour analysis has brought up tremendous attention recently. The most common method for user clustering is by applying K-means on raw profile matrix. For example, the web browsing similarity among users of a university laboratory is examined using web log data in in @cite_4 . Besides, in @cite_2 , the authors evaluate the effectiveness of K-Means and spectral clustering and discuss the effect of the profile size to the clustering results. The authors in @cite_12 discuss this personalization problem from another perspective. They apply K-Means to construct user profiles using device-side traffic logs. The authors also point out that the extra information on gender may help with the problem in future work. | {
"cite_N": [
"@cite_4",
"@cite_12",
"@cite_2"
],
"mid": [
"2031610968",
"2156151108",
"2115952768"
],
"abstract": [
"As one of the most important tasks of Web Usage Mining (WUM), web user clustering, which establishes groups of users exhibiting similar browsing patterns, provides useful knowledge to personalized web services. In this paper, we cluster web users with KMeans algorithm based on web user log data. Given a set of web users and their associated historical web usage data, we study their behavior characteristic and cluster them. Experiment results show the feasibility and efficiency of such algorithm application. Web user clusters generated in this way can provide novel and useful knowledge for various personalized web applications.",
"Mobile device can be used as a medium to send and receive the mobile internet content. However, there are several limitations using mobile internet. Content personalisation has been viewed as an important area when using mobile internet. In order for personalisation to be successful, understanding the user is important. In this paper, we explore the implementation of the user profile at client-side, which may be used whenever user connect to the mobile content provider. The client-side user profile can help to free the provider in performing analysis by using data mining technique at the mobile device. This research investigates the conceptual idea of using clustering and classification of user profile at the client-site mobile. In this paper, we applied K-means and compared several other classification algorithms like TwoStep, Kohenen and Anomaly to determine the boundaries of the important factors using information ranking separation.",
"In this paper, we introduce and evaluate a framework for a user profile-based socializing application. We model the user profile as an unordered set composed of n keywords that represent users' preferences. We evaluate the effectiveness of two clustering algorithms, k-means and spectral clustering in the scope of social groups' assessment. Through experimental results we substantiate the applicability of spectral clustering in the examined service and we evaluate the impact of profile size in terms of the quality of partitions yielded by spectral clustering. The above are performed using case studies and scenarios developed in the context of 1ST project's MAGNET Beyond pilot services definition."
]
} |
1511.05618 | 2272403233 | User behaviour analysis based on traffic log in wireless networks can be beneficial to many fields in real life: not only for commercial purposes, but also for improving network service quality and social management. We cluster users into groups marked by the most frequently visited websites to find their preferences. In this paper, we propose a user behaviour model based on Topic Model from document classification problems. We use the logarithmic TF-IDF (term frequency - inverse document frequency) weighing to form a high-dimensional sparse feature matrix. Then we apply LSA (Latent semantic analysis) to deduce the latent topic distribution and generate a low-dimensional dense feature matrix. K-means++, which is a classic clustering algorithm, is then applied to the dense feature matrix and several interpretable user clusters are found. Moreover, by combining the clustering results with additional demographical information, including age, gender, and financial information, we are able to uncover more realistic implications from the clustering results. | Besides K-means, other clustering algorithms are also investigated in the literature. The authors in @cite_6 present a Greedy Clustering Using Belief Function (GCB) algorithm. The representatives of the clusters are picked iteratively, so that the current representative is well separated from the former ones. In @cite_3 , the authors analyze the online activity and mobility for thousands of mobile users. They use Self-Organized Map (SOM) for discovering mobile users' trends. In @cite_7 , the authors use a information theoretic co-clustering method on user-domain matrix and consider the effect of location. | {
"cite_N": [
"@cite_3",
"@cite_7",
"@cite_6"
],
"mid": [
"2021799309",
"1999594085",
"2153842973"
],
"abstract": [
"User online behavior and interests will play a central role in future mobile networks. We introduce a systematic method for large-scale multi-dimensional modeling and analysis of online activity and mobility for thousands of mobile users across 79 buildings over a variety of web domains. We propose a modeling approach based on kind of neural-networks, called self-organizing maps (SOM), for discovering, organizing and visualizing different mobile users' trends from billions of WLAN records. We find surprisingly that users' trends based on domains and locations can be accurately modeled using a self-organizing map with clearly distinct characteristics. We also find many non-trivial correlations between different types of web domains and locations.",
"Design and simulation of future mobile networks will center around human interests and behavior. We propose a design paradigm for mobile networks driven by realistic models of users' on-line behavior, based on mining of billions of wireless-LAN records. We introduce a systematic method for large-scale multi-dimensional co-clustering of web activity for thousands of mobile users at 79 locations. We find surprisingly that users can be consistently modeled using ten clusters with disjoint profiles. Access patterns from multiple locations show differential user behavior. This is the first study to obtain such detailed results for mobile Internet usage.",
"In this work, we present a novel approach to clustering Web site users into different groups and generating common user profiles. These profiles can be used to make recommendations, personalize Web sites, and for other uses such as targeting users for advertising. By using the concept of mass distribution in Dempster-Shafer's theory, the belief function similarity measure in our algorithm adds to the clustering task the ability to capture the uncertainty among Web user's navigation behavior. Our algorithm is relatively simple to use and gives comparable results to other approaches reported in the literature of web mining."
]
} |
1511.05618 | 2272403233 | User behaviour analysis based on traffic log in wireless networks can be beneficial to many fields in real life: not only for commercial purposes, but also for improving network service quality and social management. We cluster users into groups marked by the most frequently visited websites to find their preferences. In this paper, we propose a user behaviour model based on Topic Model from document classification problems. We use the logarithmic TF-IDF (term frequency - inverse document frequency) weighing to form a high-dimensional sparse feature matrix. Then we apply LSA (Latent semantic analysis) to deduce the latent topic distribution and generate a low-dimensional dense feature matrix. K-means++, which is a classic clustering algorithm, is then applied to the dense feature matrix and several interpretable user clusters are found. Moreover, by combining the clustering results with additional demographical information, including age, gender, and financial information, we are able to uncover more realistic implications from the clustering results. | Many similarity measures between users are studied. The authors in @cite_10 use dataset containing traces of 34 users and mainly deal with temporal dimension. In their work, different similarity matrices including Cosine Similarity, Hellinger distance and SVD are compared with each other. The authors in @cite_6 apply a belief function as the similarity measure. | {
"cite_N": [
"@cite_10",
"@cite_6"
],
"mid": [
"1969086082",
"2153842973"
],
"abstract": [
"In this paper we explore the notion of mobile users' similarity as a key enabler of innovative applications hinging on opportunistic mobile encounters. In particular, we analyze the performance of known similarity metrics, applicable to our problem domain, as well as propose a novel temporal-based metric, in an attempt to quantify the inherently qualitative notion of similarity. Towards this objective, we first introduce generalized profile structures, beyond mere location, that aim to capture users interests and prior experiences, in the form of a probability distribution. Afterwards, we analyze known and proposed similarity metrics for the proposed profile structures using publicly available data. Apart from the classic Cosine similarity, we identify a distance metric from probability theory, namely Hellinger distance, as a strong candidate for quantifying similarity due to the probability distribution structure of the proposed profiles. In addition, we introduce a novel temporal similarity metric, based on matrix vectorization, to capitalize on the richness in the temporal dimension and maintain low complexity. Finally, the numerical results unveil a number of key insights. First, the temporal metrics yield, on the average, lower similarity indices, compared to the non-temporal ones, due to incorporating the dynamics in the temporal dimension. Second, the Hellinger distance holds great promise for quantifying similarity between probability distribution profiles. Third, vectorized metrics constitute a low-complexity approach towards temporal similarity on resource-limited mobile devices.",
"In this work, we present a novel approach to clustering Web site users into different groups and generating common user profiles. These profiles can be used to make recommendations, personalize Web sites, and for other uses such as targeting users for advertising. By using the concept of mass distribution in Dempster-Shafer's theory, the belief function similarity measure in our algorithm adds to the clustering task the ability to capture the uncertainty among Web user's navigation behavior. Our algorithm is relatively simple to use and gives comparable results to other approaches reported in the literature of web mining."
]
} |
1511.05618 | 2272403233 | User behaviour analysis based on traffic log in wireless networks can be beneficial to many fields in real life: not only for commercial purposes, but also for improving network service quality and social management. We cluster users into groups marked by the most frequently visited websites to find their preferences. In this paper, we propose a user behaviour model based on Topic Model from document classification problems. We use the logarithmic TF-IDF (term frequency - inverse document frequency) weighing to form a high-dimensional sparse feature matrix. Then we apply LSA (Latent semantic analysis) to deduce the latent topic distribution and generate a low-dimensional dense feature matrix. K-means++, which is a classic clustering algorithm, is then applied to the dense feature matrix and several interpretable user clusters are found. Moreover, by combining the clustering results with additional demographical information, including age, gender, and financial information, we are able to uncover more realistic implications from the clustering results. | A closely recent related work is @cite_5 . Same as us, the authors study user behaviour modeling in a perspective of topic modeling. By way of analogy, the authors find that there are many similarities between user behaviour profiling and document classification in natural language processing, which is also used in our modeling. But in their work, the authors generate their feature matrix by oversampling the URLs, while we use TF-IDF weights. The data comes from cellualr networks in their paper. But the datasets we use are clollected from WLAN. In @cite_11 , the author proposes a similar procedure in analyzing user preferences with our algorithm for webpage prefetching. But the data set used in the paper is small and there are only around 100 domains included. Therefore, user behaviour cannot be fully presented. | {
"cite_N": [
"@cite_5",
"@cite_11"
],
"mid": [
"2209343521",
"69386222"
],
"abstract": [
"Insights into the behavior and preference of mobile device users from their web browsing application activities are critical components of any successful dynamic content recommendation system, mobile advertisement platform, or web personalization initiative. In this paper we use an unsupervised topic model to understand the interests of the cellular users based upon their browsing profile. We posit that the length of time a user remains on a given website is positively correlated with the user’s interest in the website’s content. We propose an extended model to integrate this duration information efficiently by oversampling the URLs.",
"This study evaluates methods for clustering web users via vector space models, for the purpose of web page prefetching for possible applications of server optimization. An experiment using Latent S ..."
]
} |
1511.05547 | 2173393671 | Unlike human learning, machine learning often fails to handle changes between training (source) and test (target) input distributions. Such domain shifts, common in practical scenarios, severely damage the performance of conventional machine learning methods. Supervised domain adaptation methods have been proposed for the case when the target data have labels, including some that perform very well despite being "frustratingly easy" to implement. However, in practice, the target domain is often unlabeled, requiring unsupervised adaptation. We propose a simple, effective, and efficient method for unsupervised domain adaptation called CORrelation ALignment (CORAL). CORAL minimizes domain shift by aligning the second-order statistics of source and target distributions, without requiring any target labels. Even though it is extraordinarily simple--it can be implemented in four lines of Matlab code--CORAL performs remarkably well in extensive evaluations on standard benchmark datasets. | Domain shift is a fundamental problem in machine learning, and has also attracted a lot of attention in the speech, natural language and vision communities @cite_33 @cite_38 @cite_21 . For supervised adaptation, a variety of techniques have been proposed. Some consider the source domain as a prior that regularizes the learning problem in the sparsely labeled target domain, e.g., @cite_31 . Others minimize the distance between the target and source domains, either by re-weighting the domains or by changing the feature representation according to some explicit distribution distance metric @cite_14 . Some learn a transformation on features using a contrastive loss @cite_16 . Arguably the simplest and most prominent supervised approach is the frustratingly easy'' feature replication @cite_28 . Given a feature vector @math , it defines the augmented feature vector @math for data points in the source and @math for data points in the target. A classifier is then trained on augmented features. This approach is simple, however, it requires labeled target examples, which are often not available in real world applications. | {
"cite_N": [
"@cite_38",
"@cite_14",
"@cite_33",
"@cite_28",
"@cite_21",
"@cite_31",
"@cite_16"
],
"mid": [
"2128053425",
"2164943005",
"2163302275",
"2952362487",
"2083544878",
"",
"1722318740"
],
"abstract": [
"Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.",
"Motivation: Many problems in data integration in bioinformatics can be posed as one common question: Are two sets of observations generated by the same distribution? We propose a kernel-based statistical test for this problem, based on the fact that two distributions are different if and only if there exists at least one function having different expectation on the two distributions. Consequently we use the maximum discrepancy between function means as the basis of a test statistic. The Maximum Mean Discrepancy (MMD) can take advantage of the kernel trick, which allows us to apply it not only to vectors, but strings, sequences, graphs, and other common structured data types arising in molecular biology. Results: We study the practical feasibility of an MMD-based test on three central data integration tasks: Testing cross-platform comparability of microarray data, cancer diagnosis, and data-content based schema matching for two different protein function classification schemas. In all of these experiments, including high-dimensional ones, MMD is very accurate in finding samples that were generated from the same distribution, and outperforms its best competitors. Conclusions: We have defined a novel statistical test of whether two samples are from the same distribution, compatible with both multivariate and structured data, that is fast, easy to implement, and works well, as confirmed by our experiments. Availability: Contact: [email protected]",
"Automatic sentiment classification has been extensively studied and applied in recent years. However, sentiment is expressed differently in different domains, and annotating corpora for every possible domain of interest is impractical. We investigate domain adaptation for sentiment classifiers, focusing on online reviews for different types of products. First, we extend to sentiment classification the recently-proposed structural correspondence learning (SCL) algorithm, reducing the relative error due to adaptation between domains by an average of 30 over the original SCL algorithm and 46 over a supervised baseline. Second, we identify a measure of domain similarity that correlates well with the potential for adaptation of a classifier from one domain to another. This measure could for instance be used to select a small set of domains to annotate whose trained classifiers would transfer well to many other domains.",
"We describe an approach to domain adaptation that is appropriate exactly in the case when one has enough target'' data to do slightly better than just using only source'' data. Our approach is incredibly simple, easy to implement as a preprocessing step (10 lines of Perl!) and outperforms state-of-the-art approaches on a range of datasets. Moreover, it is trivially extended to a multi-domain adaptation problem, where one has data from a variety of different domains.",
"The most successful 2D object detection methods require a large number of images annotated with object bounding boxes to be collected for training. We present an alternative approach that trains on virtual data rendered from 3D models, avoiding the need for manual labeling. Growing demand for virtual reality applications is quickly bringing about an abundance of available 3D models for a large variety of object categories. While mainstream use of 3D models in vision has focused on predicting the 3D pose of objects, we investigate the use of such freely available 3D models for multicategory 2D object detection. To address the issue of dataset bias that arises from training on virtual data and testing on real images, we propose a simple and fast adaptation approach based on decorrelated features. We also compare two kinds of virtual data, one rendered with real-image textures and one without. Evaluation on a benchmark domain adaptation dataset demonstrates that our method performs comparably to existing methods trained on large-scale real image domains.",
"",
"Domain adaptation is an important emerging topic in computer vision. In this paper, we present one of the first studies of domain shift in the context of object recognition. We introduce a method that adapts object models acquired in a particular visual domain to new imaging conditions by learning a transformation that minimizes the effect of domain-induced changes in the feature distribution. The transformation is learned in a supervised manner and can be applied to categories for which there are no labeled examples in the new domain. While we focus our evaluation on object recognition tasks, the transform-based adaptation technique we develop is general and could be applied to nonimage data. Another contribution is a new multi-domain object database, freely available for download. We experimentally demonstrate the ability of our method to improve recognition on categories with few or no target domain labels and moderate to large changes in the imaging conditions."
]
} |
1511.05547 | 2173393671 | Unlike human learning, machine learning often fails to handle changes between training (source) and test (target) input distributions. Such domain shifts, common in practical scenarios, severely damage the performance of conventional machine learning methods. Supervised domain adaptation methods have been proposed for the case when the target data have labels, including some that perform very well despite being "frustratingly easy" to implement. However, in practice, the target domain is often unlabeled, requiring unsupervised adaptation. We propose a simple, effective, and efficient method for unsupervised domain adaptation called CORrelation ALignment (CORAL). CORAL minimizes domain shift by aligning the second-order statistics of source and target distributions, without requiring any target labels. Even though it is extraordinarily simple--it can be implemented in four lines of Matlab code--CORAL performs remarkably well in extensive evaluations on standard benchmark datasets. | Early techniques for unsupervised adaptation consisted of re-weighting the training point losses to more closely reflect those in the test distribution @cite_32 @cite_4 . Dictionary learning methods @cite_36 @cite_17 try to learn a dictionary where the difference between the source and target domain is minimized in the new representation. Recent state-of-the-art unsupervised approaches @cite_38 @cite_23 @cite_18 @cite_25 have pursued adaptation by projecting the source and target distributions into a lower-dimensional manifold, and finding a transformation that brings the subspaces closer together. Geodesic methods find a path along the subspace manifold, and either project source and target onto points along that path @cite_38 , or find a closed-form linear map that projects source points to target @cite_23 . Alternatively, the subspaces can be aligned by computing the linear map that minimizes the Frobenius norm of the difference between them @cite_2 @cite_15 . However, these approaches only align the bases of the subspaces, not the distribution of the projected points. They also require expensive subspace projection and hyperparameter selection. | {
"cite_N": [
"@cite_38",
"@cite_18",
"@cite_4",
"@cite_36",
"@cite_32",
"@cite_23",
"@cite_2",
"@cite_15",
"@cite_25",
"@cite_17"
],
"mid": [
"2128053425",
"2057266281",
"2811380766",
"1985967702",
"2111362445",
"",
"2950657435",
"2104068492",
"1892837366",
"2136995369"
],
"abstract": [
"Adapting the classifier trained on a source domain to recognize instances from a new target domain is an important problem that is receiving recent attention. In this paper, we present one of the first studies on unsupervised domain adaptation in the context of object recognition, where we have labeled data only from the source domain (and therefore do not have correspondences between object categories across domains). Motivated by incremental learning, we create intermediate representations of data between the two domains by viewing the generative subspaces (of same dimension) created from these domains as points on the Grassmann manifold, and sampling points along the geodesic between them to obtain subspaces that provide a meaningful description of the underlying domain shift. We then obtain the projections of labeled source domain data onto these subspaces, from which a discriminative classifier is learnt to classify projected data from the target domain. We discuss extensions of our approach for semi-supervised adaptation, and for cases with multiple source and target domains, and report competitive results on standard datasets.",
"Visual domain adaptation, which learns an accurate classifier for a new domain using labeled images from an old domain, has shown promising value in computer vision yet still been a challenging problem. Most prior works have explored two learning strategies independently for domain adaptation: feature matching and instance reweighting. In this paper, we show that both strategies are important and inevitable when the domain difference is substantially large. We therefore put forward a novel Transfer Joint Matching (TJM) approach to model them in a unified optimization problem. Specifically, TJM aims to reduce the domain difference by jointly matching the features and reweighting the instances across domains in a principled dimensionality reduction procedure, and construct new feature representation that is invariant to both the distribution difference and the irrelevant instances. Comprehensive experimental results verify that TJM can significantly outperform competitive methods for cross-domain image recognition problems.",
"We consider the scenario where training and test data are drawn from different distributions, commonly referred to as sample selection bias. Most algorithms for this setting try to first recover sampling distributions and then make appropriate corrections based on the distribution estimate. We present a nonparametric method which directly produces resampling weights without distribution estimation. Our method works by matching distributions between training and testing sets in feature space. Experimental results demonstrate that our method works well in practice.",
"Data-driven dictionaries have produced state-of-the-art results in various classification tasks. However, when the target data has a different distribution than the source data, the learned sparse representation may not be optimal. In this paper, we investigate if it is possible to optimally represent both source and target by a common dictionary. Specifically, we describe a technique which jointly learns projections of data in the two domains, and a latent dictionary which can succinctly represent both the domains in the projected low-dimensional space. An efficient optimization technique is presented, which can be easily kernelized and extended to multiple domains. The algorithm is modified to learn a common discriminative dictionary, which can be further used for classification. The proposed approach does not require any explicit correspondence between the source and target domains, and shows good results even when there are only a few labels available in the target domain. Various recognition experiments show that the method performs on par or better than competitive state-of-the-art methods.",
"Domain adaptation is an important problem in natural language processing (NLP) due to the lack of labeled data in novel domains. In this paper, we study the domain adaptation problem from the instance weighting perspective. We formally analyze and characterize the domain adaptation problem from a distributional view, and show that there are two distinct needs for adaptation, corresponding to the different distributions of instances and classification functions in the source and the target domains. We then propose a general instance weighting framework for domain adaptation. Our empirical results on three NLP tasks show that incorporating and exploiting more information from the target domain through instance weighting is effective.",
"",
"We propose a novel problem formulation of learning a single task when the data are provided in different feature spaces. Each such space is called an outlook, and is assumed to contain both labeled and unlabeled data. The objective is to take advantage of the data from all the outlooks to better classify each of the outlooks. We devise an algorithm that computes optimal affine mappings from different outlooks to a target outlook by matching moments of the empirical distributions. We further derive a probabilistic interpretation of the resulting algorithm and a sample complexity bound indicating how many samples are needed to adequately find the mapping. We report the results of extensive experiments on activity recognition tasks that show the value of the proposed approach in boosting performance.",
"In this paper, we introduce a new domain adaptation (DA) algorithm where the source and target domains are represented by subspaces described by eigenvectors. In this context, our method seeks a domain adaptation solution by learning a mapping function which aligns the source subspace with the target one. We show that the solution of the corresponding optimization problem can be obtained in a simple closed form, leading to an extremely fast algorithm. We use a theoretical result to tune the unique hyper parameter corresponding to the size of the subspaces. We run our method on various datasets and show that, despite its intrinsic simplicity, it outperforms state of the art DA methods.",
"Recently, a particular paradigm [18] in the domain adaptation field has received considerable attention by introducing novel and important insights to the problem. In this case, the source and target domains are represented in the form of subspaces, which are treated as points on the Grassmann manifold. The geodesic curve between them is sampled to obtain intermediate points. Then a classifier is learnt using the projections of the data onto these subspaces. Despite its relevance and popularity, this paradigm [18] contains some limitations. Firstly, in real-world applications, that simple curve (i.e. shortest path) does not provide the necessary flexibility to model the domain shift between the training and testing data sets. Secondly, by using the geodesic curve, we are restricted to only one source domain, which does not allow to take fully advantage of the multiple datasets that are available nowadays. It is then, natural to ask whether this popular concept could be extended to deal with more complex curves and to integrate multi-sources domains. This is a hard problem considering the Riemannian structure of the space, but we propose a mathematically well-founded idea that enables us to solve it. We exploit the geometric insight of rolling maps [30] to compute a spline curve on the Grassmann manifold. The benefits of the proposed idea are demonstrated through several empirical studies on standard datasets. This novel concept allows to explicitly integrate multi-source domains while the previous one [18] uses the mean of all sources. This enables to model better the domain shift and take fully advantage of the training datasets.",
"Cross-domain image synthesis and recognition are typically considered as two distinct tasks in the areas of computer vision and pattern recognition. Therefore, it is not clear whether approaches addressing one task can be easily generalized or extended for solving the other. In this paper, we propose a unified model for coupled dictionary and feature space learning. The proposed learning model not only observes a common feature space for associating cross-domain image data for recognition purposes, the derived feature space is able to jointly update the dictionaries in each image domain for improved representation. This is why our method can be applied to both cross-domain image synthesis and recognition problems. Experiments on a variety of synthesis and recognition tasks such as single image super-resolution, cross-view action recognition, and sketch-to-photo face recognition would verify the effectiveness of our proposed learning model."
]
} |
1511.05547 | 2173393671 | Unlike human learning, machine learning often fails to handle changes between training (source) and test (target) input distributions. Such domain shifts, common in practical scenarios, severely damage the performance of conventional machine learning methods. Supervised domain adaptation methods have been proposed for the case when the target data have labels, including some that perform very well despite being "frustratingly easy" to implement. However, in practice, the target domain is often unlabeled, requiring unsupervised adaptation. We propose a simple, effective, and efficient method for unsupervised domain adaptation called CORrelation ALignment (CORAL). CORAL minimizes domain shift by aligning the second-order statistics of source and target distributions, without requiring any target labels. Even though it is extraordinarily simple--it can be implemented in four lines of Matlab code--CORAL performs remarkably well in extensive evaluations on standard benchmark datasets. | Adaptive deep neural networks have recently been explored for unsupervised adaptation. DLID @cite_7 trains a joint source and target CNN architecture, but is limited to two adaptation layers. ReverseGrad @cite_30 , DAN @cite_27 , and DDC @cite_12 directly optimize the deep representation for domain invariance, using additional loss layers designed for this purpose. Training with this additional loss is costly and can be sensitive to initialization, network structure, and other optimization settings. Our approach, applied to deep features (top layer activations), achieves better or comparable performance to these more complex methods, and can be incorporated directly into the network structure. | {
"cite_N": [
"@cite_30",
"@cite_27",
"@cite_12",
"@cite_7"
],
"mid": [
"1882958252",
"2951670162",
"1565327149",
"2186639548"
],
"abstract": [
"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.",
"Recent studies reveal that a deep neural network can learn transferable features which generalize well to novel tasks for domain adaptation. However, as deep features eventually transition from general to specific along the network, the feature transferability drops significantly in higher layers with increasing domain discrepancy. Hence, it is important to formally reduce the dataset bias and enhance the transferability in task-specific layers. In this paper, we propose a new Deep Adaptation Network (DAN) architecture, which generalizes deep convolutional neural network to the domain adaptation scenario. In DAN, hidden representations of all task-specific layers are embedded in a reproducing kernel Hilbert space where the mean embeddings of different domain distributions can be explicitly matched. The domain discrepancy is further reduced using an optimal multi-kernel selection method for mean embedding matching. DAN can learn transferable features with statistical guarantees, and can scale linearly by unbiased estimate of kernel embedding. Extensive empirical evidence shows that the proposed architecture yields state-of-the-art image classification error rates on standard domain adaptation benchmarks.",
"Recent reports suggest that a generic supervised deep CNN model trained on a large-scale dataset reduces, but does not remove, dataset bias on a standard benchmark. Fine-tuning deep models in a new domain can require a significant amount of data, which for many applications is simply not available. We propose a new CNN architecture which introduces an adaptation layer and an additional domain confusion loss, to learn a representation that is both semantically meaningful and domain invariant. We additionally show that a domain confusion metric can be used for model selection to determine the dimension of an adaptation layer and the best position for the layer in the CNN architecture. Our proposed adaptation method offers empirical performance which exceeds previously published results on a standard benchmark visual domain adaptation task.",
"In many real world applications of machine learning, the distribution of the training data (on which the machine learning model is trained) is dierent from the distribution of the test data (where the learnt model is actually deployed). This is known as the problem of Domain Adaptation. We propose a novel deep learning model for domain adaptation which attempts to learn a predictively useful representation of the data by taking into account information from the distribution shift between the training and test data. Our key proposal is to successively learn multiple intermediate representations along an path\" between the train and test domains. Our experiments on a standard object recognition dataset show a signicant performance improvement over the state-of-the-art."
]
} |
1511.05659 | 2952663444 | Heterogeneous gap among different modalities emerges as one of the critical issues in modern AI problems. Unlike traditional uni-modal cases, where raw features are extracted and directly measured, the heterogeneous nature of cross modal tasks requires the intrinsic semantic representation to be compared in a unified framework. This paper studies the learning of different representations that can be retrieved across different modality contents. A novel approach for mining cross-modal representations is proposed by incorporating explicit linear semantic projecting in Hilbert space. The insight is that the discriminative structures of different modality data can be linearly represented in appropriate high dimension Hilbert spaces, where linear operations can be used to approximate nonlinear decisions in the original spaces. As a result, an efficient linear semantic down mapping is jointly learned for multimodal data, leading to a common space where they can be compared. The mechanism of "feature up-lifting and down-projecting" works seamlessly as a whole, which accomplishes crossmodal retrieval tasks very well. The proposed method, named as shared discriminative semantic representation learning (), is tested on two public multimodal dataset for both within- and inter- modal retrieval. The experiments demonstrate that it outperforms several state-of-the-art methods in most scenarios. | One typical category of methods is based on Latent Dirichlet Allocation related topic models, such as Correspondence LDA @cite_8 , and Multimodal Document Random Field (MDRF) @cite_13 . The Correspondence LDA, extends Latent Dirichlet Allocation to find a topic-level relationship between images and text annotations, in which distributions of topics act as a middle semantic layer. The MDRF suggests a mixture probabilistic graphical model using a Markov random field over LDA. These methods are conducted in an unsupervised manner. | {
"cite_N": [
"@cite_13",
"@cite_8"
],
"mid": [
"2115752676",
"2020842694"
],
"abstract": [
"Many applications involve multiple-modalities such as text and images that describe the problem of interest. In order to leverage the information present in all the modalities, one must model the relationships between them. While some techniques have been proposed to tackle this problem, they either are restricted to words describing visual objects only, or require full correspondences between the different modalities. As a consequence, they are unable to tackle more realistic scenarios where a narrative text is only loosely related to an image, and where only a few image-text pairs are available. In this paper, we propose a model that addresses both these challenges. Our model can be seen as a Markov random field of topic models, which connects the documents based on their similarity. As a consequence, the topics learned with our model are shared across connected documents, thus encoding the relations between different modalities. We demonstrate the effectiveness of our model for image retrieval from a loosely related text.",
"We consider the problem of modeling annotated data---data with multiple types where the instance of one type (such as a caption) serves as a description of the other type (such as an image). We describe three hierarchical probabilistic mixture models which aim to describe such data, culminating in correspondence latent Dirichlet allocation, a latent variable model that is effective at modeling the joint distribution of both types and the conditional distribution of the annotation given the primary type. We conduct experiments on the Corel database of images and captions, assessing performance in terms of held-out likelihood, automatic annotation, and text-based image retrieval."
]
} |
1511.05659 | 2952663444 | Heterogeneous gap among different modalities emerges as one of the critical issues in modern AI problems. Unlike traditional uni-modal cases, where raw features are extracted and directly measured, the heterogeneous nature of cross modal tasks requires the intrinsic semantic representation to be compared in a unified framework. This paper studies the learning of different representations that can be retrieved across different modality contents. A novel approach for mining cross-modal representations is proposed by incorporating explicit linear semantic projecting in Hilbert space. The insight is that the discriminative structures of different modality data can be linearly represented in appropriate high dimension Hilbert spaces, where linear operations can be used to approximate nonlinear decisions in the original spaces. As a result, an efficient linear semantic down mapping is jointly learned for multimodal data, leading to a common space where they can be compared. The mechanism of "feature up-lifting and down-projecting" works seamlessly as a whole, which accomplishes crossmodal retrieval tasks very well. The proposed method, named as shared discriminative semantic representation learning (), is tested on two public multimodal dataset for both within- and inter- modal retrieval. The experiments demonstrate that it outperforms several state-of-the-art methods in most scenarios. | Some other work, like Semantic Correlative Matching @cite_2 and Multimodal Deep Learning @cite_7 , build lower-dimensional common representations based on canonical correlation analysis (CCA). Menon @ proposed to reduce cross-modal retrieval as binary classification over pairs. The framework can be seen as a variant of Semantic Matching. It is not difficult to find that, these approaches usually exploit the symbiosis of multimodal data in the assumption of there existing one-to-one strictly paired training data. | {
"cite_N": [
"@cite_7",
"@cite_2"
],
"mid": [
"2184188583",
"2138118304"
],
"abstract": [
"Deep networks have been successfully applied to unsupervised feature learning for single modalities (e.g., text, images or audio). In this work, we propose a novel application of deep networks to learn features over multiple modalities. We present a series of tasks for multimodal learning and show how to train deep networks that learn features to address these tasks. In particular, we demonstrate cross modality feature learning, where better features for one modality (e.g., video) can be learned if multiple modalities (e.g., audio and video) are present at feature learning time. Furthermore, we show how to learn a shared representation between modalities and evaluate it on a unique task, where the classifier is trained with audio-only data but tested with video-only data and vice-versa. Our models are validated on the CUAVE and AVLetters datasets on audio-visual speech classification, demonstrating best published visual speech classification on AVLetters and effective shared representation learning.",
"The problem of cross-modal retrieval from multimedia repositories is considered. This problem addresses the design of retrieval systems that support queries across content modalities, for example, using an image to search for texts. A mathematical formulation is proposed, equating the design of cross-modal retrieval systems to that of isomorphic feature spaces for different content modalities. Two hypotheses are then investigated regarding the fundamental attributes of these spaces. The first is that low-level cross-modal correlations should be accounted for. The second is that the space should enable semantic abstraction. Three new solutions to the cross-modal retrieval problem are then derived from these hypotheses: correlation matching (CM), an unsupervised method which models cross-modal correlations, semantic matching (SM), a supervised technique that relies on semantic representation, and semantic correlation matching (SCM), which combines both. An extensive evaluation of retrieval performance is conducted to test the validity of the hypotheses. All approaches are shown successful for text retrieval in response to image queries and vice versa. It is concluded that both hypotheses hold, in a complementary form, although evidence in favor of the abstraction hypothesis is stronger than that for correlation."
]
} |
1511.05659 | 2952663444 | Heterogeneous gap among different modalities emerges as one of the critical issues in modern AI problems. Unlike traditional uni-modal cases, where raw features are extracted and directly measured, the heterogeneous nature of cross modal tasks requires the intrinsic semantic representation to be compared in a unified framework. This paper studies the learning of different representations that can be retrieved across different modality contents. A novel approach for mining cross-modal representations is proposed by incorporating explicit linear semantic projecting in Hilbert space. The insight is that the discriminative structures of different modality data can be linearly represented in appropriate high dimension Hilbert spaces, where linear operations can be used to approximate nonlinear decisions in the original spaces. As a result, an efficient linear semantic down mapping is jointly learned for multimodal data, leading to a common space where they can be compared. The mechanism of "feature up-lifting and down-projecting" works seamlessly as a whole, which accomplishes crossmodal retrieval tasks very well. The proposed method, named as shared discriminative semantic representation learning (), is tested on two public multimodal dataset for both within- and inter- modal retrieval. The experiments demonstrate that it outperforms several state-of-the-art methods in most scenarios. | There are also a great amount of work focusing on hamming space as the semantic space. Multimodal hashing methods in this category includes Inter Media Hashing (IMH) @cite_9 , Latent Semantic Sparse Hashing (LSSH) @cite_18 , CMFH @cite_16 , CSLP @cite_11 , QCH @cite_24 and Regularized CrossModal Hashing (RCMH) @cite_22 etc. The IMH adopts a linear regression model with regularization to learn view-specific hash functions. The LSSH applies sparse coding and matrix factorization to learn the latent structure of image and text, then maps them to a joint hamming space. QCH takes both hashing function learning and quantization of hash codes into consideration. RCMH is an extension of Graph Regularized Hashing used in unimodal case. Despite of the efficiency on retrieval time and memory storage brought by hashing, directly projecting raw features into short binary codes is in some extent an over compressing'' idea, which may result severe reduction in retrieval accuracy. | {
"cite_N": [
"@cite_18",
"@cite_22",
"@cite_9",
"@cite_24",
"@cite_16",
"@cite_11"
],
"mid": [
"2086958058",
"",
"2049993534",
"2267050401",
"1976258951",
"1622965039"
],
"abstract": [
"Similarity search methods based on hashing for effective and efficient cross-modal retrieval on large-scale multimedia databases with massive text and images have attracted considerable attention. The core problem of cross-modal hashing is how to effectively construct correlation between multi-modal representations which are heterogeneous intrinsically in the process of hash function learning. Analogous to Canonical Correlation Analysis (CCA), most existing cross-modal hash methods embed the heterogeneous data into a joint abstraction space by linear projections. However, these methods fail to bridge the semantic gap more effectively, and capture high-level latent semantic information which has been proved that it can lead to better performance for image retrieval. To address these challenges, in this paper, we propose a novel Latent Semantic Sparse Hashing (LSSH) to perform cross-modal similarity search by employing Sparse Coding and Matrix Factorization. In particular, LSSH uses Sparse Coding to capture the salient structures of images, and Matrix Factorization to learn the latent concepts from text. Then the learned latent semantic features are mapped to a joint abstraction space. Moreover, an iterative strategy is applied to derive optimal solutions efficiently, and it helps LSSH to explore the correlation between multi-modal representations efficiently and automatically. Finally, the unified hashcodes are generated through the high level abstraction space by quantization. Extensive experiments on three different datasets highlight the advantage of our method under cross-modal scenarios and show that LSSH significantly outperforms several state-of-the-art methods.",
"",
"In this paper, we present a new multimedia retrieval paradigm to innovate large-scale search of heterogenous multimedia data. It is able to return results of different media types from heterogeneous data sources, e.g., using a query image to retrieve relevant text documents or images from different data sources. This utilizes the widely available data from different sources and caters for the current users' demand of receiving a result list simultaneously containing multiple types of data to obtain a comprehensive understanding of the query's results. To enable large-scale inter-media retrieval, we propose a novel inter-media hashing (IMH) model to explore the correlations among multiple media types from different data sources and tackle the scalability issue. To this end, multimedia data from heterogeneous data sources are transformed into a common Hamming space, in which fast search can be easily implemented by XOR and bit-count operations. Furthermore, we integrate a linear regression model to learn hashing functions so that the hash codes for new data points can be efficiently generated. Experiments conducted on real-world large-scale multimedia datasets demonstrate the superiority of our proposed method compared with state-of-the-art techniques.",
"Cross-modal hashing is designed to facilitate fast search across domains. In this work, we present a cross-modal hashing approach, called quantized correlation hashing (QCH), which takes into consideration the quantization loss over domains and the relation between domains. Unlike previous approaches that separate the optimization of the quantizer independent of maximization of domain correlation, our approach simultaneously optimizes both processes. The underlying relation between the domains that describes the same objects is established via maximizing the correlation between the hash codes across the domains. The resulting multi-modal objective function is transformed to a unimodal formalization, which is optimized through an alternative procedure. Experimental results on three real world datasets demonstrate that our approach outperforms the state-of-the-art multi-modal hashing methods.",
"Nearest neighbor search methods based on hashing have attracted considerable attention for effective and efficient large-scale similarity search in computer vision and information retrieval community. In this paper, we study the problems of learning hash functions in the context of multimodal data for cross-view similarity search. We put forward a novel hashing method, which is referred to Collective Matrix Factorization Hashing (CMFH). CMFH learns unified hash codes by collective matrix factorization with latent factor model from different modalities of one instance, which can not only supports cross-view search but also increases the search accuracy by merging multiple view information sources. We also prove that CMFH, a similarity-preserving hashing learning method, has upper and lower boundaries. Extensive experiments verify that CMFH significantly outperforms several state-of-the-art methods on three different datasets.",
"We present a probabilistic framework for learning pair-wise similarities between objects belonging to different modalities, such as drugs and proteins, or text and images. Our framework is based on learning a binary code based representation for objects in each modality, and has the following key properties: (i) it can leverage both pairwise as well as easy-to-obtain relative preference based cross-modal constraints, (ii) the probabilistic framework naturally allows querying for the most useful informative constraints, facilitating an active learning setting (existing methods for cross-modal similarity learning do not have such a mechanism), and (iii) the binary code length is learned from the data. We demonstrate the effectiveness of the proposed approach on two problems that require computing pairwise similarities between cross-modal object pairs: cross-modal link prediction in bipartite graphs, and hashing based cross-modal similarity search."
]
} |
1511.05659 | 2952663444 | Heterogeneous gap among different modalities emerges as one of the critical issues in modern AI problems. Unlike traditional uni-modal cases, where raw features are extracted and directly measured, the heterogeneous nature of cross modal tasks requires the intrinsic semantic representation to be compared in a unified framework. This paper studies the learning of different representations that can be retrieved across different modality contents. A novel approach for mining cross-modal representations is proposed by incorporating explicit linear semantic projecting in Hilbert space. The insight is that the discriminative structures of different modality data can be linearly represented in appropriate high dimension Hilbert spaces, where linear operations can be used to approximate nonlinear decisions in the original spaces. As a result, an efficient linear semantic down mapping is jointly learned for multimodal data, leading to a common space where they can be compared. The mechanism of "feature up-lifting and down-projecting" works seamlessly as a whole, which accomplishes crossmodal retrieval tasks very well. The proposed method, named as shared discriminative semantic representation learning (), is tested on two public multimodal dataset for both within- and inter- modal retrieval. The experiments demonstrate that it outperforms several state-of-the-art methods in most scenarios. | In order to account for both the retrieval efficiency and accuracy, supervised common latent semantic features are learned in many recently published literature. Typically, The CMSRM @cite_20 considers learning a latent cross-media representation from the perspective of a listwise ranking problem. The Multimodal Stacked AutoEncoders (MSAE) @cite_4 , constructs stacked auto-encoders (SAE) to project features of different modality into a common latent space. The Correspondence Autoencoder (CorrAE) @cite_0 incorporates representation learning and correlation learning into a single process. The Bilateral Image-Text Retrieval (BITR) @cite_5 addresses the problem of learning bilateral associations between visual and textual data based on structural SVM. The data driven semantic embedding proposed by @cite_17 is trained for image and text independently. It should be noted that some of these supervisory methods have complex learning process with much parameters to be tuned, or they are unable to conveniently access different modality independently in learned mutual space. We are of the opinion that a simple and effective method is more preferred. | {
"cite_N": [
"@cite_4",
"@cite_0",
"@cite_5",
"@cite_20",
"@cite_17"
],
"mid": [
"",
"1964073652",
"2044626341",
"2132551672",
"2035650150"
],
"abstract": [
"",
"The problem of cross-modal retrieval, e.g., using a text query to search for images and vice-versa, is considered in this paper. A novel model involving correspondence autoencoder (Corr-AE) is proposed here for solving this problem. The model is constructed by correlating hidden representations of two uni-modal autoencoders. A novel optimal objective, which minimizes a linear combination of representation learning errors for each modality and correlation learning error between hidden representations of two modalities, is used to train the model as a whole. Minimization of correlation learning error forces the model to learn hidden representations with only common information in different modalities, while minimization of representation learning error makes hidden representations are good enough to reconstruct input of each modality. A parameter @math is used to balance the representation learning error and the correlation learning error. Based on two different multi-modal autoencoders, Corr-AE is extended to other two correspondence models, here we called Corr-Cross-AE and Corr-Full-AE. The proposed models are evaluated on three publicly available data sets from real scenes. We demonstrate that the three correspondence autoencoders perform significantly better than three canonical correlation analysis based models and two popular multi-modal deep models on cross-modal retrieval tasks.",
"Building bilateral semantic associations between images and texts is among the fundamental problems in computer vision. In this paper, we study two complementary cross-modal prediction tasks: (i) predicting text(s) given an image (“Im2Text”), and (ii) predicting image(s) given a piece of text (“Text2Im”). We make no assumption on the specific form of text; i.e., it could be either a set of labels, phrases, or even captions. We pose both these tasks in a retrieval framework. For Im2Text, given a query image, our goal is to retrieve a ranked list of semantically relevant texts from an independent textcorpus (i.e., texts with no corresponding images). Similarly, for Text2Im, given a query text, we aim to retrieve a ranked list of semantically relevant images from a collection of unannotated images (i.e., images without any associated textual meta-data). We propose a novel Structural SVM based unified formulation for these two tasks. For both visual and textual data, two types of representations are investigated. These are based on: (1) unimodal probability distributions over topics learned using latent Dirichlet allocation, and (2) explicitly learned multi-modal correlations using canonical correlation analysis. Extensive experiments on three popular datasets (two medium and one web-scale) demonstrate that our framework gives promising results compared to existing models under various settings, thus confirming its efficacy for both the tasks.",
"In multimedia information retrieval, most classic approaches tend to represent different modalities of media in the same feature space. Existing approaches take either one-to-one paired data or uni-directional ranking examples (i.e., utilizing only text-query-image ranking examples or image-query-text ranking examples) as training examples, which do not make full use of bi-directional ranking examples (bi-directional ranking means that both text-query-image and image-query-text ranking examples are utilized in the training period) to achieve a better performance. In this paper, we consider learning a cross-media representation model from the perspective of optimizing a listwise ranking problem while taking advantage of bi-directional ranking examples. We propose a general cross-media ranking algorithm to optimize the bi-directional listwise ranking loss with a latent space embedding, which we call Bi-directional Cross-Media Semantic Representation Model (Bi-CMSRM). The latent space embedding is discriminatively learned by the structural large margin learning for optimization with certain ranking criteria (mean average precision in this paper) directly. We evaluate Bi-CMSRM on the Wikipedia and NUS-WIDE datasets and show that the utilization of the bi-directional ranking examples achieves a much better performance than only using the uni-directional ranking examples.",
"This paper proposes a data-driven approach for cross-media retrieval by automatically learning its underlying semantic vocabulary. Different from the existing semantic vocabularies, which are manually pre-defined and annotated, we automatically discover the vocabulary concepts and their annotations from multimedia collections. To this end, we apply a probabilistic topic model on the text available in the collection to extract its semantic structure. Moreover, we propose a learning to rank framework, to effectively learn the concept classifiers from the extracted annotations. We evaluate the discovered semantic vocabulary for cross-media retrieval on three datasets of image text and video text pairs. Our experiments demonstrate that the discovered vocabulary does not require any manual labeling to outperform three recent alternatives for cross-media retrieval."
]
} |
1511.05676 | 2266930373 | Visual Question Answering (VQA) emerges as one of the most fascinating topics in computer vision recently. Many state of the art methods naively use holistic visual features with language features into a Long Short-Term Memory (LSTM) module, neglecting the sophisticated interaction between them. This coarse modeling also blocks the possibilities of exploring finer-grained local features that contribute to the question answering dynamically over time. This paper addresses this fundamental problem by directly modeling the temporal dynamics between language and all possible local image patches. When traversing the question words sequentially, our end-to-end approach explicitly fuses the features associated to the words and the ones available at multiple local patches in an attention mechanism, and further combines the fused information to generate dynamic messages, which we call episode. We then feed the episodes to a standard question answering module together with the contextual visual information and linguistic information. Motivated by recent practices in deep learning, we use auxiliary loss functions during training to improve the performance. Our experiments on two latest public datasets suggest that our method has a superior performance. Notably, on the DARQUAR dataset we advanced the state of the art by 6 @math , and we also evaluated our approach on the most recent MSCOCO-VQA dataset. | Recurrent Neural Networks (RNN) have been used for modeling temporal sequences and gained attention in speech recognition @cite_2 , machine translation @cite_15 , image captioning @cite_17 @cite_13 . The recurrent connections are feedback loops in the unfolded network, and because of these connections, RNNs are suitable for modeling time series with strong nonlinear dynamics and long time correlations. The traditional RNN is hard to train due to the vanishing gradient problem, the weight updates computed via error backpropagation through time may become very small. | {
"cite_N": [
"@cite_15",
"@cite_13",
"@cite_17",
"@cite_2"
],
"mid": [
"2133564696",
"",
"2951183276",
"2950689855"
],
"abstract": [
"Neural machine translation is a recently proposed approach to machine translation. Unlike the traditional statistical machine translation, the neural machine translation aims at building a single neural network that can be jointly tuned to maximize the translation performance. The models proposed recently for neural machine translation often belong to a family of encoder-decoders and consists of an encoder that encodes a source sentence into a fixed-length vector from which a decoder generates a translation. In this paper, we conjecture that the use of a fixed-length vector is a bottleneck in improving the performance of this basic encoder-decoder architecture, and propose to extend this by allowing a model to automatically (soft-)search for parts of a source sentence that are relevant to predicting a target word, without having to form these parts as a hard segment explicitly. With this new approach, we achieve a translation performance comparable to the existing state-of-the-art phrase-based system on the task of English-to-French translation. Furthermore, qualitative analysis reveals that the (soft-)alignments found by the model agree well with our intuition.",
"",
"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.",
"Recurrent neural networks (RNNs) are a powerful model for sequential data. End-to-end training methods such as Connectionist Temporal Classification make it possible to train RNNs for sequence labelling problems where the input-output alignment is unknown. The combination of these methods with the Long Short-term Memory RNN architecture has proved particularly fruitful, delivering state-of-the-art results in cursive handwriting recognition. However RNN performance in speech recognition has so far been disappointing, with better results returned by deep feedforward networks. This paper investigates , which combine the multiple levels of representation that have proved so effective in deep networks with the flexible use of long range context that empowers RNNs. When trained end-to-end with suitable regularisation, we find that deep Long Short-term Memory RNNs achieve a test set error of 17.7 on the TIMIT phoneme recognition benchmark, which to our knowledge is the best recorded score."
]
} |
1511.05676 | 2266930373 | Visual Question Answering (VQA) emerges as one of the most fascinating topics in computer vision recently. Many state of the art methods naively use holistic visual features with language features into a Long Short-Term Memory (LSTM) module, neglecting the sophisticated interaction between them. This coarse modeling also blocks the possibilities of exploring finer-grained local features that contribute to the question answering dynamically over time. This paper addresses this fundamental problem by directly modeling the temporal dynamics between language and all possible local image patches. When traversing the question words sequentially, our end-to-end approach explicitly fuses the features associated to the words and the ones available at multiple local patches in an attention mechanism, and further combines the fused information to generate dynamic messages, which we call episode. We then feed the episodes to a standard question answering module together with the contextual visual information and linguistic information. Motivated by recent practices in deep learning, we use auxiliary loss functions during training to improve the performance. Our experiments on two latest public datasets suggest that our method has a superior performance. Notably, on the DARQUAR dataset we advanced the state of the art by 6 @math , and we also evaluated our approach on the most recent MSCOCO-VQA dataset. | Question answering (QA) is a classical problem in natural language processing @cite_1 . When images are involved, the goal of VQA is to infer the answer of a question directly from the image @cite_19 . Multiple questions and answers can be associated to the same image during training. | {
"cite_N": [
"@cite_19",
"@cite_1"
],
"mid": [
"2950761309",
"2131494463"
],
"abstract": [
"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).",
"Most tasks in natural language processing can be cast into question answering (QA) problems over language input. We introduce the dynamic memory network (DMN), a neural network architecture which processes input sequences and questions, forms episodic memories, and generates relevant answers. Questions trigger an iterative attention process which allows the model to condition its attention on the inputs and the result of previous iterations. These results are then reasoned over in a hierarchical recurrent sequence model to generate answers. The DMN can be trained end-to-end and obtains state-of-the-art results on several types of tasks and datasets: question answering (Facebook's bAbI dataset), text classification for sentiment analysis (Stanford Sentiment Treebank) and sequence modeling for part-of-speech tagging (WSJ-PTB). The training for these different tasks relies exclusively on trained word vector representations and input-question-answer triplets."
]
} |
1511.05676 | 2266930373 | Visual Question Answering (VQA) emerges as one of the most fascinating topics in computer vision recently. Many state of the art methods naively use holistic visual features with language features into a Long Short-Term Memory (LSTM) module, neglecting the sophisticated interaction between them. This coarse modeling also blocks the possibilities of exploring finer-grained local features that contribute to the question answering dynamically over time. This paper addresses this fundamental problem by directly modeling the temporal dynamics between language and all possible local image patches. When traversing the question words sequentially, our end-to-end approach explicitly fuses the features associated to the words and the ones available at multiple local patches in an attention mechanism, and further combines the fused information to generate dynamic messages, which we call episode. We then feed the episodes to a standard question answering module together with the contextual visual information and linguistic information. Motivated by recent practices in deep learning, we use auxiliary loss functions during training to improve the performance. Our experiments on two latest public datasets suggest that our method has a superior performance. Notably, on the DARQUAR dataset we advanced the state of the art by 6 @math , and we also evaluated our approach on the most recent MSCOCO-VQA dataset. | It has been shown that VQA can borrow the idea from image captioning. Being a related area, image captioning also uses RNN for sentence generation @cite_17 . Attention mechanism is recently adopted in image captioning and proves to be a useful component @cite_0 . | {
"cite_N": [
"@cite_0",
"@cite_17"
],
"mid": [
"2950178297",
"2951183276"
],
"abstract": [
"Inspired by recent work in machine translation and object detection, we introduce an attention based model that automatically learns to describe the content of images. We describe how we can train this model in a deterministic manner using standard backpropagation techniques and stochastically by maximizing a variational lower bound. We also show through visualization how the model is able to automatically learn to fix its gaze on salient objects while generating the corresponding words in the output sequence. We validate the use of attention with state-of-the-art performance on three benchmark datasets: Flickr8k, Flickr30k and MS COCO.",
"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."
]
} |
1511.05676 | 2266930373 | Visual Question Answering (VQA) emerges as one of the most fascinating topics in computer vision recently. Many state of the art methods naively use holistic visual features with language features into a Long Short-Term Memory (LSTM) module, neglecting the sophisticated interaction between them. This coarse modeling also blocks the possibilities of exploring finer-grained local features that contribute to the question answering dynamically over time. This paper addresses this fundamental problem by directly modeling the temporal dynamics between language and all possible local image patches. When traversing the question words sequentially, our end-to-end approach explicitly fuses the features associated to the words and the ones available at multiple local patches in an attention mechanism, and further combines the fused information to generate dynamic messages, which we call episode. We then feed the episodes to a standard question answering module together with the contextual visual information and linguistic information. Motivated by recent practices in deep learning, we use auxiliary loss functions during training to improve the performance. Our experiments on two latest public datasets suggest that our method has a superior performance. Notably, on the DARQUAR dataset we advanced the state of the art by 6 @math , and we also evaluated our approach on the most recent MSCOCO-VQA dataset. | LSTM for VQA Because a VQA system needs to process language and visual information simultaneously, most recent work adopted the LSTM in their approaches. A typical LSTM-VQA uses holistic image features extracted by CNNs as visual words'', as shown in Figure . The visual word features are used either as the first or at the end of question sequence @cite_4 or they are concatenated together with question word vectors into a LSTM @cite_3 @cite_7 . | {
"cite_N": [
"@cite_4",
"@cite_7",
"@cite_3"
],
"mid": [
"",
"1488163396",
"2952246170"
],
"abstract": [
"",
"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",
"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."
]
} |
1511.05240 | 2247326188 | We derive an extension of McDiarmid’s inequality for functions f with bounded differences on a high probability set Y (instead of almost surely). The behavior of f outside Y may be arbitrary. The proof is short and elementary, and relies on an extension argument similar to Kirszbraun’s theorem [4]. | While McDiarmid's inequality sometimes gives loose concentration bounds, its strength lies in its applicability (see @cite_7 for an extensive survey): sets @math may be completely arbitrary, and, even when @math is involved, it is usually easy to check that the bounded differences assumption holds. Two notable applications of McDiarmid's inequality are combinatorics and learning theory. Two representative results are the concentration of the chromatic number of Erd o s-R 'enyi graphs @cite_2 , and the fact that stable algorithms have good generalization performance @cite_11 . Namely, if the output of a learning algorithm does not vary too much when a training example is modified, then it performs well on an unseen, randomly selected, example. | {
"cite_N": [
"@cite_11",
"@cite_7",
"@cite_2"
],
"mid": [
"2110630246",
"",
"2082249316"
],
"abstract": [
"We present a novel way of obtaining PAC-style bounds on the generalization error of learning algorithms, explicitly using their stability properties. A stable learner is one for which the learned solution does not change much with small changes in the training set. The bounds we obtain do not depend on any measure of the complexity of the hypothesis space (e.g. VC dimension) but rather depend on how the learning algorithm searches this space, and can thus be applied even when the VC dimension is infinite. We demonstrate that regularization networks possess the required stability property and apply our method to obtain new bounds on their generalization performance.",
"",
"For a fixed probabilityp, 0<p<1, almost every random graphGn,p has chromatic number @math ,"
]
} |
1511.05240 | 2247326188 | We derive an extension of McDiarmid’s inequality for functions f with bounded differences on a high probability set Y (instead of almost surely). The behavior of f outside Y may be arbitrary. The proof is short and elementary, and relies on an extension argument similar to Kirszbraun’s theorem [4]. | Motivated by the study of random graphs, @cite_8 @cite_10 @cite_4 @cite_6 have provided concentration inequalities for particular classes of functions @math (e.g. polynomials) which have bounded differences with high probability. Polynomials are of interest for combinatorics since the number of subgraphs such as triangles or cliques can be written as a polynomial in the entries of the adjacency matrix. | {
"cite_N": [
"@cite_10",
"@cite_6",
"@cite_4",
"@cite_8"
],
"mid": [
"1873349594",
"2037289762",
"1985376393",
"2077204415"
],
"abstract": [
"In this work we design a general method for proving moment inequalities for polynomials of independent random variables. Our method works for a wide range of random variables including Gaussian, Boolean, exponential, Poisson and many others. We apply our method to derive general concentration inequalities for polynomials of independent random variables. We show that our method implies concentration inequalities for some previously open problems, e.g. permanent of random symmetric matrices. We show that our concentration inequality is stronger than the well-known concentration inequality due to Kim and Vu [29]. The main advantage of our method in comparison with the existing ones is a wide range of random variables we can handle and bounds for previously intractable regimes of high degree polynomials and small expectations. On the negative side we show that even for boolean random variables each term in our concentration inequality is tight.",
"Strong concentration results play a fundamental role in probabilistic combinatorics and theoretical computer science. In this paper, we present several new concentration results developed recently by the author and collaborators. To illustrate the power of these new results, we discuss applications in many different areas of mathematics, from combinatorial number theory to the theory of random graphs.",
"",
"Suppose t1, ..., tn are independent random variables which take value either 0 or 1, and Y is a multi-variable polynomial in ti’s with positive coefficients. We give a condition which guarantees that Y concentrates strongly around its mean even when several variables could have a large effect on Y . Some applications will be discussed. §"
]
} |
1511.05240 | 2247326188 | We derive an extension of McDiarmid’s inequality for functions f with bounded differences on a high probability set Y (instead of almost surely). The behavior of f outside Y may be arbitrary. The proof is short and elementary, and relies on an extension argument similar to Kirszbraun’s theorem [4]. | On the other hand, concentration inequalities for general functions whose differences are bounded with high probability were provided in @cite_5 , @cite_1 , @cite_9 . The authors assume that there exists vectors @math and @math , with @math for all @math such that function @math has @math -bounded differences on @math and @math -bounded differences on @math . The provided concentration inequalities usually give a strong improvement over McDiarmid's inequality, but are not informative if @math is too large. Theorem shows that this is an artefact, since all the required "information" about the behaviour of @math outside of @math is contained in @math . A toy example of this phenomenon is @math , @math Ber( @math ), @math , @math and f(X) = . For all @math , Theorem guarantees that @math while previously known inequalities become uninformative for @math arbitrarily large. | {
"cite_N": [
"@cite_5",
"@cite_9",
"@cite_1"
],
"mid": [
"154704495",
"2167969281",
"1615542443"
],
"abstract": [
"",
"Concentration inequalities are fundamental tools in probabilistic combinatorics and theoretical computer science for proving that random functions are near their means. Of particular importance is the case where f(X) is a function of independent random variables X=(X_1, ..., X_n). Here the well known bounded differences inequality (also called McDiarmid's or Hoeffding-Azuma inequality) establishes sharp concentration if the function f does not depend too much on any of the variables. One attractive feature is that it relies on a very simple Lipschitz condition (L): it suffices to show that |f(X)-f(X')| c_k whenever X,X' differ only in X_k. While this is easy to check, the main disadvantage is that it considers worst-case changes c_k, which often makes the resulting bounds too weak to be useful. In this paper we prove a variant of the bounded differences inequality which can be used to establish concentration of functions f(X) where (i) the typical changes are small although (ii) the worst case changes might be very large. One key aspect of this inequality is that it relies on a simple condition that (a) is easy to check and (b) coincides with heuristic considerations why concentration should hold. Indeed, given an event that holds with very high probability, we essentially relax the Lipschitz condition (L) to situations where occurs. The point is that the resulting typical changes c_k are often much smaller than the worst case ones. To illustrate its application we consider the reverse H-free process, where H is 2-balanced. We prove that the final number of edges in this process is concentrated, and also determine its likely value up to constant factors. This answers a question of Bollob 'as and Erd o s.",
"We prove an extension of McDiarmid's inequality for metric spaces with unbounded diameter. To this end, we introduce the notion of the subgaussian diameter, which is a distribution-dependent refinement of the metric diameter. Our technique provides an alternative approach to that of Kutin and Niyogi's method of weakly difference-bounded functions, and yields nontrivial, dimension-free results in some interesting cases where the former does not. As an application, we give apparently the first generalization bound in the algorithmic stability setting that holds for unbounded loss functions. This yields a novel risk bound for some regularized metric regression algorithms. We give two extensions of the basic concentration result. The first enables one to replace the independence assumption by appropriate strong mixing. The second generalizes the subgaussian technique to other Orlicz norms."
]
} |
1511.05296 | 2178857108 | In this paper, we propose a method for ranking fashion images to find the ones which might be liked by more people. We collect two new datasets from image sharing websites (Pinterest and Polyvore). We represent fashion images based on attributes: semantic attributes and data-driven attributes. To learn semantic attributes from limited training data, we use an algorithm on multi-task convolutional neural networks to share visual knowledge among different semantic attribute categories. To discover data-driven attributes unsupervisedly, we propose an algorithm to simultaneously discover visual clusters and learn fashion-specific feature representations. Given attributes as representations, we propose to learn a ranking SPN (sum product networks) to rank pairs of fashion images. The proposed ranking SPN can capture the high-order correlations of the attributes. We show the effectiveness of our method on our two newly collected datasets. | We develop our representations based on attributes learning. Many works show the importance of attributes as mid-level descriptors in visual recognition tasks @cite_28 @cite_39 @cite_2 @cite_47 @cite_19 @cite_44 @cite_43 @cite_38 . @cite_49 learn robust attribute classifier for object description by selecting informative features. For accurate attribute prediction, @cite_28 propose a method to encourage information sharing among the closely related attributes using structured @math sparsity regularization. @cite_19 propose a method for feature sharing between object recognition and attribute perdition. To learn the semantic attributes for clothing description, @cite_39 first extract low-level features based on the pose estimation result, then further improves the attribute prediction accuracy by modeling the mutual dependencies of the attributes with conditional random field. Instead of predicting the presence of an attribute, Parikh and Grauman @cite_2 propose relative attribute to model the strength of an attribute in an image with respect to other images. All of the previous methods learn attribute classifier based on low-level features @cite_28 @cite_39 @cite_2 @cite_47 @cite_19 . | {
"cite_N": [
"@cite_38",
"@cite_28",
"@cite_39",
"@cite_44",
"@cite_19",
"@cite_43",
"@cite_49",
"@cite_2",
"@cite_47"
],
"mid": [
"2147994354",
"2040171755",
"146395692",
"2013258928",
"2063386797",
"2010987212",
"2098411764",
"",
""
],
"abstract": [
"Attribute learning has attracted a lot of interests in recent years for its advantage of being able to model high-level concepts with a compact set of midlevel attributes. Real-world objects often demand multiple attributes for effective modeling. Most existing methods learn attributes independently without explicitly considering their intrinsic relatedness. In this paper, we propose max margin multiattribute learning with low-rank constraint, which learns a set of attributes simultaneously, using only relative ranking of the attributes for the data. By learning all the attributes simultaneously through low-rank constraint, the proposed method is able to capture their intrinsic correlation for improved learning; by requiring only relative ranking, the method avoids restrictive binary labels of attributes that are often assumed by many existing techniques. The proposed method is evaluated on both synthetic data and real visual data including a challenging video data set. Experimental results demonstrate the effectiveness of the proposed method.",
"Existing methods to learn visual attributes are prone to learning the wrong thing -- namely, properties that are correlated with the attribute of interest among training samples. Yet, many proposed applications of attributes rely on being able to learn the correct semantic concept corresponding to each attribute. We propose to resolve such confusions by jointly learning decorrelated, discriminative attribute models. Leveraging side information about semantic relatedness, we develop a multi-task learning approach that uses structured sparsity to encourage feature competition among unrelated attributes and feature sharing among related attributes. On three challenging datasets, we show that accounting for structure in the visual attribute space is key to learning attribute models that preserve semantics, yielding improved generalizability that helps in the recognition and discovery of unseen object categories.",
"Describing clothing appearance with semantic attributes is an appealing technique for many important applications. In this paper, we propose a fully automated system that is capable of generating a list of nameable attributes for clothes on human body in unconstrained images. We extract low-level features in a pose-adaptive manner, and combine complementary features for learning attribute classifiers. Mutual dependencies between the attributes are then explored by a Conditional Random Field to further improve the predictions from independent classifiers. We validate the performance of our system on a challenging clothing attribute dataset, and introduce a novel application of dressing style analysis that utilizes the semantic attributes produced by our system.",
"In this paper, we utilize structured learning to simultaneously address two intertwined problems: 1) human pose estimation (HPE) and 2) garment attribute classification (GAC), which are valuable for a variety of computer vision and multimedia applications. Unlike previous works that usually handle the two problems separately, our approach aims to produce an optimal joint estimation for both HPE and GAC via a unified inference procedure. To this end, we adopt a preprocessing step to detect potential human parts from each image (i.e., a set of candidates) that allows us to have a manageable input space. In this way, the simultaneous inference of HPE and GAC is converted to a structured learning problem, where the inputs are the collections of candidate ensembles, outputs are the joint labels of human parts and garment attributes, and joint feature representation involves various cues such as pose-specific features, garment-specific features, and cross-task features that encode correlations between human parts and garment attributes. Furthermore, we explore the strong edge evidence around the potential human parts so as to derive more powerful representations for oriented human parts. Such evidences can be seamlessly integrated into our structured learning model as a kind of energy function, and the learning process could be performed by standard structured support vector machines algorithm. However, the joint structure of the two problems is a cyclic graph, which hinders efficient inference. To resolve this issue, we compute instead approximate optima using an iterative procedure, where in each iteration, the variables of one problem are fixed. In this way, satisfactory solutions can be efficiently computed by dynamic programming. Experimental results on two benchmark data sets show the state-of-the-art performance of our approach.",
"Visual attributes expose human-defined semantics to object recognition models, but existing work largely restricts their influence to mid-level cues during classifier training. Rather than treat attributes as intermediate features, we consider how learning visual properties in concert with object categories can regularize the models for both. Given a low-level visual feature space together with attribute-and object-labeled image data, we learn a shared lower-dimensional representation by optimizing a joint loss function that favors common sparsity patterns across both types of prediction tasks. We adopt a recent kernelized formulation of convex multi-task feature learning, in which one alternates between learning the common features and learning task-specific classifier parameters on top of those features. In this way, our approach discovers any structure among the image descriptors that is relevant to both tasks, and allows the top-down semantics to restrict the hypothesis space of the ultimate object classifiers. We validate the approach on datasets of animals and outdoor scenes, and show significant improvements over traditional multi-class object classifiers and direct attribute prediction models.",
"Semantic attributes have been recognized as a more spontaneous manner to describe and annotate image content. It is widely accepted that image annotation using semantic attributes is a significant improvement to the traditional binary or multiclass annotation due to its naturally continuous and relative properties. Though useful, existing approaches rely on an abundant supervision and high-quality training data, which limit their applicability. Two standard methods to overcome small amounts of guidance and low-quality training data are transfer and active learning. In the context of relative attributes, this would entail learning multiple relative attributes simultaneously and actively querying a human for additional information. This paper addresses the two main limitations in existing work: 1) it actively adds humans to the learning loop so that minimal additional guidance can be given and 2) it learns multiple relative attributes simultaneously and thereby leverages dependence amongst them. In this paper, we formulate a joint active learning to rank framework with pairwise supervision to achieve these two aims, which also has other benefits such as the ability to be kernelized. The proposed framework optimizes over a set of ranking functions (measuring the strength of the presence of attributes) simultaneously and dependently on each other. The proposed pairwise queries take the form of which one of these two pictures is more natural? These queries can be easily answered by humans. Extensive empirical study on real image data sets shows that our proposed method, compared with several state-of-the-art methods, achieves superior retrieval performance while requires significantly less human inputs.",
"We propose to shift the goal of recognition from naming to describing. Doing so allows us not only to name familiar objects, but also: to report unusual aspects of a familiar object (“spotty dog”, not just “dog”); to say something about unfamiliar objects (“hairy and four-legged”, not just “unknown”); and to learn how to recognize new objects with few or no visual examples. Rather than focusing on identity assignment, we make inferring attributes the core problem of recognition. These attributes can be semantic (“spotty”) or discriminative (“dogs have it but sheep do not”). Learning attributes presents a major new challenge: generalization across object categories, not just across instances within a category. In this paper, we also introduce a novel feature selection method for learning attributes that generalize well across categories. We support our claims by thorough evaluation that provides insights into the limitations of the standard recognition paradigm of naming and demonstrates the new abilities provided by our attribute-based framework.",
"",
""
]
} |
1511.05296 | 2178857108 | In this paper, we propose a method for ranking fashion images to find the ones which might be liked by more people. We collect two new datasets from image sharing websites (Pinterest and Polyvore). We represent fashion images based on attributes: semantic attributes and data-driven attributes. To learn semantic attributes from limited training data, we use an algorithm on multi-task convolutional neural networks to share visual knowledge among different semantic attribute categories. To discover data-driven attributes unsupervisedly, we propose an algorithm to simultaneously discover visual clusters and learn fashion-specific feature representations. Given attributes as representations, we propose to learn a ranking SPN (sum product networks) to rank pairs of fashion images. The proposed ranking SPN can capture the high-order correlations of the attributes. We show the effectiveness of our method on our two newly collected datasets. | Differently, we use deep features extracted by CNN models to learn attributes: both nameable and hidden (data-driven discoverable). The deep structure of CNN can produce robust feature generalization ability in different computer vision tasks @cite_37 @cite_50 @cite_25 @cite_29 @cite_14 . In attribute learning, we usually don't have enough training data. To tackle this problem, some vision papers use multi-task learning to effectively share visual knowledge @cite_28 @cite_27 @cite_9 @cite_3 . Different from their works, we integrate multi-task learning with the powerful CNN models to learn semantic attributes. We jointly learn feature representations and visual knowledge sharing at the same time. We use an enhanced multi-task sharing strategy, where different attribute categories can adaptively select to share from intra-group members, and compete inter-group members, through a flexible decomposition of a shared hidden latent task layer and linear combination layers. In addition, we unsupervisedly train CNN models to discover data-diven attributes. These attributes are important to provide complementary informations beside semantic ones. | {
"cite_N": [
"@cite_37",
"@cite_14",
"@cite_28",
"@cite_29",
"@cite_9",
"@cite_3",
"@cite_27",
"@cite_50",
"@cite_25"
],
"mid": [
"2163605009",
"1897260080",
"2040171755",
"",
"2156392723",
"1994762580",
"",
"",
""
],
"abstract": [
"We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 different classes. On the test data, we achieved top-1 and top-5 error rates of 37.5 and 17.0 which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implementation of the convolution operation. To reduce overriding in the fully-connected layers we employed a recently-developed regularization method called \"dropout\" that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3 , compared to 26.2 achieved by the second-best entry.",
"We adopt Convolutional Neural Networks (CNN) as our parametric model to learn discriminative features and classifiers for local patch classification. As visually similar pixels are indistinguishable from local context, we alleviate such ambiguity by introducing a global scene constraint. We estimate the global potential in a non-parametric framework. Furthermore, a large margin based CNN metric learning method is proposed for better global potential estimation. The final pixel class prediction is performed by integrating local and global beliefs. Even without any post-processing, we achieve state-of-the-art performance on SiftFlow and competitive results on Stanford Background benchmark.",
"Existing methods to learn visual attributes are prone to learning the wrong thing -- namely, properties that are correlated with the attribute of interest among training samples. Yet, many proposed applications of attributes rely on being able to learn the correct semantic concept corresponding to each attribute. We propose to resolve such confusions by jointly learning decorrelated, discriminative attribute models. Leveraging side information about semantic relatedness, we develop a multi-task learning approach that uses structured sparsity to encourage feature competition among unrelated attributes and feature sharing among related attributes. On three challenging datasets, we show that accounting for structure in the visual attribute space is key to learning attribute models that preserve semantics, yielding improved generalizability that helps in the recognition and discovery of unseen object categories.",
"",
"Sharing knowledge for multiple related machine learning tasks is an effective strategy to improve the generalization performance. In this paper, we investigate knowledge sharing across categories for action recognition in videos. The motivation is that many action categories are related, where common motion pattern are shared among them (e.g. diving and high jump share the jump motion). We propose a new multi-task learning method to learn latent tasks shared across categories, and reconstruct a classifier for each category from these latent tasks. Compared to previous methods, our approach has two advantages: (1) The learned latent tasks correspond to basic motion patterns instead of full actions, thus enhancing discrimination power of the classifiers. (2) Categories are selected to share information with a sparsity regularizer, avoiding falsely forcing all categories to share knowledge. Experimental results on multiple public data sets show that the proposed approach can effectively transfer knowledge between different action categories to improve the performance of conventional single task learning methods.",
"Model-free object tracking is still challenging because of the limited prior knowledge and the unexpected variation of the target object. In this paper, we propose a feature learning algorithm for model-free multiple object tracking. First, we pre-learn generic features invariant to diverse motion transformations from auxiliary video data by using a deep network of anto-encoder. Then, we adapt the pre-learned features according to multiple target objects respectively in a multi-task learning manner. We treat the feature adaptation for each target object as one single task. We simultaneously learn the common feature shared by all target objects and the individual feature of each object. Experimental results demonstrate that our feature learning algorithm can significantly improve multiple object tracking performance.",
"",
"",
""
]
} |
1511.05296 | 2178857108 | In this paper, we propose a method for ranking fashion images to find the ones which might be liked by more people. We collect two new datasets from image sharing websites (Pinterest and Polyvore). We represent fashion images based on attributes: semantic attributes and data-driven attributes. To learn semantic attributes from limited training data, we use an algorithm on multi-task convolutional neural networks to share visual knowledge among different semantic attribute categories. To discover data-driven attributes unsupervisedly, we propose an algorithm to simultaneously discover visual clusters and learn fashion-specific feature representations. Given attributes as representations, we propose to learn a ranking SPN (sum product networks) to rank pairs of fashion images. The proposed ranking SPN can capture the high-order correlations of the attributes. We show the effectiveness of our method on our two newly collected datasets. | Based on these learned representations, we exploit the nature of SPN to model the high-order correlations between attribute representations. SPN is first introduced in @cite_4 as a deep probabilistic model, that can capture the deep relationships. The theoretical works @cite_23 @cite_17 @cite_20 motivate the research of SPN. Delalleau and Bengio @cite_17 prove that deep SPN is very efficiently in representing some functions. Gens and Domingos @cite_23 propose an algorithm for SPN structure learning. Rooshenas and Lowd @cite_20 propose another method for learning SPN structures that can capture both indirect and direct interactions between variables. SPN has been successfully applied in computer vision tasks, including image classification @cite_24 , image completing @cite_4 , and facial attribute analysis @cite_36 . For the first time, we extend SPN to be a ranking machine by modeling the high-order correlations of its leaf nodes. We also propose a method to learn the parameters of SPN based on pairs of images. | {
"cite_N": [
"@cite_4",
"@cite_36",
"@cite_24",
"@cite_23",
"@cite_20",
"@cite_17"
],
"mid": [
"2949869425",
"2143352446",
"2128744540",
"2159687189",
"2119900738",
"2141473882"
],
"abstract": [
"The key limiting factor in graphical model inference and learning is the complexity of the partition function. We thus ask the question: what are general conditions under which the partition function is tractable? The answer leads to a new kind of deep architecture, which we call sum-product networks (SPNs). SPNs are directed acyclic graphs with variables as leaves, sums and products as internal nodes, and weighted edges. We show that if an SPN is complete and consistent it represents the partition function and all marginals of some graphical model, and give semantics to its nodes. Essentially all tractable graphical models can be cast as SPNs, but SPNs are also strictly more general. We then propose learning algorithms for SPNs, based on backpropagation and EM. Experiments show that inference and learning with SPNs can be both faster and more accurate than with standard deep networks. For example, SPNs perform image completion better than state-of-the-art deep networks for this task. SPNs also have intriguing potential connections to the architecture of the cortex.",
"Recent works have shown that facial attributes are useful in a number of applications such as face recognition and retrieval. However, estimating attributes in images with large variations remains a big challenge. This challenge is addressed in this paper. Unlike existing methods that assume the independence of attributes during their estimation, our approach captures the interdependencies of local regions for each attribute, as well as the high-order correlations between different attributes, which makes it more robust to occlusions and misdetection of face regions. First, we have modeled region interdependencies with a discriminative decision tree, where each node consists of a detector and a classifier trained on a local region. The detector allows us to locate the region, while the classifier determines the presence or absence of an attribute. Second, correlations of attributes and attribute predictors are modeled by organizing all of the decision trees into a large sum-product network (SPN), which is learned by the EM algorithm and yields the most probable explanation (MPE) of the facial attributes in terms of the region's localization and classification. Experimental results on a large data set with 22,400 images show the effectiveness of the proposed approach.",
"Sum-product networks are a new deep architecture that can perform fast, exact inference on high-treewidth models. Only generative methods for training SPNs have been proposed to date. In this paper, we present the first discriminative training algorithms for SPNs, combining the high accuracy of the former with the representational power and tractability of the latter. We show that the class of tractable discriminative SPNs is broader than the class of tractable generative ones, and propose an efficient backpropagation-style algorithm for computing the gradient of the conditional log likelihood. Standard gradient descent suffers from the diffusion problem, but networks with many layers can be learned reliably using \"hard\" gradient descent, where marginal inference is replaced by MPE inference (i.e., inferring the most probable state of the non-evidence variables). The resulting updates have a simple and intuitive form. We test discriminative SPNs on standard image classification tasks. We obtain the best results to date on the CIFAR-10 dataset, using fewer features than prior methods with an SPN architecture that learns local image structure discriminatively. We also report the highest published test accuracy on STL-10 even though we only use the labeled portion of the dataset.",
"Sum-product networks (SPNs) are a new class of deep probabilistic models. SPNs can have unbounded treewidth but inference in them is always tractable. An SPN is either a univariate distribution, a product of SPNs over disjoint variables, or a weighted sum of SPNs over the same variables. We propose the first algorithm for learning the structure of SPNs that takes full advantage of their expressiveness. At each step, the algorithm attempts to divide the current variables into approximately independent subsets. If successful, it returns the product of recursive calls on the subsets; otherwise it returns the sum of recursive calls on subsets of similar instances from the current training set. A comprehensive empirical study shows that the learned SPNs are typically comparable to graphical models in likelihood but superior in inference speed and accuracy.",
"Sum-product networks (SPNs) are a deep probabilistic representation that allows for efficient, exact inference. SPNs generalize many other tractable models, including thin junction trees, latent tree models, and many types of mixtures. Previous work on learning SPN structure has mainly focused on using top-down or bottom-up clustering to find mixtures, which capture variable interactions indirectly through implicit latent variables. In contrast, most work on learning graphical models, thin junction trees, and arithmetic circuits has focused on finding direct interactions among variables. In this paper, we present ID-SPN, a new algorithm for learning SPN structure that unifies the two approaches. In experiments on 20 benchmark datasets, we find that the combination of direct and indirect interactions leads to significantly better accuracy than several state-of-the-art algorithms for learning SPNs and other tractable models.",
"We investigate the representational power of sum-product networks (computation networks analogous to neural networks, but whose individual units compute either products or weighted sums), through a theoretical analysis that compares deep (multiple hidden layers) vs. shallow (one hidden layer) architectures. We prove there exist families of functions that can be represented much more efficiently with a deep network than with a shallow one, i.e. with substantially fewer hidden units. Such results were not available until now, and contribute to motivate recent research involving learning of deep sum-product networks, and more generally motivate research in Deep Learning."
]
} |
1511.05296 | 2178857108 | In this paper, we propose a method for ranking fashion images to find the ones which might be liked by more people. We collect two new datasets from image sharing websites (Pinterest and Polyvore). We represent fashion images based on attributes: semantic attributes and data-driven attributes. To learn semantic attributes from limited training data, we use an algorithm on multi-task convolutional neural networks to share visual knowledge among different semantic attribute categories. To discover data-driven attributes unsupervisedly, we propose an algorithm to simultaneously discover visual clusters and learn fashion-specific feature representations. Given attributes as representations, we propose to learn a ranking SPN (sum product networks) to rank pairs of fashion images. The proposed ranking SPN can capture the high-order correlations of the attributes. We show the effectiveness of our method on our two newly collected datasets. | Our work is related to a number of existing high-level image understanding papers, including fashion modeling @cite_16 , interestingness prediction @cite_6 , image importance prediction @cite_30 and image memorability @cite_32 . But different from them, we are focusing on ranking fashion images, which is considered as an important task for on-line shopping and social websites. | {
"cite_N": [
"@cite_30",
"@cite_16",
"@cite_32",
"@cite_6"
],
"mid": [
"2168117362",
"",
"2158155099",
"2080754665"
],
"abstract": [
"Photos are becoming prominent means of communication online. Despite photos' pervasive presence in social media and online world, we know little about how people interact and engage with their content. Understanding how photo content might signify engagement, can impact both science and design, influencing production and distribution. One common type of photo content that is shared on social media, is the photos of people. From studies of offline behavior, we know that human faces are powerful channels of non-verbal communication. In this paper, we study this behavioral phenomena online. We ask how presence of a face, it's age and gender might impact social engagement on the photo. We use a corpus of 1 million Instagram images and organize our study around two social engagement feedback factors, likes and comments. Our results show that photos with faces are 38 more likely to receive likes and 32 more likely to receive comments, even after controlling for social network reach and activity. We find, however, that the number of faces, their age and gender do not have an effect. This work presents the first results on how photos with human faces relate to engagement on large scale image sharing communities. In addition to contributing to the research around online user behavior, our findings offer a new line of future work using visual analysis.",
"",
"Artists, advertisers, and photographers are routinely presented with the task of creating an image that a viewer will remember. While it may seem like image memorability is purely subjective, recent work shows that it is not an inexplicable phenomenon: variation in memorability of images is consistent across subjects, suggesting that some images are intrinsically more memorable than others, independent of a subjects' contexts and biases. In this paper, we used the publicly available memorability dataset of [13], and augmented the object and scene annotations with interpretable spatial, content, and aesthetic image properties. We used a feature-selection scheme with desirable explaining-away properties to determine a compact set of attributes that characterizes the memorability of any individual image. We find that images of enclosed spaces containing people with visible faces are memorable, while images of vistas and peaceful scenes are not. Contrary to popular belief, unusual or aesthetically pleasing scenes do not tend to be highly memorable. This work represents one of the first attempts at understanding intrinsic image memorability, and opens a new domain of investigation at the interface between human cognition and computer vision.",
"With the rise in popularity of digital cameras, the amount of visual data available on the web is growing exponentially. Some of these pictures are extremely beautiful and aesthetically pleasing, but the vast majority are uninteresting or of low quality. This paper demonstrates a simple, yet powerful method to automatically select high aesthetic quality images from large image collections. Our aesthetic quality estimation method explicitly predicts some of the possible image cues that a human might use to evaluate an image and then uses them in a discriminative approach. These cues or high level describable image attributes fall into three broad types: 1) compositional attributes related to image layout or configuration, 2) content attributes related to the objects or scene types depicted, and 3) sky-illumination attributes related to the natural lighting conditions. We demonstrate that an aesthetics classifier trained on these describable attributes can provide a significant improvement over baseline methods for predicting human quality judgments. We also demonstrate our method for predicting the “interestingness” of Flickr photos, and introduce a novel problem of estimating query specific “interestingness”."
]
} |
1511.05110 | 2179091859 | We consider a population structured by a spacevariable and a phenotypical trait, submitted to dispersion,mutations, growth and nonlocal competition. We introduce theclimate shift due to Global Warming and discuss the dynamicsof the population by studying the long time behavior of thesolution of the Cauchy problem. We consider three sets ofassumptions on the growth function. In the so-called confinedcase we determine a critical climate change speed for theextinction or survival of the population, the latter case taking place by "strictly following the climate shift". In the so-called gradient case , or unconfined case , we additionally determine the propagation speedof the population when it survives: thanks to a combination of migration and evolution, it can here be different from the speed of the climate shift. Finally, we consider mixed scenarios , that are complex situations, where thegrowth function satisfies the conditions of the confined case on the right, and the conditions of the unconfined case on the left.The main difficulty comes from the nonlocal competition term that prevents the use of classical methods based on comparison arguments. This difficulty is overcome thanks to estimates on the tails of the solution, and a careful application of the parabolic Harnack inequality. | The main difficulty in the mathematical analysis of is to handle the nonlocal competition term. When the competition term is replaced by a local (in @math and @math ) density regulation, many techniques based on the comparison principle --- such as some monotone iterative schemes or the sliding method --- can be used to get, among other things, monotonicity properties of the solution. Since integro-differential equations with a nonlocal competition term do not satisfy the comparison principle, it is unlikely that such techniques apply here. Problem shares this difficulty with the nonlocal Fisher-KPP equation which describes a population structured by a spatial variable only, and submitted to nonlocal competition modelized by the kernel @math . As far as equation is concerned, let us mention the possible destabilization of the steady state @math by some kernels @cite_17 , the construction of travelling waves @cite_12 , additional properties of these waves @cite_2 , @cite_1 , and a spreading speed result @cite_6 . We also refer to @cite_15 , @cite_9 for the construction of travelling waves for a bistable nonlocal equation, for an epidemiological system with mutations respectively. | {
"cite_N": [
"@cite_9",
"@cite_1",
"@cite_6",
"@cite_2",
"@cite_15",
"@cite_12",
"@cite_17"
],
"mid": [
"1609258433",
"2010503314",
"1967904114",
"2093569619",
"",
"1970661866",
"2130513524"
],
"abstract": [
"In this article, we are interested in a non-monotone system of logistic reaction-diffusion equations. This system of equations models an epidemics where two types of pathogens are competing, and a mutation can change one type into the other with a certain rate. We show the existence of minimal speed travelling waves, that are usually non monotonic. We then provide a description of the shape of those constructed travelling waves, and relate them to some Fisher-KPP fronts with non-minimal speed.",
"Abstract In this note, we give a positive answer to a question addressed in (2011) [7] . To be precise, we prove that, for any kernel and any slope at the origin, there exist traveling wave solutions (actually those which are “rapid”) of the nonlocal Fisher equation that connect the two homogeneous steady states 0 (dynamically unstable) and 1. In particular, this allows situations where 1 is unstable in the sense of Turing. Our proof does not involve any maximum principle argument and applies to kernels with fat tails .",
"We consider the Fisher–KPP (for Kolmogorov–Petrovsky–Piskunov) equation with a nonlocal interaction term. We establish a condition on the interaction that allows for existence of non-constant periodic solutions, and prove uniform upper bounds for the solutions of the Cauchy problem, as well as upper and lower bounds on the spreading rate of the solutions with compactly supported initial data.",
"A sufficient and necessary condition for the existence of monotone travelling waves in the nonlocal Fisher–KPP equation is established, and the uniqueness of travelling wavefronts (up to translation) is also proved.",
"",
"We consider the Fisher–KPP equation with a non-local saturation effect defined through an interaction kernel (x) and investigate the possible differences with the standard Fisher–KPP equation. Our first concern is the existence of steady states. We prove that if the Fourier transform is positive or if the length σ of the non-local interaction is short enough, then the only steady states are u ≡ 0 and u ≡ 1. Next, we study existence of the travelling waves. We prove that this equation admits travelling wave solutions that connect u = 0 to an unknown positive steady state u∞(x), for all speeds c ≥ c*. The travelling wave connects to the standard state u∞(x) ≡ 1 under the aforementioned conditions: 0 SRC=http: ej.iop.org images 0951-7715 22 12 002 non313053in002.gif > or σ is sufficiently small. However, the wave is not monotonic for σ large.",
"We study a reaction-diffusion equation with an integral term describing nonlocal consumption of resources in population dynamics. We show that a homogeneous equilibrium can lose its stability resulting in appearance of stationary spatial structures. They can be related to the emergence of biological species due to the intra-specific competition and random mutations. Various types of travelling waves are observed."
]
} |
1511.05202 | 2257177216 | Adequate evaluation of an information retrieval system to estimate future performance is a crucial task. Area under the ROC curve (AUC) is widely used to evaluate the generalization of a retrieval system. However, the objective function optimized in many retrieval systems is the error rate and not the AUC value. This paper provides an efficient and effective non-linear approach to optimize AUC using additive regression trees, with a special emphasis on the use of multi-class AUC (MAUC) because multiple relevance levels are widely used in many ranking applications. Compared to a conventional linear approach, the performance of the non-linear approach is comparable on binary-relevance benchmark datasets and is better on multi-relevance benchmark datasets. | This work relates to two areas: LambdaMART and AUC optimization in ranking. LambdaMART was originally proposed in @cite_11 and is overviewed in @cite_0 . The LambdaRank algorithm, upon which LambdaMART is based, was shown to find a locally optimal model for the IR metrics NDCG@10, mean NDCG, MAP, and MRR @cite_2 . Svore @cite_1 propose a modification to LambdaMART that allows for simultaneous optimization of NDCG and a measure based on click-through rate. | {
"cite_N": [
"@cite_0",
"@cite_1",
"@cite_2",
"@cite_11"
],
"mid": [
"2115584760",
"2145664829",
"22063930",
"2162059449"
],
"abstract": [
"LambdaMART is the boosted tree version of LambdaRank, which is based on RankNet. RankNet, LambdaRank, and LambdaMART have proven to be very successful algorithms for solving real world ranking problems: for example an ensemble of LambdaMART rankers won Track 1 of the 2010 Yahoo! Learning To Rank Challenge. The details of these algorithms are spread across several papers and reports, and so here we give a self-contained, detailed and complete description of them.",
"We investigate the problem of learning to rank with document retrieval from the perspective of learning for multiple objective functions. We present solutions to two open problems in learning to rank: first, we show how multiple measures can be combined into a single graded measure that can be learned. This solves the problem of learning from a 'scorecard' of measures by making such scorecards comparable, and we show results where a standard web relevance measure (NDCG) is used for the top-tier measure, and a relevance measure derived from click data is used for the second-tier measure; the second-tier measure is shown to significantly improve while leaving the top-tier measure largely unchanged. Second, we note that the learning-to-rank problem can itself be viewed as changing as the ranking model learns: for example, early in learning, adjusting the rank of all documents can be advantageous, but later during training, it becomes more desirable to concentrate on correcting the top few documents for each query. We show how an analysis of these problems leads to an improved, iteration-dependent cost function that interpolates between a cost function that is more appropriate for early learning, with one that is more appropriate for late-stage learning. The approach results in a significant improvement in accuracy with the same size models. We investigate these ideas using LambdaMART, a state-of-the-art ranking algorithm.",
"",
"We present a new ranking algorithm that combines the strengths of two previous methods: boosted tree classification, and LambdaRank, which has been shown to be empirically optimal for a widely used information retrieval measure. Our algorithm is based on boosted regression trees, although the ideas apply to any weak learners, and it is significantly faster in both train and test phases than the state of the art, for comparable accuracy. We also show how to find the optimal linear combination for any two rankers, and we use this method to solve the line search problem exactly during boosting. In addition, we show that starting with a previously trained model, and boosting using its residuals, furnishes an effective technique for model adaptation, and we give significantly improved results for a particularly pressing problem in web search--training rankers for markets for which only small amounts of labeled data are available, given a ranker trained on much more data from a larger market."
]
} |
1511.05202 | 2257177216 | Adequate evaluation of an information retrieval system to estimate future performance is a crucial task. Area under the ROC curve (AUC) is widely used to evaluate the generalization of a retrieval system. However, the objective function optimized in many retrieval systems is the error rate and not the AUC value. This paper provides an efficient and effective non-linear approach to optimize AUC using additive regression trees, with a special emphasis on the use of multi-class AUC (MAUC) because multiple relevance levels are widely used in many ranking applications. Compared to a conventional linear approach, the performance of the non-linear approach is comparable on binary-relevance benchmark datasets and is better on multi-relevance benchmark datasets. | Various approaches have been developed for optimizing AUC in binary-class settings. Cortes and Mohri @cite_5 show that minimum error rate training may be insufficient for optimizing AUC, and demonstrate that the RankBoost algorithm globally optimizes AUC. Calders and Jaroszewicz @cite_8 propose a smooth polynomial approximation of AUC that can be optimized with a gradient descent method. Joachims @cite_4 proposes an SVM method for various IR measures including AUC, and evaluates the system on text classification datasets. The SVM method is used as the comparison baseline in this paper. | {
"cite_N": [
"@cite_5",
"@cite_4",
"@cite_8"
],
"mid": [
"2120100126",
"2070771761",
"2165880761"
],
"abstract": [
"The area under an ROC curve (AUC) is a criterion used in many applications to measure the quality of a classification algorithm. However, the objective function optimized in most of these algorithms is the error rate and not the AUC value. We give a detailed statistical analysis of the relationship between the AUC and the error rate, including the first exact expression of the expected value and the variance of the AUC for a fixed error rate. Our results show that the average AUC is monotonically increasing as a function of the classification accuracy, but that the standard deviation for uneven distributions and higher error rates is noticeable. Thus, algorithms designed to minimize the error rate may not lead to the best possible AUC values. We show that, under certain conditions, the global function optimized by the RankBoost algorithm is exactly the AUC. We report the results of our experiments with RankBoost in several datasets demonstrating the benefits of an algorithm specifically designed to globally optimize the AUC over other existing algorithms optimizing an approximation of the AUC or only locally optimizing the AUC.",
"This paper presents a Support Vector Method for optimizing multivariate nonlinear performance measures like the F 1 -score. Taking a multivariate prediction approach, we give an algorithm with which such multivariate SVMs can be trained in polynomial time for large classes of potentially non-linear performance measures, in particular ROCArea and all measures that can be computed from the contingency table. The conventional classification SVM arises as a special case of our method.",
"In this paper we show an efficient method for inducing classifiers that directly optimize the area under the ROC curve. Recently, AUC gained importance in the classification community as a mean to compare the performance of classifiers. Because most classification methods do not optimize this measure directly, several classification learning methods are emerging that directly optimize the AUC. These methods, however, require many costly computations of the AUC, and hence, do not scale well to large datasets. In this paper, we develop a method to increase the efficiency of computing AUC based on a polynomial approximation of the AUC. As a proof of concept, the approximation is plugged into the construction of a scalable linear classifier that directly optimizes AUC using a gradient descent method. Experiments on real-life datasets show a high accuracy and efficiency of the polynomial approximation."
]
} |
1511.05045 | 2256809320 | Image and video classification research has made great progress through the development of handcrafted local features and learning based features. These two architectures were proposed roughly at the same time and have flourished at overlapping stages of history. However, they are typically viewed as distinct approaches. In this paper, we emphasize their structural similarities and show how such a unified view helps us in designing features that balance efficiency and effectiveness. As an example, we study the problem of designing efficient video feature learning algorithms for action recognition. We approach this problem by first showing that local handcrafted features and Convolutional Neural Networks (CNNs) share the same convolution-pooling network structure. We then propose a two-stream Convolutional ISA (ConvISA) that adopts the convolution-pooling structure of the state-of-the-art handcrafted video feature with greater modeling capacities and a cost-effective training algorithm. Through custom designed network structures for pixels and optical flow, our method also reflects distinctive characteristics of these two data sources. Our experimental results on standard action recognition benchmarks show that by focusing on the structure of CNNs, rather than end-to-end training methods, we are able to design an efficient and powerful video feature learning algorithm. | Features and encoding are the major sources of breakthroughs in conventional video representations. Among them the trajectory based approaches ( @cite_30 @cite_9 ), especially the Dense Trajectory (DT) and IDT ( @cite_25 @cite_2 ), are the basis of current state-of-the-art handcrafted algorithms. These trajectory-based methods are designed to address the flaws of image-extended video features. Their superior performance validates the need for a unique representation of motion features. | {
"cite_N": [
"@cite_30",
"@cite_9",
"@cite_25",
"@cite_2"
],
"mid": [
"2105101328",
"1211924006",
"2126574503",
""
],
"abstract": [
"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.",
"Human action recognition in videos is a challenging problem with wide applications. State-of-the-art approaches often adopt the popular bag-of-features representation based on isolated local patches or temporal patch trajectories, where motion patterns like object relationships are mostly discarded. This paper proposes a simple representation specifically aimed at the modeling of such motion relationships. We adopt global and local reference points to characterize motion information, so that the final representation can be robust to camera movement. Our approach operates on top of visual codewords derived from local patch trajectories, and therefore does not require accurate foreground-background separation, which is typically a necessary step to model object relationships. Through an extensive experimental evaluation, we show that the proposed representation offers very competitive performance on challenging benchmark datasets, and combining it with the bag-of-features representation leads to substantial improvement. On Hollywood2, Olympic Sports, and HMDB51 datasets, we obtain 59.5 , 80.6 and 40.7 respectively, which are the best reported results to date.",
"Feature trajectories have shown to be efficient for representing videos. Typically, they are extracted using the KLT tracker or matching SIFT descriptors between frames. However, the quality as well as quantity of these trajectories is often not sufficient. Inspired by the recent success of dense sampling in image classification, we propose an approach to describe videos by dense trajectories. We sample dense points from each frame and track them based on displacement information from a dense optical flow field. Given a state-of-the-art optical flow algorithm, our trajectories are robust to fast irregular motions as well as shot boundaries. Additionally, dense trajectories cover the motion information in videos well. We, also, investigate how to design descriptors to encode the trajectory information. We introduce a novel descriptor based on motion boundary histograms, which is robust to camera motion. This descriptor consistently outperforms other state-of-the-art descriptors, in particular in uncontrolled realistic videos. We evaluate our video description in the context of action classification with a bag-of-features approach. Experimental results show a significant improvement over the state of the art on four datasets of varying difficulty, i.e. KTH, YouTube, Hollywood2 and UCF sports.",
""
]
} |
1511.05045 | 2256809320 | Image and video classification research has made great progress through the development of handcrafted local features and learning based features. These two architectures were proposed roughly at the same time and have flourished at overlapping stages of history. However, they are typically viewed as distinct approaches. In this paper, we emphasize their structural similarities and show how such a unified view helps us in designing features that balance efficiency and effectiveness. As an example, we study the problem of designing efficient video feature learning algorithms for action recognition. We approach this problem by first showing that local handcrafted features and Convolutional Neural Networks (CNNs) share the same convolution-pooling network structure. We then propose a two-stream Convolutional ISA (ConvISA) that adopts the convolution-pooling structure of the state-of-the-art handcrafted video feature with greater modeling capacities and a cost-effective training algorithm. Through custom designed network structures for pixels and optical flow, our method also reflects distinctive characteristics of these two data sources. Our experimental results on standard action recognition benchmarks show that by focusing on the structure of CNNs, rather than end-to-end training methods, we are able to design an efficient and powerful video feature learning algorithm. | There have been many works attempting to improve IDT due to its popularity. @cite_22 enhanced the performance of IDT by increasing codebook sizes and fusing multiple coding methods. @cite_15 explored ways to sub-sample and generate vocabularies for DT features. @cite_18 achieved state-of-the-art performance on several action recognition datasets through applying three techniques including data augmentation, modeling score distribution over video subsequences, and capturing the relationship among action classes. @cite_10 modeled the evolution of appearance in the video and achieved state-of-the-art results on the Hollywood2 dataset. @cite_6 proposed to extract features from videos with multiple playback speeds to achieve speed invariances. However, none of them dealt with the fact that IDT relies on very simple handcrafted descriptors. In contrast, data-driven approaches have demonstrated their modeling power over image recognition ( @cite_0 ) and are gradually replacing traditional handcrafted methods. | {
"cite_N": [
"@cite_18",
"@cite_22",
"@cite_6",
"@cite_0",
"@cite_15",
"@cite_10"
],
"mid": [
"2176316098",
"2951552696",
"2951893483",
"",
"160239212",
"1926645898"
],
"abstract": [
"We propose two complementary techniques to improve the performance of action recognition systems. The first technique addresses the temporal interval ambiguity of actions by learning a classifier score distribution over video subsequences. A classifier based on this score distribution is shown to be more effective than using the maximum or average scores. The second technique learns a classifier for the relative values of action scores, capturing the correlation and exclusion between action classes. Both techniques are simple and have efficient implementations using a Least-Squares SVM. We demonstrate that taken together the techniques exceed the state-of-the-art performance by a wide margin on challenging benchmarks for human actions.",
"Video based action recognition is one of the important and challenging problems in computer vision research. Bag of Visual Words model (BoVW) with local features has become the most popular method and obtained the state-of-the-art performance on several realistic datasets, such as the HMDB51, UCF50, and UCF101. BoVW is a general pipeline to construct a global representation from a set of local features, which is mainly composed of five steps: (i) feature extraction, (ii) feature pre-processing, (iii) codebook generation, (iv) feature encoding, and (v) pooling and normalization. Many efforts have been made in each step independently in different scenarios and their effect on action recognition is still unknown. Meanwhile, video data exhibits different views of visual pattern, such as static appearance and motion dynamics. Multiple descriptors are usually extracted to represent these different views. Many feature fusion methods have been developed in other areas and their influence on action recognition has never been investigated before. This paper aims to provide a comprehensive study of all steps in BoVW and different fusion methods, and uncover some good practice to produce a state-of-the-art action recognition system. Specifically, we explore two kinds of local features, ten kinds of encoding methods, eight kinds of pooling and normalization strategies, and three kinds of fusion methods. We conclude that every step is crucial for contributing to the final recognition rate. Furthermore, based on our comprehensive study, we propose a simple yet effective representation, called hybrid representation, by exploring the complementarity of different BoVW frameworks and local descriptors. Using this representation, we obtain the state-of-the-art on the three challenging datasets: HMDB51 (61.1 ), UCF50 (92.3 ), and UCF101 (87.9 ).",
"Most state-of-the-art action feature extractors involve differential operators, which act as highpass filters and tend to attenuate low frequency action information. This attenuation introduces bias to the resulting features and generates ill-conditioned feature matrices. The Gaussian Pyramid has been used as a feature enhancing technique that encodes scale-invariant characteristics into the feature space in an attempt to deal with this attenuation. However, at the core of the Gaussian Pyramid is a convolutional smoothing operation, which makes it incapable of generating new features at coarse scales. In order to address this problem, we propose a novel feature enhancing technique called Multi-skIp Feature Stacking (MIFS), which stacks features extracted using a family of differential filters parameterized with multiple time skips and encodes shift-invariance into the frequency space. MIFS compensates for information lost from using differential operators by recapturing information at coarse scales. This recaptured information allows us to match actions at different speeds and ranges of motion. We prove that MIFS enhances the learnability of differential-based features exponentially. The resulting feature matrices from MIFS have much smaller conditional numbers and variances than those from conventional methods. Experimental results show significantly improved performance on challenging action recognition and event detection tasks. Specifically, our method exceeds the state-of-the-arts on Hollywood2, UCF101 and UCF50 datasets and is comparable to state-of-the-arts on HMDB51 and Olympics Sports datasets. MIFS can also be used as a speedup strategy for feature extraction with minimal or no accuracy cost.",
"",
"The recent trend in action recognition is towards larger datasets, an increasing number of action classes and larger visual vocabularies. State-of-the-art human action classification in challenging video data is currently based on a bag-of-visual-words pipeline in which space-time features are aggregated globally to form a histogram. The strategies chosen to sample features and construct a visual vocabulary are critical to performance, in fact often dominating performance. In this work we provide a critical evaluation of various approaches to building a vocabulary and show that good practises do have a significant impact. By subsampling and partitioning features strategically, we are able to achieve state-of-the-art results on 5 major action recognition datasets using relatively small visual vocabularies.",
"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."
]
} |
1511.05045 | 2256809320 | Image and video classification research has made great progress through the development of handcrafted local features and learning based features. These two architectures were proposed roughly at the same time and have flourished at overlapping stages of history. However, they are typically viewed as distinct approaches. In this paper, we emphasize their structural similarities and show how such a unified view helps us in designing features that balance efficiency and effectiveness. As an example, we study the problem of designing efficient video feature learning algorithms for action recognition. We approach this problem by first showing that local handcrafted features and Convolutional Neural Networks (CNNs) share the same convolution-pooling network structure. We then propose a two-stream Convolutional ISA (ConvISA) that adopts the convolution-pooling structure of the state-of-the-art handcrafted video feature with greater modeling capacities and a cost-effective training algorithm. Through custom designed network structures for pixels and optical flow, our method also reflects distinctive characteristics of these two data sources. Our experimental results on standard action recognition benchmarks show that by focusing on the structure of CNNs, rather than end-to-end training methods, we are able to design an efficient and powerful video feature learning algorithm. | There has been limited number of works on unsupervised methods for learning video features. Among them the Independent Component Analysis (ICA) ( @cite_29 ) was the first approach to learn representations of videos in an unsupervised way. @cite_26 addressed the issue using stacked ConvISA. @cite_19 applied unsupervised feature learning through long-short term memory. Since these methods rely purely on pixel data, they struggled to capture motion information and generally performed no better than state-of-the-art handcrafted methods. Our previous work ( @cite_27 ) tried to use stacked ConvISA to learn filters for optical flow data, but also could not outperform handcrafted methods due to the fact that stacked ConvISA does not have a good network structure for optical-flow feature learning. | {
"cite_N": [
"@cite_19",
"@cite_29",
"@cite_26",
"@cite_27"
],
"mid": [
"2952453038",
"2142457795",
"1999192586",
"2952580834"
],
"abstract": [
"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.",
"Recently, statistical models of natural images have shown the emergence of several properties of the visual cortex. Most models have considered the nongaussian properties of static image patches, leading to sparse coding or independent component analysis. Here we consider the basic time dependencies of image sequences instead of their nongaussianity. We show that simple-cell-type receptive fields emerge when temporal response strength correlation is maximized for natural image sequences. Thus, temporal response strength correlation, which is a nonlinear measure of temporal coherence, provides an alternative to sparseness in modeling simple-cell receptive field properties. Our results also suggest an interpretation of simple cells in terms of invariant coding principles, which have previously been used to explain complex-cell receptive fields.",
"Previous work on action recognition has focused on adapting hand-designed local features, such as SIFT or HOG, from static images to the video domain. In this paper, we propose using unsupervised feature learning as a way to learn features directly from video data. More specifically, we present an extension of the Independent Subspace Analysis algorithm to learn invariant spatio-temporal features from unlabeled video data. We discovered that, despite its simplicity, this method performs surprisingly well when combined with deep learning techniques such as stacking and convolution to learn hierarchical representations. By replacing hand-designed features with our learned features, we achieve classification results superior to all previous published results on the Hollywood2, UCF, KTH and YouTube action recognition datasets. On the challenging Hollywood2 and YouTube action datasets we obtain 53.3 and 75.8 respectively, which are approximately 5 better than the current best published results. Further benefits of this method, such as the ease of training and the efficiency of training and prediction, will also be discussed. You can download our code and learned spatio-temporal features here: http: ai.stanford.edu ∼wzou",
"Motivated by the success of data-driven convolutional neural networks (CNNs) in object recognition on static images, researchers are working hard towards developing CNN equivalents for learning video features. However, learning video features globally has proven to be quite a challenge due to its high dimensionality, the lack of labelled data and the difficulty in processing large-scale video data. Therefore, we propose to leverage effective techniques from both data-driven and data-independent approaches to improve action recognition system. Our contribution is three-fold. First, we propose a two-stream Stacked Convolutional Independent Subspace Analysis (ConvISA) architecture to show that unsupervised learning methods can significantly boost the performance of traditional local features extracted from data-independent models. Second, we demonstrate that by learning on video volumes detected by Improved Dense Trajectory (IDT), we can seamlessly combine our novel local descriptors with hand-crafted descriptors. Thus we can utilize available feature enhancing techniques developed for hand-crafted descriptors. Finally, similar to multi-class classification framework in CNNs, we propose a training-free re-ranking technique that exploits the relationship among action classes to improve the overall performance. Our experimental results on four benchmark action recognition datasets show significantly improved performance."
]
} |
1511.05292 | 2176125451 | Recognizing actions from still images is popularly studied recently. In this paper, we model an action class as a flexible number of spatial configurations of body parts by proposing a new spatial SPN (Sum-Product Networks). First, we discover a set of parts in image collections via unsupervised learning. Then, our new spatial SPN is applied to model the spatial relationship and also the high-order correlations of parts. To learn robust networks, we further develop a hierarchical spatial SPN method, which models pairwise spatial relationship between parts inside sub-images and models the correlation of sub-images via extra layers of SPN. Our method is shown to be effective on two benchmark datasets. | To recognize actions from images, researchers propose methods to learn discriminative representations for humans under different poses. Ikizler- @cite_2 learn HOG-feature based representations for different action classes based on the images collected from Web. Thurau and Halvac @cite_35 train a set of pose primitives by non-negative matrix decomposition of HOG-descriptor and represent images using these pose primitives. @cite_24 propose a technique for deformable matching of edges from a pair of images. These methods @cite_2 @cite_35 @cite_24 extract features from the whole image and obtain a global template. However, these global templates are not effective for the action recognition due to the significant pose variations in images. | {
"cite_N": [
"@cite_24",
"@cite_35",
"@cite_2"
],
"mid": [
"2148948295",
"2103822353",
"2120601899"
],
"abstract": [
"In this paper we consider the problem of describing the action being performed by human figures in still images. We will attack this problem using an unsupervised learning approach, attempting to discover the set of action classes present in a large collection of training images. These action classes will then be used to label test images. Our approach uses the coarse shape of the human figures to match pairs of images. The distance between a pair of images is computed using a linear programming relaxation technique. This is a computationally expensive process, and we employ a fast pruning method to enable its use on a large collection of images. Spectral clustering is then performed using the resulting distances. We present clustering and image labeling results on a variety of datasets.",
"This paper presents a method for recognizing human actions based on pose primitives. In learning mode, the parameters representing poses and activities are estimated from videos. In run mode, the method can be used both for videos or still images. For recognizing pose primitives, we extend a Histogram of Oriented Gradient (HOG) based descriptor to better cope with articulated poses and cluttered background. Action classes are represented by histograms of poses primitives. For sequences, we incorporate the local temporal context by means of n-gram expressions. Action recognition is based on a simple histogram comparison. Unlike the mainstream video surveillance approaches, the proposed method does not rely on background subtraction or dynamic features and thus allows for action recognition in still images.",
"This paper proposes a generic method for action recognition in uncontrolled videos. The idea is to use images collected from the Web to learn representations of actions and use this knowledge to automatically annotate actions in videos. Our approach is unsupervised in the sense that it requires no human intervention other than the text querying. Its benefits are two-fold: 1) we can improve retrieval of action images, and 2) we can collect a large generic database of action poses, which can then be used in tagging videos. We present experimental evidence that using action images collected from the Web, annotating actions is possible."
]
} |
1511.05292 | 2176125451 | Recognizing actions from still images is popularly studied recently. In this paper, we model an action class as a flexible number of spatial configurations of body parts by proposing a new spatial SPN (Sum-Product Networks). First, we discover a set of parts in image collections via unsupervised learning. Then, our new spatial SPN is applied to model the spatial relationship and also the high-order correlations of parts. To learn robust networks, we further develop a hierarchical spatial SPN method, which models pairwise spatial relationship between parts inside sub-images and models the correlation of sub-images via extra layers of SPN. Our method is shown to be effective on two benchmark datasets. | SPN (Sum-Product Networks) @cite_23 is a newly proposed deep structure that can capture the high-order correlations between its leaf nodes. A number of papers later investigate this structure theoretically @cite_37 @cite_44 . In addition, SPN is proved to be successful in several tasks of computer vision, including image classification @cite_11 , facial attribute analysis @cite_7 , and action recognition in videos @cite_29 . In this paper, we characterize an action class by several configurations of body parts, and represent it using SPN. Amer and Todorovic @cite_29 propose to learn a Bag-of-Words representation for a video and model the deep correlations of parts using SPN with a stochastic structure. However, the spatial relationships of the parts are not taken into consideration. For the first time, in this work, we propose spatial SPN to explicitly model the spatial relationship of parts for robust action recognition. | {
"cite_N": [
"@cite_37",
"@cite_7",
"@cite_29",
"@cite_44",
"@cite_23",
"@cite_11"
],
"mid": [
"2141473882",
"2143352446",
"2064052975",
"2119900738",
"2949869425",
"2128744540"
],
"abstract": [
"We investigate the representational power of sum-product networks (computation networks analogous to neural networks, but whose individual units compute either products or weighted sums), through a theoretical analysis that compares deep (multiple hidden layers) vs. shallow (one hidden layer) architectures. We prove there exist families of functions that can be represented much more efficiently with a deep network than with a shallow one, i.e. with substantially fewer hidden units. Such results were not available until now, and contribute to motivate recent research involving learning of deep sum-product networks, and more generally motivate research in Deep Learning.",
"Recent works have shown that facial attributes are useful in a number of applications such as face recognition and retrieval. However, estimating attributes in images with large variations remains a big challenge. This challenge is addressed in this paper. Unlike existing methods that assume the independence of attributes during their estimation, our approach captures the interdependencies of local regions for each attribute, as well as the high-order correlations between different attributes, which makes it more robust to occlusions and misdetection of face regions. First, we have modeled region interdependencies with a discriminative decision tree, where each node consists of a detector and a classifier trained on a local region. The detector allows us to locate the region, while the classifier determines the presence or absence of an attribute. Second, correlations of attributes and attribute predictors are modeled by organizing all of the decision trees into a large sum-product network (SPN), which is learned by the EM algorithm and yields the most probable explanation (MPE) of the facial attributes in terms of the region's localization and classification. Experimental results on a large data set with 22,400 images show the effectiveness of the proposed approach.",
"This paper addresses recognition of human activities with stochastic structure, characterized by variable spacetime arrangements of primitive actions, and conducted by a variable number of actors. We demonstrate that modeling aggregate counts of visual words is surprisingly expressive enough for such a challenging recognition task. An activity is represented by a sum-product network (SPN). SPN is a mixture of bags-of-words (BoWs) with exponentially many mixture components, where subcomponents are reused by larger ones. SPN consists of terminal nodes representing BoWs, and product and sum nodes organized in a number of layers. The products are aimed at encoding particular configurations of primitive actions, and the sums serve to capture their alternative configurations. The connectivity of SPN and parameters of BoW distributions are learned under weak supervision using the EM algorithm. SPN inference amounts to parsing the SPN graph, which yields the most probable explanation (MPE) of the video in terms of activity detection and localization. SPN inference has linear complexity in the number of nodes, under fairly general conditions, enabling fast and scalable recognition. A new Volleyball dataset is compiled and annotated for evaluation. Our classification accuracy and localization precision and recall are superior to those of the state-of-the-art on the benchmark and our Volleyball datasets.",
"Sum-product networks (SPNs) are a deep probabilistic representation that allows for efficient, exact inference. SPNs generalize many other tractable models, including thin junction trees, latent tree models, and many types of mixtures. Previous work on learning SPN structure has mainly focused on using top-down or bottom-up clustering to find mixtures, which capture variable interactions indirectly through implicit latent variables. In contrast, most work on learning graphical models, thin junction trees, and arithmetic circuits has focused on finding direct interactions among variables. In this paper, we present ID-SPN, a new algorithm for learning SPN structure that unifies the two approaches. In experiments on 20 benchmark datasets, we find that the combination of direct and indirect interactions leads to significantly better accuracy than several state-of-the-art algorithms for learning SPNs and other tractable models.",
"The key limiting factor in graphical model inference and learning is the complexity of the partition function. We thus ask the question: what are general conditions under which the partition function is tractable? The answer leads to a new kind of deep architecture, which we call sum-product networks (SPNs). SPNs are directed acyclic graphs with variables as leaves, sums and products as internal nodes, and weighted edges. We show that if an SPN is complete and consistent it represents the partition function and all marginals of some graphical model, and give semantics to its nodes. Essentially all tractable graphical models can be cast as SPNs, but SPNs are also strictly more general. We then propose learning algorithms for SPNs, based on backpropagation and EM. Experiments show that inference and learning with SPNs can be both faster and more accurate than with standard deep networks. For example, SPNs perform image completion better than state-of-the-art deep networks for this task. SPNs also have intriguing potential connections to the architecture of the cortex.",
"Sum-product networks are a new deep architecture that can perform fast, exact inference on high-treewidth models. Only generative methods for training SPNs have been proposed to date. In this paper, we present the first discriminative training algorithms for SPNs, combining the high accuracy of the former with the representational power and tractability of the latter. We show that the class of tractable discriminative SPNs is broader than the class of tractable generative ones, and propose an efficient backpropagation-style algorithm for computing the gradient of the conditional log likelihood. Standard gradient descent suffers from the diffusion problem, but networks with many layers can be learned reliably using \"hard\" gradient descent, where marginal inference is replaced by MPE inference (i.e., inferring the most probable state of the non-evidence variables). The resulting updates have a simple and intuitive form. We test discriminative SPNs on standard image classification tasks. We obtain the best results to date on the CIFAR-10 dataset, using fewer features than prior methods with an SPN architecture that learns local image structure discriminatively. We also report the highest published test accuracy on STL-10 even though we only use the labeled portion of the dataset."
]
} |
1511.04897 | 2469208367 | In the last 10 years, cache attacks on Intel x86 CPUs have gained increasing attention among the scientific community and powerful techniques to exploit cache side channels have been developed. However, modern smartphones use one or more multi-core ARM CPUs that have a different cache organization and instruction set than Intel x86 CPUs. So far, no cross-core cache attacks have been demonstrated on non-rooted Android smartphones. In this work, we demonstrate how to solve key challenges to perform the most powerful cross-core cache attacks Prime+Probe, Flush+Reload, Evict+Reload, and Flush+Flush on non-rooted ARM-based devices without any privileges. Based on our techniques, we demonstrate covert channels that outperform state-of-the-art covert channels on Android by several orders of magnitude. Moreover, we present attacks to monitor tap and swipe events as well as keystrokes, and even derive the lengths of words entered on the touchscreen. Eventually, we are the first to attack cryptographic primitives implemented in Java. Our attacks work across CPUs and can even monitor cache activity in the ARM TrustZone from the normal world. The techniques we present can be used to attack hundreds of millions of Android devices. | Modern caches organize cache lines in multiple sets, which is also known as set-associative caches. Each memory address maps to one of these cache sets and addresses that map to the same cache set are considered congruent. Congruent addresses compete for cache lines within the same set and a predefined replacement policy determines which cache line is replaced. For instance, the last generations of Intel CPUs employ an undocumented variant of least-recently used (LRU) replacement policy @cite_51 . ARM processors use a pseudo-LRU replacement policy for the L1 cache and they support two different cache replacement policies for L2 caches, namely round-robin and pseudo-random replacement policy. In practice, however, only the pseudo-random replacement policy is used due to performance reasons. Switching the cache replacement policy is only possible in privileged mode. The implementation details for the pseudo-random policy are not documented. | {
"cite_N": [
"@cite_51"
],
"mid": [
"2953300605"
],
"abstract": [
"A fundamental assumption in software security is that a memory location can only be modified by processes that may write to this memory location. However, a recent study has shown that parasitic effects in DRAM can change the content of a memory cell without accessing it, but by accessing other memory locations in a high frequency. This so-called Rowhammer bug occurs in most of today's memory modules and has fatal consequences for the security of all affected systems, e.g., privilege escalation attacks. All studies and attacks related to Rowhammer so far rely on the availability of a cache flush instruction in order to cause accesses to DRAM modules at a sufficiently high frequency. We overcome this limitation by defeating complex cache replacement policies. We show that caches can be forced into fast cache eviction to trigger the Rowhammer bug with only regular memory accesses. This allows to trigger the Rowhammer bug in highly restricted and even scripting environments. We demonstrate a fully automated attack that requires nothing but a website with JavaScript to trigger faults on remote hardware. Thereby we can gain unrestricted access to systems of website visitors. We show that the attack works on off-the-shelf systems. Existing countermeasures fail to protect against this new Rowhammer attack."
]
} |
1511.04897 | 2469208367 | In the last 10 years, cache attacks on Intel x86 CPUs have gained increasing attention among the scientific community and powerful techniques to exploit cache side channels have been developed. However, modern smartphones use one or more multi-core ARM CPUs that have a different cache organization and instruction set than Intel x86 CPUs. So far, no cross-core cache attacks have been demonstrated on non-rooted Android smartphones. In this work, we demonstrate how to solve key challenges to perform the most powerful cross-core cache attacks Prime+Probe, Flush+Reload, Evict+Reload, and Flush+Flush on non-rooted ARM-based devices without any privileges. Based on our techniques, we demonstrate covert channels that outperform state-of-the-art covert channels on Android by several orders of magnitude. Moreover, we present attacks to monitor tap and swipe events as well as keystrokes, and even derive the lengths of words entered on the touchscreen. Eventually, we are the first to attack cryptographic primitives implemented in Java. Our attacks work across CPUs and can even monitor cache activity in the ARM TrustZone from the normal world. The techniques we present can be used to attack hundreds of millions of Android devices. | CPUs have multiple cache levels, with the lower levels being faster and smaller than the higher levels. ARM processors typically have two levels of cache. If all cache lines from lower levels are also stored in a higher-level cache, the higher-level cache is called . If a cache line can only reside in one of the cache levels at any point in time, the caches are called . If the cache is neither inclusive nor exclusive, it is called . The last-level cache is often shared among all cores to enhance the performance upon transitioning threads between cores and to simplify cross-core cache lookups. However, with shared last-level caches, one core can (intentionally) influence the cache content of all other cores. This represents the basis for cache attacks like @cite_40 . | {
"cite_N": [
"@cite_40"
],
"mid": [
"1427174644"
],
"abstract": [
"Sharing memory pages between non-trusting processes is a common method of reducing the memory footprint of multi-tenanted systems. In this paper we demonstrate that, due to a weakness in the Intel X86 processors, page sharing exposes processes to information leaks. We present FLUSH+RELOAD, a cache side-channel attack technique that exploits this weakness to monitor access to memory lines in shared pages. Unlike previous cache side-channel attacks, FLUSH+RELOAD targets the Last-Level Cache (i.e. L3 on processors with three cache levels). Consequently, the attack program and the victim do not need to share the execution core. We demonstrate the efficacy of the FLUSH+RELOAD attack by using it to extract the private encryption keys from a victim program running GnuPG 1.4.13. We tested the attack both between two unrelated processes in a single operating system and between processes running in separate virtual machines. On average, the attack is able to recover 96.7 of the bits of the secret key by observing a single signature or decryption round."
]
} |
1511.04897 | 2469208367 | In the last 10 years, cache attacks on Intel x86 CPUs have gained increasing attention among the scientific community and powerful techniques to exploit cache side channels have been developed. However, modern smartphones use one or more multi-core ARM CPUs that have a different cache organization and instruction set than Intel x86 CPUs. So far, no cross-core cache attacks have been demonstrated on non-rooted Android smartphones. In this work, we demonstrate how to solve key challenges to perform the most powerful cross-core cache attacks Prime+Probe, Flush+Reload, Evict+Reload, and Flush+Flush on non-rooted ARM-based devices without any privileges. Based on our techniques, we demonstrate covert channels that outperform state-of-the-art covert channels on Android by several orders of magnitude. Moreover, we present attacks to monitor tap and swipe events as well as keystrokes, and even derive the lengths of words entered on the touchscreen. Eventually, we are the first to attack cryptographic primitives implemented in Java. Our attacks work across CPUs and can even monitor cache activity in the ARM TrustZone from the normal world. The techniques we present can be used to attack hundreds of millions of Android devices. | Both approaches allow an adversary to determine which cache sets are used during the victim's computations and have been exploited to attack cryptographic implementations @cite_35 @cite_24 @cite_33 @cite_5 and to build cross-VM covert channels @cite_21 . Yarom and Falkner @cite_40 proposed , a significantly more fine-grained attack that exploits three fundamental concepts of modern system architectures. First, the availability of shared memory between the victim process and the adversary. Second, last-level caches are typically shared among all cores. Third, Intel platforms use inclusive last-level caches, meaning that the eviction of information from the last-level cache leads to the eviction of this data from all lower-level caches of other cores, which allows any program to evict data from other programs on other cores. While the basic idea of this attack has been proposed by Gullasch al @cite_25 , Yarom and Falkner extended this idea to shared last-level caches, allowing cross-core attacks. works as follows. | {
"cite_N": [
"@cite_35",
"@cite_33",
"@cite_21",
"@cite_24",
"@cite_40",
"@cite_5",
"@cite_25"
],
"mid": [
"1488058190",
"1503814339",
"1494212869",
"2103289002",
"1427174644",
"1934458198",
"2172060328"
],
"abstract": [
"We describe several software side-channel attacks based on inter-process leakage through the state of the CPU’s memory cache. This leakage reveals memory access patterns, which can be used for cryptanalysis of cryptographic primitives that employ data-dependent table lookups. The attacks allow an unprivileged process to attack other processes running in parallel on the same processor, despite partitioning methods such as memory protection, sandboxing and virtualization. Some of our methods require only the ability to trigger services that perform encryption or MAC using the unknown key, such as encrypted disk partitions or secure network links. Moreover, we demonstrate an extremely strong type of attack, which requires knowledge of neither the specific plaintexts nor ciphertexts, and works by merely monitoring the effect of the cryptographic process on the cache. We discuss in detail several such attacks on AES, and experimentally demonstrate their applicability to real systems, such as OpenSSL and Linux’s dm-crypt encrypted partitions (in the latter case, the full key can be recovered after just 800 writes to the partition, taking 65 milliseconds). Finally, we describe several countermeasures for mitigating such attacks.",
"The cloud computing infrastructure relies on virtualized servers that provide isolation across guest OS's through sand boxing. This isolation was demonstrated to be imperfect in past work which exploited hardware level information leakages to gain access to sensitive information across co-located virtual machines (VMs). In response virtualization companies and cloud services providers have disabled features such as deduplication to prevent such attacks. In this work, we introduce a fine-grain cross-core cache attack that exploits access time variations on the last level cache. The attack exploits huge pages to work across VM boundaries without requiring deduplication. No configuration changes on the victim OS are needed, making the attack quite viable. Furthermore, only machine co-location is required, while the target and victim OS can still reside on different cores of the machine. Our new attack is a variation of the prime and probe cache attack whose applicability at the time is limited to L1 cache. In contrast, our attack works in the spirit of the flush and reload attack targeting the shared L3 cache instead. Indeed, by adjusting the huge page size our attack can be customized to work virtually at any cache level size. We demonstrate the viability of the attack by targeting an Open SSL1.0.1f implementation of AES. The attack recovers AES keys in the cross-VM setting on Xen 4.1 with deduplication disabled, being only slightly less efficient than the flush and reload attack. Given that huge pages are a standard feature enabled in the memory management unit of OS's and that besides co-location no additional assumptions are needed, the attack we present poses a significant risk to existing cloud servers.",
"Cloud computing relies on hypervisors to isolate virtual machines running on shared hardware. Since perfect isolation is difficult to achieve, sharing hardware induces threats. Covert channels were demonstrated to violate isolation and, typically, allow data exfiltration. Several covert channels have been proposed that rely on the processor's cache. However, these covert channels are either slow or impractical due to the addressing uncertainty. This uncertainty exists in particular in virtualized environments and with recent L3 caches which are using complex addressing. Using shared memory would elude addressing uncertainty, but shared memory is not available in most practical setups. We build C5, a covert channel that tackles addressing uncertainty without requiring any shared memory, making the covert channel fast and practical. We are able to transfer messages on modern hardware across any cores of the same processor. The covert channel targets the last level cache that is shared across all cores. It exploits the inclusive feature of caches, allowing a core to evict lines in the private first level cache of another core. We experimentally evaluate the covert channel in native and virtualized environments. In particular, we successfully establish a covert channel between virtual machines running on different cores. We measure a bitrate of 1291i¾?bps for a native setup, and 751i¾?bps for a virtualized setup. This is one order of magnitude above previous cache-based covert channels in the same setup.",
"We describe several software side-channel attacks based on inter-process leakage through the state of the CPU’s memory cache. This leakage reveals memory access patterns, which can be used for cryptanalysis of cryptographic primitives that employ data-dependent table lookups. The attacks allow an unprivileged process to attack other processes running in parallel on the same processor, despite partitioning methods such as memory protection, sandboxing, and virtualization. Some of our methods require only the ability to trigger services that perform encryption or MAC using the unknown key, such as encrypted disk partitions or secure network links. Moreover, we demonstrate an extremely strong type of attack, which requires knowledge of neither the specific plaintexts nor ciphertexts and works by merely monitoring the effect of the cryptographic process on the cache. We discuss in detail several attacks on AES and experimentally demonstrate their applicability to real systems, such as OpenSSL and Linux’s dm-crypt encrypted partitions (in the latter case, the full key was recovered after just 800 writes to the partition, taking 65 milliseconds). Finally, we discuss a variety of countermeasures which can be used to mitigate such attacks.",
"Sharing memory pages between non-trusting processes is a common method of reducing the memory footprint of multi-tenanted systems. In this paper we demonstrate that, due to a weakness in the Intel X86 processors, page sharing exposes processes to information leaks. We present FLUSH+RELOAD, a cache side-channel attack technique that exploits this weakness to monitor access to memory lines in shared pages. Unlike previous cache side-channel attacks, FLUSH+RELOAD targets the Last-Level Cache (i.e. L3 on processors with three cache levels). Consequently, the attack program and the victim do not need to share the execution core. We demonstrate the efficacy of the FLUSH+RELOAD attack by using it to extract the private encryption keys from a victim program running GnuPG 1.4.13. We tested the attack both between two unrelated processes in a single operating system and between processes running in separate virtual machines. On average, the attack is able to recover 96.7 of the bits of the secret key by observing a single signature or decryption round.",
"We present an effective implementation of the Prime Probe side-channel attack against the last-level cache. We measure the capacity of the covert channel the attack creates and demonstrate a cross-core, cross-VM attack on multiple versions of GnuPG. Our technique achieves a high attack resolution without relying on weaknesses in the OS or virtual machine monitor or on sharing memory between attacker and victim.",
"Side channel attacks on cryptographic systems exploit information gained from physical implementations rather than theoretical weaknesses of a scheme. In recent years, major achievements were made for the class of so called access-driven cache attacks. Such attacks exploit the leakage of the memory locations accessed by a victim process. In this paper we consider the AES block cipher and present an attack which is capable of recovering the full secret key in almost real time for AES-128, requiring only a very limited number of observed encryptions. Unlike previous attacks, we do not require any information about the plaintext (such as its distribution, etc.). Moreover, for the first time, we also show how the plaintext can be recovered without having access to the cipher text at all. It is the first working attack on AES implementations using compressed tables. There, no efficient techniques to identify the beginning of AES rounds is known, which is the fundamental assumption underlying previous attacks. We have a fully working implementation of our attack which is able to recover AES keys after observing as little as 100 encryptions. It works against the OpenS SL 0.9.8n implementation of AES on Linux systems. Our spy process does not require any special privileges beyond those of a standard Linux user. A contribution of probably independent interest is a denial of service attack on the task scheduler of current Linux systems (CFS), which allows one to observe (on average) every single memory access of a victim process."
]
} |
1511.04897 | 2469208367 | In the last 10 years, cache attacks on Intel x86 CPUs have gained increasing attention among the scientific community and powerful techniques to exploit cache side channels have been developed. However, modern smartphones use one or more multi-core ARM CPUs that have a different cache organization and instruction set than Intel x86 CPUs. So far, no cross-core cache attacks have been demonstrated on non-rooted Android smartphones. In this work, we demonstrate how to solve key challenges to perform the most powerful cross-core cache attacks Prime+Probe, Flush+Reload, Evict+Reload, and Flush+Flush on non-rooted ARM-based devices without any privileges. Based on our techniques, we demonstrate covert channels that outperform state-of-the-art covert channels on Android by several orders of magnitude. Moreover, we present attacks to monitor tap and swipe events as well as keystrokes, and even derive the lengths of words entered on the touchscreen. Eventually, we are the first to attack cryptographic primitives implemented in Java. Our attacks work across CPUs and can even monitor cache activity in the ARM TrustZone from the normal world. The techniques we present can be used to attack hundreds of millions of Android devices. | Thereby, allows an attacker to determine which specific instructions are executed and also which specific data is accessed by the victim program. Thus, rather fine-grained attacks are possible and have already been demonstrated against cryptographic implementations @cite_59 @cite_47 @cite_30 . Furthermore, Gruss al @cite_37 demonstrated the possibility to automatically exploit cache-based side-channel information based on the approach. Besides attacking cryptographic implementations like AES T-table implementations, they showed how to infer keystroke information and even how to build a keylogger by exploiting the cache side channel. Similarly, Oren al @cite_8 demonstrated the possibility to exploit cache attacks on Intel platforms from JavaScript and showed how to infer visited websites and how to track the user's mouse activity. | {
"cite_N": [
"@cite_30",
"@cite_37",
"@cite_8",
"@cite_59",
"@cite_47"
],
"mid": [
"2296391043",
"1555558540",
"2061354941",
"1981260134",
"1964389195"
],
"abstract": [
"Cloud's unrivaled cost effectiveness and on the fly operation versatility is attractive to enterprise and personal users. However, the cloud inherits a dangerous behavior from virtualization systems that poses a serious security risk: resource sharing. This work exploits a shared resource optimization technique called memory deduplication to mount a powerful known-ciphertext only cache side-channel attack on a popular OpenSSL implementation of AES. In contrast to the other cross-VM cache attacks, our attack does not require synchronization with the target server and is fully asynchronous, working in a more realistic scenario with much weaker assumption. Also, our attack succeeds in just 15 seconds working across cores in the cross-VM setting. Our results show that there is strong information leakage through cache in virtualized systems and the memory deduplication should be approached with caution.",
"Recent work on cache attacks has shown that CPU caches represent a powerful source of information leakage. However, existing attacks require manual identification of vulnerabilities, i.e., data accesses or instruction execution depending on secret information. In this paper, we present Cache Template Attacks. This generic attack technique allows us to profile and exploit cache-based information leakage of any program automatically, without prior knowledge of specific software versions or even specific system information. Cache Template Attacks can be executed online on a remote system without any prior offline computations or measurements. Cache Template Attacks consist of two phases. In the profiling phase, we determine dependencies between the processing of secret information, e.g., specific key inputs or private keys of cryptographic primitives, and specific cache accesses. In the exploitation phase, we derive the secret values based on observed cache accesses. We illustrate the power of the presented approach in several attacks, but also in a useful application for developers. Among the presented attacks is the application of Cache Template Attacks to infer keystrokes and--even more severe--the identification of specific keys on Linux and Windows user interfaces. More specifically, for lowercase only passwords, we can reduce the entropy per character from log2(26) = 4.7 to 1.4 bits on Linux systems. Furthermore, we perform an automated attack on the T-table-based AES implementation of OpenSSL that is as efficient as state-of-the-art manual cache attacks.",
"We present a micro-architectural side-channel attack that runs entirely in the browser. In contrast to previous work in this genre, our attack does not require the attacker to install software on the victim's machine; to facilitate the attack, the victim needs only to browse to an untrusted webpage that contains attacker-controlled content. This makes our attack model highly scalable, and extremely relevant and practical to today's Web, as most desktop browsers currently used to access the Internet are affected by such side channel threats. Our attack, which is an extension to the last-level cache attacks of , allows a remote adversary to recover information belonging to other processes, users, and even virtual machines running on the same physical host with the victim web browser. We describe the fundamentals behind our attack, and evaluate its performance characteristics. In addition, we show how it can be used to compromise user privacy in a common setting, letting an attacker spy after a victim that uses private browsing. Defending against this side channel is possible, but the required countermeasures can exact an impractical cost on benign uses of the browser.",
"",
"In this work we show how the Lucky 13 attack can be resurrected in the cloud by gaining access to a virtual machine co-located with the target. Our version of the attack exploits distinguishable cache access times enabled by VM deduplication to detect dummy function calls that only happen in case of an incorrectly CBC-padded TLS packet. Thereby, we gain back a new covert channel not considered in the original paper that enables the Lucky 13 attack. In fact, the new side channel is significantly more accurate, thus yielding a much more effective attack. We briefly survey prominent cryptographic libraries for this vulnerability. The attack currently succeeds to compromise PolarSSL, GnuTLS and CyaSSL on deduplication enabled platforms while the Lucky 13 patches in OpenSSL, Mozilla NSS and MatrixSSL are immune to this vulnerability. We conclude that, any program that follows secret data dependent execution flow is exploitable by side-channel attacks as shown in (but not limited to) our version of the Lucky 13 attack."
]
} |
1511.04897 | 2469208367 | In the last 10 years, cache attacks on Intel x86 CPUs have gained increasing attention among the scientific community and powerful techniques to exploit cache side channels have been developed. However, modern smartphones use one or more multi-core ARM CPUs that have a different cache organization and instruction set than Intel x86 CPUs. So far, no cross-core cache attacks have been demonstrated on non-rooted Android smartphones. In this work, we demonstrate how to solve key challenges to perform the most powerful cross-core cache attacks Prime+Probe, Flush+Reload, Evict+Reload, and Flush+Flush on non-rooted ARM-based devices without any privileges. Based on our techniques, we demonstrate covert channels that outperform state-of-the-art covert channels on Android by several orders of magnitude. Moreover, we present attacks to monitor tap and swipe events as well as keystrokes, and even derive the lengths of words entered on the touchscreen. Eventually, we are the first to attack cryptographic primitives implemented in Java. Our attacks work across CPUs and can even monitor cache activity in the ARM TrustZone from the normal world. The techniques we present can be used to attack hundreds of millions of Android devices. | Gruss al @cite_37 proposed the technique that replaces the flush instruction in by eviction. While it has no practical application on x86 CPUs, we show that it can be used on ARM CPUs. Recently, @cite_7 has been proposed. Unlike other techniques, it does not perform any memory access but relies on the timing of the flush instruction to determine whether a line has been loaded by a victim. We show that the execution time of the ARMv8-A flush instruction also depends on whether or not data is cached and, thus, can be used to implement this attack. | {
"cite_N": [
"@cite_37",
"@cite_7"
],
"mid": [
"1555558540",
"2949505625"
],
"abstract": [
"Recent work on cache attacks has shown that CPU caches represent a powerful source of information leakage. However, existing attacks require manual identification of vulnerabilities, i.e., data accesses or instruction execution depending on secret information. In this paper, we present Cache Template Attacks. This generic attack technique allows us to profile and exploit cache-based information leakage of any program automatically, without prior knowledge of specific software versions or even specific system information. Cache Template Attacks can be executed online on a remote system without any prior offline computations or measurements. Cache Template Attacks consist of two phases. In the profiling phase, we determine dependencies between the processing of secret information, e.g., specific key inputs or private keys of cryptographic primitives, and specific cache accesses. In the exploitation phase, we derive the secret values based on observed cache accesses. We illustrate the power of the presented approach in several attacks, but also in a useful application for developers. Among the presented attacks is the application of Cache Template Attacks to infer keystrokes and--even more severe--the identification of specific keys on Linux and Windows user interfaces. More specifically, for lowercase only passwords, we can reduce the entropy per character from log2(26) = 4.7 to 1.4 bits on Linux systems. Furthermore, we perform an automated attack on the T-table-based AES implementation of OpenSSL that is as efficient as state-of-the-art manual cache attacks.",
"Research on cache attacks has shown that CPU caches leak significant information. Proposed detection mechanisms assume that all cache attacks cause more cache hits and cache misses than benign applications and use hardware performance counters for detection. In this article, we show that this assumption does not hold by developing a novel attack technique: the Flush+Flush attack. The Flush+Flush attack only relies on the execution time of the flush instruction, which depends on whether data is cached or not. Flush+Flush does not make any memory accesses, contrary to any other cache attack. Thus, it causes no cache misses at all and the number of cache hits is reduced to a minimum due to the constant cache flushes. Therefore, Flush+Flush attacks are stealthy, i.e., the spy process cannot be detected based on cache hits and misses, or state-of-the-art detection mechanisms. The Flush+Flush attack runs in a higher frequency and thus is faster than any existing cache attack. With 496 KB s in a cross-core covert channel it is 6.7 times faster than any previously published cache covert channel."
]
} |
1511.04897 | 2469208367 | In the last 10 years, cache attacks on Intel x86 CPUs have gained increasing attention among the scientific community and powerful techniques to exploit cache side channels have been developed. However, modern smartphones use one or more multi-core ARM CPUs that have a different cache organization and instruction set than Intel x86 CPUs. So far, no cross-core cache attacks have been demonstrated on non-rooted Android smartphones. In this work, we demonstrate how to solve key challenges to perform the most powerful cross-core cache attacks Prime+Probe, Flush+Reload, Evict+Reload, and Flush+Flush on non-rooted ARM-based devices without any privileges. Based on our techniques, we demonstrate covert channels that outperform state-of-the-art covert channels on Android by several orders of magnitude. Moreover, we present attacks to monitor tap and swipe events as well as keystrokes, and even derive the lengths of words entered on the touchscreen. Eventually, we are the first to attack cryptographic primitives implemented in Java. Our attacks work across CPUs and can even monitor cache activity in the ARM TrustZone from the normal world. The techniques we present can be used to attack hundreds of millions of Android devices. | While the attacks discussed above have been proposed and investigated for Intel processors, the same attacks were considered not applicable to modern smartphones due to differences in the instruction set, the cache organization @cite_40 , and in the multi-core and multi-CPU architecture. Thus, only same-core cache attacks have been demonstrated on smartphones so far. For instance, Wei al @cite_4 investigated Bernstein 's cache-timing attack @cite_19 on a Beagleboard employing an ARM Cortex-A8 processor. Later on, Wei al @cite_54 investigated this timing attack in a multi-core setting on a development board. As Wei al @cite_4 claimed that noise makes the attack difficult, Spreitzer and Plos @cite_15 investigated the applicability of Bernstein's cache-timing attack on different ARM Cortex-A8 and ARM Cortex-A9 smartphones running Android. Both investigations @cite_4 @cite_15 confirmed that timing information is leaking, but the attack takes several hours due to the high number of measurement samples that are required, about @math AES encryptions. Later on, Spreitzer and G ' e rard @cite_41 improved upon these results and managed to reduce the key space to a complexity which is practically relevant. | {
"cite_N": [
"@cite_4",
"@cite_41",
"@cite_54",
"@cite_19",
"@cite_40",
"@cite_15"
],
"mid": [
"58308990",
"49522230",
"2404549987",
"1764679026",
"1427174644",
"2188789433"
],
"abstract": [
"We show in this paper that the isolation characteristic of system virtualization can be bypassed by the use of a cache timing attack. Using Bernstein’s correlation in this attack, an adversary is able to extract sensitive keying material from an isolated trusted execution domain. We demonstrate this cache timing attack on an embedded ARM-based platform running an L4 microkernel as virtualization layer. An attacker who gained access to the untrusted domain can extract the key of an AES-based authentication protocol used for a financial transaction. We provide measurements for different public domain AES implementations. Our results indicate that cache timing attacks are highly relevant in virtualization-based security architectures, such as trusted execution environments.",
"Side-channel attacks are usually performed by employing the \"divide-and-conquer\" approach, meaning that leaking information is collected in a divide step, and later on exploited in the conquer step. The idea is to extract as much information as possible during the divide step, and to exploit the gathered information as efficiently as possible within the conquer step. Focusing on both of these steps, we discuss potential enhancements of Bernstein's cache-timing attack against the Advanced Encryption Standard (AES). Concerning the divide part, we analyze the impact of attacking different key-chunk sizes, aiming at the extraction of more information from the overall encryption time. Furthermore, we analyze the most recent improvement of time-driven cache attacks, presented by Aly and ElGayyar, according to its applicability on ARM Cortex-A platforms. For the conquer part, we employ the optimal key-enumeration algorithm as proposed by Veyrat-Charvillonaet al to significantly reduce the complexity of the exhaustive key-search phase compared to the currently employed threshold-based approach. This in turn leads to more practical attacks. Additionally, we provide extensive experimental results of the proposed enhancements on two Android-based smartphones, namely a Google Nexus S and a Samsung Galaxy SII.",
"Virtualization has become one of the most important security enhancing techniques for embedded systems during the last years, both for mobile devices and cyber-physical system CPS. One of the major security threats in this context is posed by side channel attacks. In this work, Bernstein's time-driven cache-based attack against AES is revisited in a virtualization scenario based on an actual CPS using the PikeOS microkernel virtualization framework. The attack is conducted in the context of the implemented virtualization scenario using different scheduler configurations. We provide experimental results which show that using dedicated cores for crypto routines will have a high impact on the vulnerability of such systems. We also compare the results to previous work in that field and our visualization directly shows the differences between cache architectures of the ARM Cortex-A8 and Cortex-A9. Further, a non-invasive countermeasure against timing attacks based on the scheduler of PikeOS is devised, which in fact increases the system's security against cache timing attacks.",
"",
"Sharing memory pages between non-trusting processes is a common method of reducing the memory footprint of multi-tenanted systems. In this paper we demonstrate that, due to a weakness in the Intel X86 processors, page sharing exposes processes to information leaks. We present FLUSH+RELOAD, a cache side-channel attack technique that exploits this weakness to monitor access to memory lines in shared pages. Unlike previous cache side-channel attacks, FLUSH+RELOAD targets the Last-Level Cache (i.e. L3 on processors with three cache levels). Consequently, the attack program and the victim do not need to share the execution core. We demonstrate the efficacy of the FLUSH+RELOAD attack by using it to extract the private encryption keys from a victim program running GnuPG 1.4.13. We tested the attack both between two unrelated processes in a single operating system and between processes running in separate virtual machines. On average, the attack is able to recover 96.7 of the bits of the secret key by observing a single signature or decryption round.",
"Cache attacks are known to be sophisticated attacks against cryptographic implementations on desktop computers. Recently, investigations of such attacks on specific testbeds with processors that are employed in mobile devices have been done. In this work we investigate the applicability of Bernstein’s [2] timing attack and the cache-collision attack by Bogdanov ( ) [4] in real environments on three state-of-the-art mobile devices: an Acer Iconia A510, a Google Nexus S, and a Samsung Galaxy SIII. We show that T-table based implementations of the Advanced Encryption Standard (AES) leak enough timing information on these devices in order to recover parts of the used secret key using Bernstein’s timing attack. We also show that systems with a cache-line size larger than 32 bytes exacerbate the cache-collision attack of Bogdanov ( ) [4]."
]
} |
1511.04814 | 2949683196 | In this paper, we implement an application-aware scheduler that differentiates users running real-time applications and delay-tolerant applications while allocating resources. This approach ensures that the priority is given to real-time applications over delay-tolerant applications. In our system model, we include realistic channel effects of Long Term Evolution (LTE) system. Our application-aware scheduler runs in two stages, the first stage is resource block allocation and the second stage is power allocation. In the optimal solution of resource block allocation problem, each user is inherently guaranteed a minimum Quality of Experience (QoE) while ensuring priority given to users with real-time applications. In the power allocation problem, a new power allocation method is proposed which utilizes the optimal solution of the application-aware resource block scheduling problem. As a proof of concept, we run a simulation comparison between a conventional proportional fairness scheduler and the application-aware scheduler. The simulation results show better QoE with the application-aware scheduler. | In @cite_3 , the authors present a distributed approach for the joint allocation of modulation coding schemes, resource blocks, and power for LTE systems. Near optimal solutions are proposed to solve the optimization problem. In @cite_2 - @cite_12 , the authors present optimal rate allocation algorithms for users covered by a single carrier eNodeB. The authors use logarithmic and sigmoidal-like utility functions to represent delay-tolerant and real-time applications, respectively. In @cite_2 , the rate allocation algorithm gives priority to real-time applications over delay-tolerant applications when allocating resources. The proposed approach is not specifically designed for LTE systems. Moreover, no channel effects are included in the results. | {
"cite_N": [
"@cite_12",
"@cite_3",
"@cite_2"
],
"mid": [
"",
"2113643803",
"2013722958"
],
"abstract": [
"",
"This article investigates the problem of the allocation of modulation and coding, subcarriers and power to users in LTE. The proposed model achieves inter-cell interference mitigation through the dynamic and distributed self-organization of cells. Therefore, there is no need for any a prior frequency planning. Moreover, a two-level decomposition method able to find near optimal solutions is proposed to solve the optimization problem. Finally, simulation results show that compared to classic reuse schemes the proposed approach is able to pack more users into the same bandwidth, decreasing the probability of user outage.",
"In this paper, we introduce an approach for resource allocation of elastic and inelastic adaptive real-time traffic in fourth generation long term evolution (4G-LTE) system. In our model, we use logarithmic and sigmoidal-like utility functions to represent the users applications running on different user equipments (UE)s. We present a resource allocation optimization problem with utility proportional fairness policy, where the fairness among users is in utility percentage (i.e user satisfaction with the service) of the corresponding applications. Our objective is to allocate the resources to the users with priority given to the adaptive real-time application users. In addition, a minimum resource allocation for users with elastic and inelastic traffic should be guaranteed. Our goal is that every user subscribing for the mobile service should have a minimum quality-of-service (QoS) with a priority criterion. We prove that our resource allocation optimization problem is convex and therefore the optimal solution is tractable. We present a distributed algorithm to allocate evolved NodeB (eNodeB) resources optimally with a priority criterion. Finally, we present simulation results for the performance of our rate allocation algorithm."
]
} |
1511.04814 | 2949683196 | In this paper, we implement an application-aware scheduler that differentiates users running real-time applications and delay-tolerant applications while allocating resources. This approach ensures that the priority is given to real-time applications over delay-tolerant applications. In our system model, we include realistic channel effects of Long Term Evolution (LTE) system. Our application-aware scheduler runs in two stages, the first stage is resource block allocation and the second stage is power allocation. In the optimal solution of resource block allocation problem, each user is inherently guaranteed a minimum Quality of Experience (QoE) while ensuring priority given to users with real-time applications. In the power allocation problem, a new power allocation method is proposed which utilizes the optimal solution of the application-aware resource block scheduling problem. As a proof of concept, we run a simulation comparison between a conventional proportional fairness scheduler and the application-aware scheduler. The simulation results show better QoE with the application-aware scheduler. | In @cite_13 , a distributed solution of resource allocation for proportional fairness is provided for multi-band wireless systems. The proposed approach is not specific to the LTE systems. In @cite_1 , a distributed protocol that aims to achieve weighted proportional fairness by setting priority weights among UEs for LTE systems is presented. A resource block scheduling problem is formulated as a convex optimization problem. The weights play an important role while solving the optimization problem; however, the optimal resource scheduling which takes the user QoE into account is not guaranteed. All the UE applications are treated the same whenever the initial weights are set equal. A power allocation heuristic is also proposed in this paper. | {
"cite_N": [
"@cite_1",
"@cite_13"
],
"mid": [
"2038091796",
"2169551554"
],
"abstract": [
"We consider the problem of LTE network self organization and optimization of resource allocation. One particular challenge for LTE systems is that, by applying OFDMA, a transmission may use multiple resource blocks scheduled over the frequency and time. There are three key components involved in the resource allocation and network optimization: resource block scheduling, power control, and client association. We propose a distributed protocol that aims to achieve weighted proportional fairness (WPF) among clients by jointly consider them. The cross-layer design includes: (i) an optimal online policy for resource block scheduling, (ii) a heuristic for transmit power control, and (iii) a selfish strategy for client association. The proposed scheme only requires limited local information exchange and thus can be easily implemented for large networks. Simulation results have shown its effectiveness in both the system throughput and user fairness.",
"A challenging problem in multi-band multi-cell self-organized wireless systems, such as multi-channel Wi-Fi networks, femto pico cells in 3G 4G cellular networks, and more recent wireless networks over TV white spaces, is of distributed resource allocation. This involves four components: channel selection, client association, channel access, and client scheduling. In this paper, we present a unified framework for jointly addressing the four components with the global system objective of maximizing the clients throughput in a proportionally fair manner. Our formulation allows a natural dissociation of the problem into two sub-parts. We show that the first part, involving channel access and client scheduling, is convex and derive a distributed adaptation procedure for achieving Pareto-optimal solution. For the second part, involving channel selection and client association, we develop a Gibbs-sampler based approach for local adaptation to achieve the global objective, as well as derive fast greedy algorithms from it that achieve good solutions."
]
} |
1511.04814 | 2949683196 | In this paper, we implement an application-aware scheduler that differentiates users running real-time applications and delay-tolerant applications while allocating resources. This approach ensures that the priority is given to real-time applications over delay-tolerant applications. In our system model, we include realistic channel effects of Long Term Evolution (LTE) system. Our application-aware scheduler runs in two stages, the first stage is resource block allocation and the second stage is power allocation. In the optimal solution of resource block allocation problem, each user is inherently guaranteed a minimum Quality of Experience (QoE) while ensuring priority given to users with real-time applications. In the power allocation problem, a new power allocation method is proposed which utilizes the optimal solution of the application-aware resource block scheduling problem. As a proof of concept, we run a simulation comparison between a conventional proportional fairness scheduler and the application-aware scheduler. The simulation results show better QoE with the application-aware scheduler. | An optimal application-aware resource block scheduling is proposed in @cite_7 . Initially, eNodeB makes scheduling decisions for each client assuming that the power allocations are fixed. Utility functions are used to represent the user application types and they are incorporated into the objective function while formulating the optimization problem. This approach ensures that different throughput expectations of each user are considered while making the resource block scheduling decisions. | {
"cite_N": [
"@cite_7"
],
"mid": [
"2953083309"
],
"abstract": [
"In this paper, we introduce an approach for application-aware resource block scheduling of elastic and inelastic adaptive real-time traffic in fourth generation Long Term Evolution (LTE) systems. The users are assigned to resource blocks. A transmission may use multiple resource blocks scheduled over frequency and time. In our model, we use logarithmic and sigmoidal-like utility functions to represent the users applications running on different user equipments (UE)s. We present an optimal problem with utility proportional fairness policy, where the fairness among users is in utility percentage (i.e user satisfaction with the service) of the corresponding applications. Our objective is to allocate the resources to the users with priority given to the adaptive real-time application users. In addition, a minimum resource allocation for users with elastic and inelastic traffic should be guaranteed. Every user subscribing for the mobile service should have a minimum quality-of-service (QoS) with a priority criterion. We prove that our scheduling policy exists and achieves the maximum. Therefore the optimal solution is tractable. We present a centralized scheduling algorithm to allocate evolved NodeB (eNodeB) resources optimally with a priority criterion. Finally, we present simulation results for the performance of our scheduling algorithm and compare our results with conventional proportional fairness approaches. The results show that the user satisfaction is higher with our proposed method."
]
} |
1511.04664 | 2259263553 | Despite the widespread installation of accelerometers in almost all mobile phones and wearable devices, activity recognition using accelerometers is still immature due to the poor recognition accuracy of existing recognition methods and the scarcity of labeled training data. We consider the problem of human activity recognition using triaxial accelerometers and deep learning paradigms. This paper shows that deep activity recognition models (a) provide better recognition accuracy of human activities, (b) avoid the expensive design of handcrafted features in existing systems, and (c) utilize the massive unlabeled acceleration samples for unsupervised feature extraction. Moreover, a hybrid approach of deep learning and hidden Markov models (DL-HMM) is presented for sequential activity recognition. This hybrid approach integrates the hierarchical representations of deep activity recognition models with the stochastic modeling of temporal sequences in the hidden Markov models. We show substantial recognition improvement on real world datasets over state-of-the-art methods of human activity recognition using triaxial accelerometers. | In this section, we will focus on classification and feature engineering methods for activity recognition using accelerometers. For a more comprehensive review of the field, we refer interested readers to recent survey papers @cite_6 @cite_18 . | {
"cite_N": [
"@cite_18",
"@cite_6"
],
"mid": [
"2059732136",
"2054780155"
],
"abstract": [
"Research on sensor-based activity recognition has, recently, made significant progress and is attracting growing attention in a number of disciplines and application domains. However, there is a lack of high-level overview on this topic that can inform related communities of the research state of the art. In this paper, we present a comprehensive survey to examine the development and current status of various aspects of sensor-based activity recognition. We first discuss the general rationale and distinctions of vision-based and sensor-based activity recognition. Then, we review the major approaches and methods associated with sensor-based activity monitoring, modeling, and recognition from which strengths and weaknesses of those approaches are highlighted. We make a primary distinction in this paper between data-driven and knowledge-driven approaches, and use this distinction to structure our survey. We also discuss some promising directions for future research.",
"Providing accurate and opportune information on people's activities and behaviors is one of the most important tasks in pervasive computing. Innumerable applications can be visualized, for instance, in medical, security, entertainment, and tactical scenarios. Despite human activity recognition (HAR) being an active field for more than a decade, there are still key aspects that, if addressed, would constitute a significant turn in the way people interact with mobile devices. This paper surveys the state of the art in HAR based on wearable sensors. A general architecture is first presented along with a description of the main components of any HAR system. We also propose a two-level taxonomy in accordance to the learning approach (either supervised or semi-supervised) and the response time (either offline or online). Then, the principal issues and challenges are discussed, as well as the main solutions to each one of them. Twenty eight systems are qualitatively evaluated in terms of recognition performance, energy consumption, obtrusiveness, and flexibility, among others. Finally, we present some open problems and ideas that, due to their high relevance, should be addressed in future research."
]
} |
1511.04664 | 2259263553 | Despite the widespread installation of accelerometers in almost all mobile phones and wearable devices, activity recognition using accelerometers is still immature due to the poor recognition accuracy of existing recognition methods and the scarcity of labeled training data. We consider the problem of human activity recognition using triaxial accelerometers and deep learning paradigms. This paper shows that deep activity recognition models (a) provide better recognition accuracy of human activities, (b) avoid the expensive design of handcrafted features in existing systems, and (c) utilize the massive unlabeled acceleration samples for unsupervised feature extraction. Moreover, a hybrid approach of deep learning and hidden Markov models (DL-HMM) is presented for sequential activity recognition. This hybrid approach integrates the hierarchical representations of deep activity recognition models with the stochastic modeling of temporal sequences in the hidden Markov models. We show substantial recognition improvement on real world datasets over state-of-the-art methods of human activity recognition using triaxial accelerometers. | Machine learning algorithms have been used for a wide range of activity recognition applications @cite_0 @cite_2 @cite_9 @cite_22 , allowing the mapping between feature sets and various human activities. The classification of accelerometer samples into static and dynamic activities using MLPs is presented in @cite_2 . Conventional neural networks, including MLPs, often stuck in local optima @cite_10 which leads to poor performance of activity recognition systems. Moreover, training MLPs using backpropagation @cite_10 only hinders the addition of many hidden layers due to the vanishing gradient problem. The authors in @cite_0 used decision trees and MLPs to classify daily human activities. @cite_17 , a fuzzy inference system is designed to detect human activities. @cite_22 compared the recognition accuracy of decision tree (C4.5), logistic regression, and MLPs, where MLPs are found to outperform the other methods. In this paper, we show significant recognition accuracy improvement on real world datasets over state-of-the-art methods for human activity recognition using triaxial accelerometers. Additionally, even though some previous works have purportedly reported promising results of activity recognition accuracy, they still require a degree of handcrafted features as discussed below. | {
"cite_N": [
"@cite_22",
"@cite_9",
"@cite_0",
"@cite_2",
"@cite_10",
"@cite_17"
],
"mid": [
"2017634428",
"2610156087",
"2148217011",
"2168538666",
"1498436455",
"1980526801"
],
"abstract": [
"Mobile devices are becoming increasingly sophisticated and the latest generation of smart cell phones now incorporates many diverse and powerful sensors. These sensors include GPS sensors, vision sensors (i.e., cameras), audio sensors (i.e., microphones), light sensors, temperature sensors, direction sensors (i.e., magnetic compasses), and acceleration sensors (i.e., accelerometers). The availability of these sensors in mass-marketed communication devices creates exciting new opportunities for data mining and data mining applications. In this paper we describe and evaluate a system that uses phone-based accelerometers to perform activity recognition, a task which involves identifying the physical activity a user is performing. To implement our system we collected labeled accelerometer data from twenty-nine users as they performed daily activities such as walking, jogging, climbing stairs, sitting, and standing, and then aggregated this time series data into examples that summarize the user activity over 10- second intervals. We then used the resulting training data to induce a predictive model for activity recognition. This work is significant because the activity recognition model permits us to gain useful knowledge about the habits of millions of users passively---just by having them carry cell phones in their pockets. Our work has a wide range of applications, including automatic customization of the mobile device's behavior based upon a user's activity (e.g., sending calls directly to voicemail if a user is jogging) and generating a daily weekly activity profile to determine if a user (perhaps an obese child) is performing a healthy amount of exercise.",
"",
"Automatic classification of everyday activities can be used for promotion of health-enhancing physical activities and a healthier lifestyle. In this paper, methods used for classification of everyday activities like walking, running, and cycling are described. The aim of the study was to find out how to recognize activities, which sensors are useful and what kind of signal processing and classification is required. A large and realistic data library of sensor data was collected. Sixteen test persons took part in the data collection, resulting in approximately 31 h of annotated, 35-channel data recorded in an everyday environment. The test persons carried a set of wearable sensors while performing several activities during the 2-h measurement session. Classification results of three classifiers are shown: custom decision tree, automatically generated decision tree, and artificial neural network. The classification accuracies using leave-one-subject-out cross validation range from 58 to 97 for custom decision tree classifier, from 56 to 97 for automatically generated decision tree, and from 22 to 96 for artificial neural network. Total classification accuracy is 82 for custom decision tree classifier, 86 for automatically generated decision tree, and 82 for artificial neural network",
"Physical-activity recognition via wearable sensors can provide valuable information regarding an individual's degree of functional ability and lifestyle. In this paper, we present an accelerometer sensor-based approach for human-activity recognition. Our proposed recognition method uses a hierarchical scheme. At the lower level, the state to which an activity belongs, i.e., static, transition, or dynamic, is recognized by means of statistical signal features and artificial-neural nets (ANNs). The upper level recognition uses the autoregressive (AR) modeling of the acceleration signals, thus, incorporating the derived AR-coefficients along with the signal-magnitude area and tilt angle to form an augmented-feature vector. The resulting feature vector is further processed by the linear-discriminant analysis and ANNs to recognize a particular human activity. Our proposed activity-recognition method recognizes three states and 15 activities with an average accuracy of 97.9 using only a single triaxial accelerometer attached to the subject's chest.",
"We describe a new learning procedure, back-propagation, for networks of neurone-like units. The procedure repeatedly adjusts the weights of the connections in the network so as to minimize a measure of the difference between the actual output vector of the net and the desired output vector. As a result of the weight adjustments, internal ‘hidden’ units which are not part of the input or output come to represent important features of the task domain, and the regularities in the task are captured by the interactions of these units. The ability to create useful new features distinguishes back-propagation from earlier, simpler methods such as the perceptron-convergence procedure1.",
"Smart phones have become a powerful platform for wearable context recognition. We present a service-based recognition architecture which creates an evolving classification system using feedback from the user community. The approach utilizes classifiers based on fuzzy inference systems which use live annotation to personalize the classifier instance on the device. Our recognition system is designed for everyday use: it allows flexible placement of the device (no assumed or fixed position), requires only minimal personalization effort from the user (1–3 minutes per activity) and is capable of detecting a high number of activities. The components of the service are shown in an evaluation scenario, in which recognition rates up to 97 can be achieved for ten activity classes."
]
} |
1511.04762 | 2242543816 | In the Colored Bin Packing problem a set of items with varying weights and colors must be packed into bins of uniform weight limit such that no two items of the same color may be packed adjacently within a bin. We solve this problem for the case where there are two or more colors when the items have zero weight and when the items have unit weight. Our algorithms are optimal and run in linear time. Since our algorithms apply for two or more colors, they demonstrate that the problem does not get harder as the number of colors increases. We also provide closed-form expressions for the optimal number of bins. | study a version of Colored Bin Packing where items have weights and colors and bins have weight and color constraints, specifically no more than @math items of the same color may be packed in a bin @cite_7 . They prove upper and lower bounds for several variants of this problem in both the offline and online settings. A more generalized version of constrained bin packing where the constraints among items are defined in a conflict graph have been studied by Jansen @cite_8 and @cite_4 . | {
"cite_N": [
"@cite_4",
"@cite_7",
"@cite_8"
],
"mid": [
"2048947076",
"2050319410",
"1847958188"
],
"abstract": [
"We consider a particular bin packing problem in which some pairs of items may be in conflict and cannot be assigned to the same bin. The problem, denoted as the bin packing problem with conflicts, is of practical and theoretical interest because of its many real-world applications and because it generalizes both the bin packing problem and the vertex coloring problem. We present new lower bounds, upper bounds, and an exact approach, based on a set covering formulation solved through a branch-and-price algorithm. We investigate the behavior of the proposed procedures by means of extensive computational results on benchmark instances from the literature.",
"We study the following variant of the bin packing problem. We are given a set of items, where each item has a (non-negative) size and a color. We are also given an integer parameter k, and the goal is to partition the items into a minimum number of subsets such that for each subset S in the solution, the total size of the items in S is at most 1 (as in the classical bin packing problem) and the total number of colors of the items in S is at most k (which distinguishes our problem from the classical version). We follow earlier work on this problem and study the problem in both offline and online scenarios.",
"In this paper we consider the following bin packing problem with conflicts. Given a set of items V = 1,...,n with sizes s 1 ,...,s n ∈ (0,1) and a conflict graph G = (V, E), we consider the problem to find a packing for the items into bins of size one such that adjacent items (j, j') ∈ E are assigned to different bins. The goal is to find an assignment with a minimum number of bins. This problem is a natural generalization of the classical bin packing problem. We propose an asymptotic approximation scheme for the bin packing problem with conflicts restricted to d-inductive graphs with constant d. This graph class contains trees, grid graphs, planar graphs and graphs with constant treewidth. The algorithm finds an assignment for the items such that the generated number of bins is within a factor of (1 + ∈) of optimal provided that the optimum number is sufficiently large. The running time of the algorithm is polynomial both in n and in 1 2."
]
} |
1511.04762 | 2242543816 | In the Colored Bin Packing problem a set of items with varying weights and colors must be packed into bins of uniform weight limit such that no two items of the same color may be packed adjacently within a bin. We solve this problem for the case where there are two or more colors when the items have zero weight and when the items have unit weight. Our algorithms are optimal and run in linear time. Since our algorithms apply for two or more colors, they demonstrate that the problem does not get harder as the number of colors increases. We also provide closed-form expressions for the optimal number of bins. | In this paper, we investigate Colored Bin Packing with constraints - i.e. two items with the same color cannot be packed adjacently to each other in a bin. This 2-color variant was first introduced by , who studied both the offline and online versions of the problem @cite_2 . For the offline version the authors presented a 2.5-approximation algorithm. For the online version, they proved that classical algorithms (First, Best, Worst, Next, Harmonic) are not constant competitive, and hence do not perform well. They further proved a universal lower bound of approximately 1.7213. This exceeds classical Bin Packing's upper bound of 1.5889, proving that Colored Bin Packing with alternation is harder @cite_1 . The authors' main result was a 3-competitive online algorithm, . The main concept of the algorithm is quite elegant. Using pseudo", i.e. dummy, items of different color and zero weight, their algorithm creates an acceptable sequence of colors that are later substituted with actually released items. If any bin overflows, Pseudo redistributes the extra items using Any Fit into available bins, and creates new bins if necessary. | {
"cite_N": [
"@cite_1",
"@cite_2"
],
"mid": [
"2078659331",
"2177689085"
],
"abstract": [
"A new framework for analyzing online bin packing algorithms is presented. This framework presents a unified way of explaining the performance of algorithms based on the Harmonic approach. Within this framework, it is shown that a new algorithm, Harmonic++, has asymptotic performance ratio at most 1.58889. It is also shown that the analysis of Harmonic+1 presented in Richey [1991] is incorrect; this is a fundamental logical flaw, not an error in calculation or an omitted case. The asymptotic performance ratio of Harmonic+1 is at least 1.59217. Thus, Harmonic++ provides the best upper bound for the online bin packing problem to date.",
"We introduce the following version of bin packing. The items are of two types (black and white), and in each bin the item types must alternate. We mostly investigate the online scenario. We study the competitiveness of some classical algorithms (First Best Worst Next Fit, Harmonic) — they do not perform very well — and for all online algorithms we also prove the universal lower bound (1+ 1 2 .7213 ) which significantly exceeds the known upper bound 1.58889 on classical online bin packing. We also design an online algorithm which is 3-competitive in the absolute sense. A 2.5-approximation algorithm and an APTAS is also given for the offline version."
]
} |
1511.04762 | 2242543816 | In the Colored Bin Packing problem a set of items with varying weights and colors must be packed into bins of uniform weight limit such that no two items of the same color may be packed adjacently within a bin. We solve this problem for the case where there are two or more colors when the items have zero weight and when the items have unit weight. Our algorithms are optimal and run in linear time. Since our algorithms apply for two or more colors, they demonstrate that the problem does not get harder as the number of colors increases. We also provide closed-form expressions for the optimal number of bins. | explored a zero-weight version of the 2-color variant of this problem @cite_0 . They show that in this case, the optimal number of bins equals the color - the absolute difference between the number of items of the two colors. The paper's main result is an optimal algorithm, named Balancing Any Fit, which operates on the simple principle that an item should be packed in an available bin containing the most oppositely-colored items. This is made easier since weight is not an issue. | {
"cite_N": [
"@cite_0"
],
"mid": [
"1781979225"
],
"abstract": [
"In the Colored Bin Packing problem a sequence of items of sizes up to (1 ) arrives to be packed into bins of unit capacity. Each item has one of (c 2 ) colors and an additional constraint is that we cannot pack two items of the same color next to each other in the same bin. The objective is to minimize the number of bins."
]
} |
1511.04762 | 2242543816 | In the Colored Bin Packing problem a set of items with varying weights and colors must be packed into bins of uniform weight limit such that no two items of the same color may be packed adjacently within a bin. We solve this problem for the case where there are two or more colors when the items have zero weight and when the items have unit weight. Our algorithms are optimal and run in linear time. Since our algorithms apply for two or more colors, they demonstrate that the problem does not get harder as the number of colors increases. We also provide closed-form expressions for the optimal number of bins. | Dosa and Epstein @cite_5 extend the results from @cite_2 . They prove that online Colored Bin Packing with alternation for 3 or more colors is harder than the 2-color version. Furthermore, the authors show that 2-color algorithms do not apply to 3 or more color problems. Lastly, they present a 4-competitive algorithm for the problem. This is a modified version of the earlier Pseudo, appropriately named Balanced-Pseudo. Instead of randomly assigning items to available bins, Balanced-Pseudo assigns an item to the bin whose top color is the most frequent. | {
"cite_N": [
"@cite_5",
"@cite_2"
],
"mid": [
"166203068",
"2177689085"
],
"abstract": [
"We study a variant of online bin packing, called colorful bin packing. In this problem, items that are presented one by one are to be packed into bins of size 1. Each item i has a size s i ∈ [0,1] and a color (c_i C ), where ( C ) is a set of colors (that is not necessarily known in advance). The total size of items packed into a bin cannot exceed its size, thus an item i can always be packed into a new bin, but an item cannot be packed into a non-empty bin if the previous item packed into that bin has the same color, or if the occupied space in it is larger than 1 − s i . This problem generalizes standard online bin packing and online black and white bin packing (where (| C |=2 )). We prove that colorful bin packing is harder than black and white bin packing in the sense that an online algorithm for zero size items that packs the input into the smallest possible number of bins cannot exist for (| C | 3 ), while it is known that such an algorithm exists for (| C |=2 ). We show that natural generalizations of classic algorithms for bin packing fail to work for the case (| C | 3 ), and moreover, algorithms that perform well for black and white bin packing do not perform well either, already for the case (| C |=3 ). Our main results are a new algorithm for colorful bin packing that we design and analyze, whose absolute competitive ratio is 4, and a new lower bound of 2 on the asymptotic competitive ratio of any algorithm, that is valid even for black and white bin packing.",
"We introduce the following version of bin packing. The items are of two types (black and white), and in each bin the item types must alternate. We mostly investigate the online scenario. We study the competitiveness of some classical algorithms (First Best Worst Next Fit, Harmonic) — they do not perform very well — and for all online algorithms we also prove the universal lower bound (1+ 1 2 .7213 ) which significantly exceeds the known upper bound 1.58889 on classical online bin packing. We also design an online algorithm which is 3-competitive in the absolute sense. A 2.5-approximation algorithm and an APTAS is also given for the offline version."
]
} |
1511.04762 | 2242543816 | In the Colored Bin Packing problem a set of items with varying weights and colors must be packed into bins of uniform weight limit such that no two items of the same color may be packed adjacently within a bin. We solve this problem for the case where there are two or more colors when the items have zero weight and when the items have unit weight. Our algorithms are optimal and run in linear time. Since our algorithms apply for two or more colors, they demonstrate that the problem does not get harder as the number of colors increases. We also provide closed-form expressions for the optimal number of bins. | It was proven in @cite_5 , that no competitive algorithm exists for online Colored Bin Packing when there are more than two colors even when the items have zero-weight. Therefore, in this paper we consider offline Colored Bin Packing when there are more than two colors. We present optimal linear time algorithms that solve this problem for two or more colors when the items have zero-weight and when the items have unit-weight. Since our algorithms are optimal for two or more colors, they demonstrate that the problem does not get harder as the number of colors increases. | {
"cite_N": [
"@cite_5"
],
"mid": [
"166203068"
],
"abstract": [
"We study a variant of online bin packing, called colorful bin packing. In this problem, items that are presented one by one are to be packed into bins of size 1. Each item i has a size s i ∈ [0,1] and a color (c_i C ), where ( C ) is a set of colors (that is not necessarily known in advance). The total size of items packed into a bin cannot exceed its size, thus an item i can always be packed into a new bin, but an item cannot be packed into a non-empty bin if the previous item packed into that bin has the same color, or if the occupied space in it is larger than 1 − s i . This problem generalizes standard online bin packing and online black and white bin packing (where (| C |=2 )). We prove that colorful bin packing is harder than black and white bin packing in the sense that an online algorithm for zero size items that packs the input into the smallest possible number of bins cannot exist for (| C | 3 ), while it is known that such an algorithm exists for (| C |=2 ). We show that natural generalizations of classic algorithms for bin packing fail to work for the case (| C | 3 ), and moreover, algorithms that perform well for black and white bin packing do not perform well either, already for the case (| C |=3 ). Our main results are a new algorithm for colorful bin packing that we design and analyze, whose absolute competitive ratio is 4, and a new lower bound of 2 on the asymptotic competitive ratio of any algorithm, that is valid even for black and white bin packing."
]
} |
1511.04510 | 2179259799 | Semantic object parsing is a fundamental task for understanding objects in detail in computer vision community, where incorporating multi-level contextual information is critical for achieving such fine-grained pixel-level recognition. Prior methods often leverage the contextual information through post-processing predicted confidence maps. In this work, we propose a novel deep Local-Global Long Short-Term Memory (LG-LSTM) architecture to seamlessly incorporate short-distance and long-distance spatial dependencies into the feature learning over all pixel positions. In each LG-LSTM layer, local guidance from neighboring positions and global guidance from the whole image are imposed on each position to better exploit complex local and global contextual information. Individual LSTMs for distinct spatial dimensions are also utilized to intrinsically capture various spatial layouts of semantic parts in the images, yielding distinct hidden and memory cells of each position for each dimension. In our parsing approach, several LG-LSTM layers are stacked and appended to the intermediate convolutional layers to directly enhance visual features, allowing network parameters to be learned in an end-to-end way. The long chains of sequential computation by stacked LG-LSTM layers also enable each pixel to sense a much larger region for inference benefiting from the memorization of previous dependencies in all positions along all dimensions. Comprehensive evaluations on three public datasets well demonstrate the significant superiority of our LG-LSTM over other state-of-the-art methods. | There have been increasing research works in the semantic object parsing problem including the general object parsing @cite_2 @cite_24 @cite_5 @cite_30 @cite_37 and human parsing @cite_20 @cite_0 @cite_17 @cite_4 @cite_7 @cite_14 . Recent state-of-the-art approaches often rely on deep convolutional neural networks along with advanced network architectures @cite_37 @cite_24 @cite_21 . Instead of learning features only from local convolutional kernels as in these previous methods, we solve this problem through incorporating the novel LG-LSTM layers into CNNs to capture both long-distance and short-distance spatial dependencies. In addition, the adoption of hidden and memory cells in LSTMs makes it possible to memorize the previous contextual interactions from local neighboring positions and the whole image in previous LG-LSTM layers. It should be noted that while @cite_25 models mean-field approximate inference as recurrent networks, it can only refine results based on predicted pixel-wise confidence maps. In contrast, our LG-LSTM layers can progressively improve visual features to directly boost the performance of object parsing. | {
"cite_N": [
"@cite_30",
"@cite_37",
"@cite_14",
"@cite_4",
"@cite_7",
"@cite_21",
"@cite_24",
"@cite_0",
"@cite_2",
"@cite_5",
"@cite_25",
"@cite_20",
"@cite_17"
],
"mid": [
"2104408738",
"1948751323",
"",
"2094311777",
"1195044660",
"2204578866",
"792160549",
"2121339428",
"1903370114",
"2020206277",
"",
"2074621908",
"2010785631"
],
"abstract": [
"Detecting objects becomes difficult when we need to deal with large shape deformation, occlusion and low resolution. We propose a novel approach to i) handle large deformations and partial occlusions in animals (as examples of highly deformable objects), ii) describe them in terms of body parts, and iii) detect them when their body parts are hard to detect (e.g., animals depicted at low resolution). We represent the holistic object and body parts separately and use a fully connected model to arrange templates for the holistic object and body parts. Our model automatically decouples the holistic object or body parts from the model when they are hard to detect. This enables us to represent a large number of holistic object and body part combinations to better deal with different \"detectability\" patterns caused by deformations, occlusion and or low resolution. We apply our method to the six animal categories in the PASCAL VOC dataset and show that our method significantly improves state-of-the-art (by 4.1 AP) and provides a richer representation for objects. During training we use annotations for body parts (e.g., head, torso, etc.), making use of a new dataset of fully annotated object parts for PASCAL VOC 2010, which provides a mask for each part.",
"Recognition algorithms based on convolutional networks (CNNs) typically use the output of the last layer as a feature representation. However, the information in this layer may be too coarse spatially to allow precise localization. On the contrary, earlier layers may be precise in localization but will not capture semantics. To get the best of both worlds, we define the hypercolumn at a pixel as the vector of activations of all CNN units above that pixel. Using hypercolumns as pixel descriptors, we show results on three fine-grained localization tasks: simultaneous detection and segmentation [22], where we improve state-of-the-art from 49.7 mean APr [22] to 60.0, keypoint localization, where we get a 3.3 point boost over [20], and part labeling, where we show a 6.6 point gain over a strong baseline.",
"",
"Clothing is one of the most informative cues of human appearance. In this paper, we propose a novel multi-person clothing segmentation algorithm for highly occluded images. The key idea is combining blocking models to address the person-wise occlusions. In contrary to the traditional layered model that tries to solve the full layer ranking problem, the proposed blocking model partitions the problem into a series of pair-wise ones and then determines the local blocking relationship based on individual and contextual information. Thus, it is capable of dealing with cases with a large number of people. Additionally, we propose a layout model formulated as Markov Network which incorporates the blocking relationship to pursue an approximately optimal clothing layout for group people. Experiments demonstrated on a group images dataset show the effectiveness of our algorithm.",
"In this paper we tackle the problem of clothing parsing: Our goal is to segment and classify different garments a person is wearing. We frame the problem as the one of inference in a pose-aware Conditional Random Field (CRF) which exploits appearance, figure ground segmentation, shape and location priors for each garment as well as similarities between segments, and symmetries between different human body parts. We demonstrate the effectiveness of our approach on the Fashionista dataset [1] and show that we can obtain a significant improvement over the state-of-the-art.",
"In this work, we address the human parsing task with a novel Contextualized Convolutional Neural Network (Co-CNN) architecture, which well integrates the cross-layer context, global image-level context, within-super-pixel context and cross-super-pixel neighborhood context into a unified network. Given an input human image, Co-CNN produces the pixel-wise categorization in an end-to-end way. First, the cross-layer context is captured by our basic local-to-global-to-local structure, which hierarchically combines the global semantic structure and the local fine details within the cross-layers. Second, the global image-level label prediction is used as an auxiliary objective in the intermediate layer of the Co-CNN, and its outputs are further used for guiding the feature learning in subsequent convolutional layers to leverage the global image-level context. Finally, to further utilize the local super-pixel contexts, the within-super-pixel smoothing and cross-super-pixel neighbourhood voting are formulated as natural sub-components of the Co-CNN to achieve the local label consistency in both training and testing process. Comprehensive evaluations on two public datasets well demonstrate the significant superiority of our Co-CNN architecture over other state-of-the-arts for human parsing. In particular, the F-1 score on the large dataset [15] reaches 76.95 by Co-CNN, significantly higher than 62.81 and 64.38 by the state-of-the-art algorithms, M-CNN [21] and ATR [15], respectively.",
"Segmenting semantic objects from images and parsing them into their respective semantic parts are fundamental steps towards detailed object understanding in computer vision. In this paper, we propose a joint solution that tackles semantic object and part segmentation simultaneously, in which higher object-level context is provided to guide part segmentation, and more detailed part-level localization is utilized to refine object segmentation. Specifically, we first introduce the concept of semantic compositional parts (SCP) in which similar semantic parts are grouped and shared among different objects. A two-channel fully convolutional network (FCN) is then trained to provide the SCP and object potentials at each pixel. At the same time, a compact set of segments can also be obtained from the SCP predictions of the network. Given the potentials and the generated segments, in order to explore long-range context, we finally construct an efficient fully connected conditional random field (FCRF) to jointly predict the final object and part labels. Extensive evaluation on three different datasets shows that our approach can mutually enhance the performance of object and part segmentation, and outperforms the current state-of-the-art on both tasks.",
"Clothing recognition is an extremely challenging problem due to wide variation in clothing item appearance, layering, and style. In this paper, we tackle the clothing parsing problem using a retrieval based approach. For a query image, we find similar styles from a large database of tagged fashion images and use these examples to parse the query. Our approach combines parsing from: pre-trained global clothing models, local clothing models learned on the fly from retrieved examples, and transferred parse masks (paper doll item transfer) from retrieved examples. Experimental evaluation shows that our approach significantly outperforms state of the art in parsing accuracy.",
"In this paper, we study the problem of semantic part segmentation for animals. This is more challenging than standard object detection, object segmentation and pose estimation tasks because semantic parts of animals often have similar appearance and highly varying shapes. To tackle these challenges, we build a mixture of compositional models to represent the object boundary and the boundaries of semantic parts. And we incorporate edge, appearance, and semantic part cues into the compositional model. Given part-level segmentation annotation, we develop a novel algorithm to learn a mixture of compositional models under various poses and viewpoints for certain animal classes. Furthermore, a linear complexity algorithm is offered for efficient inference of the compositional model using dynamic programming. We evaluate our method for horse and cow using a newly annotated dataset on Pascal VOC 2010 which has pixelwise part labels. Experimental results demonstrate the effectiveness of our method.",
"This paper addresses the problem of semantic part parsing (segmentation) of cars, i.e.assigning every pixel within the car to one of the parts (e.g.body, window, lights, license plates and wheels). We formulate this as a landmark identification problem, where a set of landmarks specifies the boundaries of the parts. A novel mixture of graphical models is proposed, which dynamically couples the landmarks to a hierarchy of segments. When modeling pairwise relation between landmarks, this coupling enables our model to exploit the local image contents in addition to spatial deformation, an aspect that most existing graphical models ignore. In particular, our model enforces appearance consistency between segments within the same part. Parsing the car, including finding the optimal coupling between landmarks and segments in the hierarchy, is performed by dynamic programming. We evaluate our method on a subset of PASCAL VOC 2010 car images and on the car subset of 3D Object Category dataset (CAR3D). We show good results and, in particular, quantify the effectiveness of using the segment appearance consistency in terms of accuracy of part localization and segmentation.",
"",
"In this paper we demonstrate an effective method for parsing clothing in fashion photographs, an extremely challenging problem due to the large number of possible garment items, variations in configuration, garment appearance, layering, and occlusion. In addition, we provide a large novel dataset and tools for labeling garment items, to enable future research on clothing estimation. Finally, we present intriguing initial results on using clothing estimates to improve pose identification, and demonstrate a prototype application for pose-independent visual garment retrieval.",
"In this work, we address the problem of human parsing, namely partitioning the human body into semantic regions, by using the novel Parselet representation. Previous works often consider solving the problem of human pose estimation as the prerequisite of human parsing. We argue that these approaches cannot obtain optimal pixel level parsing due to the inconsistent targets between these tasks. In this paper, we propose to use Parselets as the building blocks of our parsing model. Parselets are a group of parsable segments which can generally be obtained by low-level over-segmentation algorithms and bear strong semantic meaning. We then build a Deformable Mixture Parsing Model (DMPM) for human parsing to simultaneously handle the deformation and multi-modalities of Parselets. The proposed model has two unique characteristics: (1) the possible numerous modalities of Parse let ensembles are exhibited as the And-Or\" structure of sub-trees, (2) to further solve the practical problem of Parselet occlusion or absence, we directly model the visibility property at some leaf nodes. The DMPM thus directly solves the problem of human parsing by searching for the best graph configuration from a pool of Parse let hypotheses without intermediate tasks. Comprehensive evaluations demonstrate the encouraging performance of the proposed approach."
]
} |
1511.04517 | 2952302801 | In this work, we propose a novel Reversible Recursive Instance-level Object Segmentation (R2-IOS) framework to address the challenging instance-level object segmentation task. R2-IOS consists of a reversible proposal refinement sub-network that predicts bounding box offsets for refining the object proposal locations, and an instance-level segmentation sub-network that generates the foreground mask of the dominant object instance in each proposal. By being recursive, R2-IOS iteratively optimizes the two sub-networks during joint training, in which the refined object proposals and improved segmentation predictions are alternately fed into each other to progressively increase the network capabilities. By being reversible, the proposal refinement sub-network adaptively determines an optimal number of refinement iterations required for each proposal during both training and testing. Furthermore, to handle multiple overlapped instances within a proposal, an instance-aware denoising autoencoder is introduced into the segmentation sub-network to distinguish the dominant object from other distracting instances. Extensive experiments on the challenging PASCAL VOC 2012 benchmark well demonstrate the superiority of R2-IOS over other state-of-the-art methods. In particular, the @math over @math classes at @math IoU achieves @math , which significantly outperforms the results of @math by PFN PFN and @math by liu2015multi . | Object detection aims to recognize and localize each object instance with a bounding box. Generally, most of the detection pipelines @cite_8 @cite_22 @cite_5 @cite_18 @cite_20 begin with producing object proposals from the input image, and then the classification and the bounding box regression are performed to identify the target objects. Many hand-designed approaches such as selective search @cite_15 , Edge Boxes @cite_16 and MCG @cite_21 , or CNN-based methods such as DeepMask @cite_26 and RPN @cite_8 have been proposed for object proposal extraction. Those detection approaches often treat the proposal generation and object detection as two separate techniques, yielding suboptimal results. In contrast, the proposed R2-IOS adaptively learns the optimal number of refinement iterations for each object proposal. Meanwhile, the reversible proposal refinement and instance-level segmentation sub-networks are jointly trained to mutually boost each other. | {
"cite_N": [
"@cite_18",
"@cite_26",
"@cite_22",
"@cite_8",
"@cite_21",
"@cite_5",
"@cite_15",
"@cite_16",
"@cite_20"
],
"mid": [
"2212123867",
"809122546",
"",
"2953106684",
"1991367009",
"1932624639",
"2088049833",
"7746136",
"2102605133"
],
"abstract": [
"Intuitive observations show that a baby may inherently possess the capability of recognizing a new visual concept (e.g., chair, dog) by learning from only very few positive instances taught by parent(s) or others, and this recognition capability can be gradually further improved by exploring and or interacting with the real instances in the physical world. Inspired by these observations, we propose a computational model for weakly-supervised object detection, based on prior knowledge modelling, exemplar learning and learning with video contexts. The prior knowledge is modeled with a pre-trained Convolutional Neural Network (CNN). When very few instances of a new concept are given, an initial concept detector is built by exemplar learning over the deep features the pre-trained CNN. The well-designed tracking solution is then used to discover more diverse instances from the massive online weakly labeled videos. Once a positive instance is detected identified with high score in each video, more instances possibly from different view-angles and or different distances are tracked and accumulated. Then the concept detector can be fine-tuned based on these new instances. This process can be repeated again and again till we obtain a very mature concept detector. Extensive experiments on Pascal VOC-07 10 12 object detection datasets [9] well demonstrate the effectiveness of our framework. It can beat the state-of-the-art full-training based performances by learning from very few samples for each object category, along with about 20,000 weakly labeled videos.",
"Recent object detection systems rely on two critical steps: (1) a set of object proposals is predicted as efficiently as possible, and (2) this set of candidate proposals is then passed to an object classifier. Such approaches have been shown they can be fast, while achieving the state of the art in detection performance. In this paper, we propose a new way to generate object proposals, introducing an approach based on a discriminative convolutional network. Our model is trained jointly with two objectives: given an image patch, the first part of the system outputs a class-agnostic segmentation mask, while the second part of the system outputs the likelihood of the patch being centered on a full object. At test time, the model is efficiently applied on the whole test image and generates a set of segmentation masks, each of them being assigned with a corresponding object likelihood score. We show that our model yields significant improvements over state-of-the-art object proposal algorithms. In particular, compared to previous approaches, our model obtains substantially higher object recall using fewer proposals. We also show that our model is able to generalize to unseen categories it has not seen during training. Unlike all previous approaches for generating object masks, we do not rely on edges, superpixels, or any other form of low-level segmentation.",
"",
"State-of-the-art object detection networks depend on region proposal algorithms to hypothesize object locations. Advances like SPPnet and Fast R-CNN have reduced the running time of these detection networks, exposing region proposal computation as a bottleneck. In this work, we introduce a Region Proposal Network (RPN) that shares full-image convolutional features with the detection network, thus enabling nearly cost-free region proposals. An RPN is a fully convolutional network that simultaneously predicts object bounds and objectness scores at each position. The RPN is trained end-to-end to generate high-quality region proposals, which are used by Fast R-CNN for detection. We further merge RPN and Fast R-CNN into a single network by sharing their convolutional features---using the recently popular terminology of neural networks with 'attention' mechanisms, the RPN component tells the unified network where to look. For the very deep VGG-16 model, our detection system has a frame rate of 5fps (including all steps) on a GPU, while achieving state-of-the-art object detection accuracy on PASCAL VOC 2007, 2012, and MS COCO datasets with only 300 proposals per image. In ILSVRC and COCO 2015 competitions, Faster R-CNN and RPN are the foundations of the 1st-place winning entries in several tracks. Code has been made publicly available.",
"We propose a unified approach for bottom-up hierarchical image segmentation and object candidate generation for recognition, called Multiscale Combinatorial Grouping (MCG). For this purpose, we first develop a fast normalized cuts algorithm. We then propose a high-performance hierarchical segmenter that makes effective use of multiscale information. Finally, we propose a grouping strategy that combines our multiscale regions into highly-accurate object candidates by exploring efficiently their combinatorial space. We conduct extensive experiments on both the BSDS500 and on the PASCAL 2012 segmentation datasets, showing that MCG produces state-of-the-art contours, hierarchical regions and object candidates.",
"We propose an object detection system that relies on a multi-region deep convolutional neural network (CNN) that also encodes semantic segmentation-aware features. The resulting CNN-based representation aims at capturing a diverse set of discriminative appearance factors and exhibits localization sensitivity that is essential for accurate object localization. We exploit the above properties of our recognition module by integrating it on an iterative localization mechanism that alternates between scoring a box proposal and refining its location with a deep CNN regression model. Thanks to the efficient use of our modules, we detect objects with very high localization accuracy. On the detection challenges of PASCAL VOC2007 and PASCAL VOC2012 we achieve mAP of 78.2 and 73.9 correspondingly, surpassing any other published work by a significant margin.",
"This paper addresses the problem of generating possible object locations for use in object recognition. We introduce selective search which combines the strength of both an exhaustive search and segmentation. Like segmentation, we use the image structure to guide our sampling process. Like exhaustive search, we aim to capture all possible object locations. Instead of a single technique to generate possible object locations, we diversify our search and use a variety of complementary image partitionings to deal with as many image conditions as possible. Our selective search results in a small set of data-driven, class-independent, high quality locations, yielding 99 recall and a Mean Average Best Overlap of 0.879 at 10,097 locations. The reduced number of locations compared to an exhaustive search enables the use of stronger machine learning techniques and stronger appearance models for object recognition. In this paper we show that our selective search enables the use of the powerful Bag-of-Words model for recognition. The selective search software is made publicly available (Software: http: disi.unitn.it uijlings SelectiveSearch.html ).",
"The use of object proposals is an effective recent approach for increasing the computational efficiency of object detection. We propose a novel method for generating object bounding box proposals using edges. Edges provide a sparse yet informative representation of an image. Our main observation is that the number of contours that are wholly contained in a bounding box is indicative of the likelihood of the box containing an object. We propose a simple box objectness score that measures the number of edges that exist in the box minus those that are members of contours that overlap the box’s boundary. Using efficient data structures, millions of candidate boxes can be evaluated in a fraction of a second, returning a ranked set of a few thousand top-scoring proposals. Using standard metrics, we show results that are significantly more accurate than the current state-of-the-art while being faster to compute. In particular, given just 1000 proposals we achieve over 96 object recall at overlap threshold of 0.5 and over 75 recall at the more challenging overlap of 0.7. Our approach runs in 0.25 seconds and we additionally demonstrate a near real-time variant with only minor loss in accuracy.",
"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."
]
} |
1511.04512 | 2952567519 | Zero-shot recognition (ZSR) deals with the problem of predicting class labels for target domain instances based on source domain side information (e.g. attributes) of unseen classes. We formulate ZSR as a binary prediction problem. Our resulting classifier is class-independent. It takes an arbitrary pair of source and target domain instances as input and predicts whether or not they come from the same class, i.e. whether there is a match. We model the posterior probability of a match since it is a sufficient statistic and propose a latent probabilistic model in this context. We develop a joint discriminative learning framework based on dictionary learning to jointly learn the parameters of our model for both domains, which ultimately leads to our class-independent classifier. Many of the existing embedding methods can be viewed as special cases of our probabilistic model. On ZSR our method shows 4.90 improvement over the state-of-the-art in accuracy averaged across four benchmark datasets. We also adapt ZSR method for zero-shot retrieval and show 22.45 improvement accordingly in mean average precision (mAP). | ( 4 ) Other methods: Less related to our method includes approaches based on semantic transfer propagation @cite_40 , transductive multi-view embedding @cite_36 , random forest approach @cite_14 , and semantic manifold distance @cite_32 . | {
"cite_N": [
"@cite_36",
"@cite_40",
"@cite_14",
"@cite_32"
],
"mid": [
"43954826",
"2151575489",
"2950497525",
"1960364170"
],
"abstract": [
"Most existing zero-shot learning approaches exploit transfer learning via an intermediate-level semantic representation such as visual attributes or semantic word vectors. Such a semantic representation is shared between an annotated auxiliary dataset and a target dataset with no annotation. A projection from a low-level feature space to the semantic space is learned from the auxiliary dataset and is applied without adaptation to the target dataset. In this paper we identify an inherent limitation with this approach. That is, due to having disjoint and potentially unrelated classes, the projection functions learned from the auxiliary dataset domain are biased when applied directly to the target dataset domain. We call this problem the projection domain shift problem and propose a novel framework, transductive multi-view embedding, to solve it. It is ‘transductive’ in that unlabelled target data points are explored for projection adaptation, and ‘multi-view’ in that both low-level feature (view) and multiple semantic representations (views) are embedded to rectify the projection shift. We demonstrate through extensive experiments that our framework (1) rectifies the projection shift between the auxiliary and target domains, (2) exploits the complementarity of multiple semantic representations, (3) achieves state-of-the-art recognition results on image and video benchmark datasets, and (4) enables novel cross-view annotation tasks.",
"Category models for objects or activities typically rely on supervised learning requiring sufficiently large training sets. Transferring knowledge from known categories to novel classes with no or only a few labels is far less researched even though it is a common scenario. In this work, we extend transfer learning with semi-supervised learning to exploit unlabeled instances of (novel) categories with no or only a few labeled instances. Our proposed approach Propagated Semantic Transfer combines three techniques. First, we transfer information from known to novel categories by incorporating external knowledge, such as linguistic or expert-specified information, e.g., by a mid-level layer of semantic attributes. Second, we exploit the manifold structure of novel classes. More specifically we adapt a graph-based learning algorithm - so far only used for semi-supervised learning -to zero-shot and few-shot learning. Third, we improve the local neighborhood in such graph structures by replacing the raw feature-based representation with a mid-level object- or attribute-based representation. We evaluate our approach on three challenging datasets in two different applications, namely on Animals with Attributes and ImageNet for image classification and on MPII Composites for activity recognition. Our approach consistently outperforms state-of-the-art transfer and semi-supervised approaches on all datasets.",
"In principle, zero-shot learning makes it possible to train a recognition model simply by specifying the category's attributes. For example, with classifiers for generic attributes like and , one can construct a classifier for the zebra category by enumerating which properties it possesses---even without providing zebra training images. In practice, however, the standard zero-shot paradigm suffers because attribute predictions in novel images are hard to get right. We propose a novel random forest approach to train zero-shot models that explicitly accounts for the unreliability of attribute predictions. By leveraging statistics about each attribute's error tendencies, our method obtains more robust discriminative models for the unseen classes. We further devise extensions to handle the few-shot scenario and unreliable attribute descriptions. On three datasets, we demonstrate the benefit for visual category learning with zero or few training examples, a critical domain for rare categories or categories defined on the fly.",
"Object recognition by zero-shot learning (ZSL) aims to recognise objects without seeing any visual examples by learning knowledge transfer between seen and unseen object classes. This is typically achieved by exploring a semantic embedding space such as attribute space or semantic word vector space. In such a space, both seen and unseen class labels, as well as image features can be embedded (projected), and the similarity between them can thus be measured directly. Existing works differ in what embedding space is used and how to project the visual data into the semantic embedding space. Yet, they all measure the similarity in the space using a conventional distance metric (e.g. cosine) that does not consider the rich intrinsic structure, i.e. semantic manifold, of the semantic categories in the embedding space. In this paper we propose to model the semantic manifold in an embedding space using a semantic class label graph. The semantic manifold structure is used to redefine the distance metric in the semantic embedding space for more effective ZSL. The proposed semantic manifold distance is computed using a novel absorbing Markov chain process (AMP), which has a very efficient closed-form solution. The proposed new model improves upon and seamlessly unifies various existing ZSL algorithms. Extensive experiments on both the large scale ImageNet dataset and the widely used Animal with Attribute (AwA) dataset show that our model outperforms significantly the state-of-the-arts."
]
} |
1511.04150 | 2178632969 | The use of distributions and high-level features from deep architecture has become commonplace in modern computer vision. Both of these methodologies have separately achieved a great deal of success in many computer vision tasks. However, there has been little work attempting to leverage the power of these to methodologies jointly. To this end, this paper presents the Deep Mean Maps (DMMs) framework, a novel family of methods to non-parametrically represent distributions of features in convolutional neural network models. DMMs are able to both classify images using the distribution of top-level features, and to tune the top-level features for performing this task. We show how to implement DMMs using a special mean map layer composed of typical CNN operations, making both forward and backward propagation simple. We illustrate the efficacy of DMMs at analyzing distributional patterns in image data in a synthetic data experiment. We also show that we extending existing deep architectures with DMMs improves the performance of existing CNNs on several challenging real-world datasets. | As previously mentioned, deep architectures for learning high-level features has been explored in a myriad of work @cite_25 @cite_22 @cite_0 @cite_39 @cite_36 . While such high-level features may be extracted in a variety of different deep architectures, we focus on deep convolution neural networks (CNNs) @cite_27 @cite_15 @cite_17 @cite_22 . Modern CNNs have been able to employ more layers @cite_0 @cite_3 without unsupervised pretraining thanks to their reduced parameterization (as compared to general neural networks), recent advents in training methods and nonlinear units @cite_5 @cite_7 , and large datasets @cite_14 . | {
"cite_N": [
"@cite_14",
"@cite_22",
"@cite_7",
"@cite_36",
"@cite_3",
"@cite_39",
"@cite_0",
"@cite_27",
"@cite_5",
"@cite_15",
"@cite_25",
"@cite_17"
],
"mid": [
"2117539524",
"104211377",
"",
"2950789693",
"",
"2130325614",
"",
"",
"1904365287",
"2310919327",
"2100495367",
"2134557905"
],
"abstract": [
"The ImageNet Large Scale Visual Recognition Challenge is a benchmark in object category classification and detection on hundreds of object categories and millions of images. The challenge has been run annually from 2010 to present, attracting participation from more than fifty institutions. This paper describes the creation of this benchmark dataset and the advances in object recognition that have been possible as a result. We discuss the challenges of collecting large-scale ground truth annotation, highlight key breakthroughs in categorical object recognition, provide a detailed analysis of the current state of the field of large-scale image classification and object detection, and compare the state-of-the-art computer vision accuracy with human accuracy. We conclude with lessons learned in the 5 years of the challenge, and propose future directions and improvements.",
"Fast visual recognition in the mammalian cortex seems to be a hierarchical process by which the representation of the visual world is transformed in multiple stages from low-level retinotopic features to high-level, global and invariant features, and to object categories. Every single step in this hierarchy seems to be subject to learning. How does the visual cortex learn such hierarchical representations by just looking at the world? How could computers learn such representations from data? Computer vision models that are weakly inspired by the visual cortex will be described. A number of unsupervised learning algorithms to train these models will be presented, which are based on the sparse auto-encoder concept. The effectiveness of these algorithms for learning invariant feature hierarchies will be demonstrated with a number of practical tasks such as scene parsing, pedestrian detection, and object classification.",
"",
"We consider the problem of building high- level, class-specific feature detectors from only unlabeled data. For example, is it possible to learn a face detector using only unlabeled images? To answer this, we train a 9-layered locally connected sparse autoencoder with pooling and local contrast normalization on a large dataset of images (the model has 1 bil- lion connections, the dataset has 10 million 200x200 pixel images downloaded from the Internet). We train this network using model parallelism and asynchronous SGD on a clus- ter with 1,000 machines (16,000 cores) for three days. Contrary to what appears to be a widely-held intuition, our experimental re- sults reveal that it is possible to train a face detector without having to label images as containing a face or not. Control experiments show that this feature detector is robust not only to translation but also to scaling and out-of-plane rotation. We also find that the same network is sensitive to other high-level concepts such as cat faces and human bod- ies. Starting with these learned features, we trained our network to obtain 15.8 accu- racy in recognizing 20,000 object categories from ImageNet, a leap of 70 relative im- provement over the previous state-of-the-art.",
"",
"There has been much interest in unsupervised learning of hierarchical generative models such as deep belief networks. Scaling such models to full-sized, high-dimensional images remains a difficult problem. To address this problem, we present the convolutional deep belief network, a hierarchical generative model which scales to realistic image sizes. This model is translation-invariant and supports efficient bottom-up and top-down probabilistic inference. Key to our approach is probabilistic max-pooling, a novel technique which shrinks the representations of higher layers in a probabilistically sound way. Our experiments show that the algorithm learns useful high-level visual features, such as object parts, from unlabeled images of objects and natural scenes. We demonstrate excellent performance on several visual recognition tasks and show that our model can perform hierarchical (bottom-up and top-down) inference over full-sized images.",
"",
"",
"When a large feedforward neural network is trained on a small training set, it typically performs poorly on held-out test data. This \"overfitting\" is greatly reduced by randomly omitting half of the feature detectors on each training case. This prevents complex co-adaptations in which a feature detector is only helpful in the context of several other specific feature detectors. Instead, each neuron learns to detect a feature that is generally helpful for producing the correct answer given the combinatorially large variety of internal contexts in which it must operate. Random \"dropout\" gives big improvements on many benchmark tasks and sets new records for speech and object recognition.",
"",
"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.",
"We assess the applicability of several popular learning methods for the problem of recognizing generic visual categories with invariance to pose, lighting, and surrounding clutter. A large dataset comprising stereo image pairs of 50 uniform-colored toys under 36 azimuths, 9 elevations, and 6 lighting conditions was collected (for a total of 194,400 individual images). The objects were 10 instances of 5 generic categories: four-legged animals, human figures, airplanes, trucks, and cars. Five instances of each category were used for training, and the other five for testing. Low-resolution grayscale images of the objects with various amounts of variability and surrounding clutter were used for training and testing. Nearest neighbor methods, support vector machines, and convolutional networks, operating on raw pixels or on PCA-derived features were tested. Test error rates for unseen object instances placed on uniform backgrounds were around 13 for SVM and 7 for convolutional nets. On a segmentation recognition task with highly cluttered images, SVM proved impractical, while convolutional nets yielded 16 7 error. A real-time version of the system was implemented that can detect and classify objects in natural scenes at around 10 frames per second."
]
} |
1511.04150 | 2178632969 | The use of distributions and high-level features from deep architecture has become commonplace in modern computer vision. Both of these methodologies have separately achieved a great deal of success in many computer vision tasks. However, there has been little work attempting to leverage the power of these to methodologies jointly. To this end, this paper presents the Deep Mean Maps (DMMs) framework, a novel family of methods to non-parametrically represent distributions of features in convolutional neural network models. DMMs are able to both classify images using the distribution of top-level features, and to tune the top-level features for performing this task. We show how to implement DMMs using a special mean map layer composed of typical CNN operations, making both forward and backward propagation simple. We illustrate the efficacy of DMMs at analyzing distributional patterns in image data in a synthetic data experiment. We also show that we extending existing deep architectures with DMMs improves the performance of existing CNNs on several challenging real-world datasets. | There has been work that explores the use of distributions of high-level deep architecture features with ad-hoc features from distributions of fixed, pre-trained high-level features @cite_24 . In contrast, this paper learns high-level features and uses their distributions in a scalable, nonparametric fashion with mean maps. | {
"cite_N": [
"@cite_24"
],
"mid": [
"2107325234"
],
"abstract": [
"In computer vision, an entity such as an image or video is often represented as a set of instance vectors, which can be SIFT, motion, or deep learning feature vectors extracted from different parts of that entity. Thus, it is essential to design efficient and effective methods to compare two sets of instance vectors. Existing methods such as FV, VLAD or Super Vectors have achieved excellent results. However, this paper shows that these methods are designed based on a generative perspective, and a discriminative method can be more effective in categorizing images or videos. The proposed D3 (discriminative distribution distance) method effectively compares two sets as two distributions, and proposes a directional total variation distance (DTVD) to measure how separated are they. Furthermore, a robust classifier-based method is proposed to estimate DTVD robustly. The D3 method is evaluated in action and image recognition tasks and has achieved excellent accuracy and speed. D3 also has a synergy with FV. The combination of D3 and FV has advantages over D3, FV, and VLAD."
]
} |
1511.04150 | 2178632969 | The use of distributions and high-level features from deep architecture has become commonplace in modern computer vision. Both of these methodologies have separately achieved a great deal of success in many computer vision tasks. However, there has been little work attempting to leverage the power of these to methodologies jointly. To this end, this paper presents the Deep Mean Maps (DMMs) framework, a novel family of methods to non-parametrically represent distributions of features in convolutional neural network models. DMMs are able to both classify images using the distribution of top-level features, and to tune the top-level features for performing this task. We show how to implement DMMs using a special mean map layer composed of typical CNN operations, making both forward and backward propagation simple. We illustrate the efficacy of DMMs at analyzing distributional patterns in image data in a synthetic data experiment. We also show that we extending existing deep architectures with DMMs improves the performance of existing CNNs on several challenging real-world datasets. | Mean map embeddings @cite_38 @cite_35 @cite_2 have served as a practical and flexible method to learn using datasets of sample sets drawn from distributions @cite_18 @cite_28 . Furthermore, the use of random features @cite_8 @cite_21 has allowed for scalable learning over distributions @cite_32 @cite_1 @cite_20 @cite_4 . Mean map embeddings have also been used extensively for the comparison and two-sample testing of distributions @cite_35 @cite_29 @cite_6 . Recent research has also explored the use of mean maps in deep architectures for building fast sampling mechanisms, by conducting two-sample tests between generated sample points and original data @cite_13 @cite_34 ; this work differs from ours in using mean maps in the loss function, rather than in the architecture itself. | {
"cite_N": [
"@cite_38",
"@cite_18",
"@cite_35",
"@cite_4",
"@cite_8",
"@cite_28",
"@cite_29",
"@cite_21",
"@cite_1",
"@cite_32",
"@cite_6",
"@cite_2",
"@cite_34",
"@cite_13",
"@cite_20"
],
"mid": [
"1946137962",
"",
"2950536412",
"2073020428",
"",
"1689407543",
"2212660284",
"2123395972",
"1780422458",
"2151244157",
"2100600008",
"",
"2949995983",
"2950292946",
"2737193326"
],
"abstract": [
"We describe a technique for comparing distributions without the need for density estimation as an intermediate step. Our approach relies on mapping the distributions into a reproducing kernel Hilbert space. Applications of this technique can be found in two-sample tests, which are used for determining whether two sets of observations arise from the same distribution, covariate shift correction, local learning, measures of independence, and density estimation.",
"",
"We propose a framework for analyzing and comparing distributions, allowing us to design statistical tests to determine if two samples are drawn from different distributions. Our test statistic is the largest difference in expectations over functions in the unit ball of a reproducing kernel Hilbert space (RKHS). We present two tests based on large deviation bounds for the test statistic, while a third is based on the asymptotic distribution of this statistic. The test statistic can be computed in quadratic time, although efficient linear time approximations are available. Several classical metrics on distributions are recovered when the function space used to compute the difference in expectations is allowed to be more general (eg. a Banach space). We apply our two-sample tests to a variety of problems, including attribute matching for databases using the Hungarian marriage method, where they perform strongly. Excellent performance is also obtained when comparing distributions over graphs, for which these are the first such tests.",
"We present a new solution to the ecological inference'' problem, of learning individual-level associations from aggregate data. This problem has a long history and has attracted much attention, debate, claims that it is unsolvable, and purported solutions. Unlike other ecological inference techniques, our method makes use of unlabeled individual-level data by embedding the distribution over these predictors into a vector in Hilbert space. Our approach relies on recent learning theory results for distribution regression, using kernel embeddings of distributions. Our novel approach to distribution regression exploits the connection between Gaussian process regression and kernel ridge regression, giving us a coherent, Bayesian approach to learning and inference and a convenient way to include prior information in the form of a spatial covariance function. Our approach is highly scalable as it relies on FastFood, a randomized explicit feature representation for kernel embeddings. We apply our approach to the challenging political science problem of modeling the voting behavior of demographic groups based on aggregate voting data. We consider the 2012 US Presidential election, and ask: what was the probability that members of various demographic groups supported Barack Obama, and how did this vary spatially across the country? Our results match standard survey-based exit polling data for the small number of states for which it is available, and serve to fill in the large gaps in this data, at a much higher degree of granularity.",
"",
"We focus on the distribution regression problem: regressing to vector-valued outputs from probability measures. Many important machine learning and statistical tasks fit into this framework, including multi-instance learning and point estimation problems without analytical solution (such as hyperparameter or entropy estimation). Despite the large number of available heuristics in the literature, the inherent two-stage sampled nature of the problem makes the theoretical analysis quite challenging, since in practice only samples from sampled distributions are observable, and the estimates have to rely on similarities computed between sets of points. To the best of our knowledge, the only existing technique with consistency guarantees for distribution regression requires kernel density estimation as an intermediate step (which often performs poorly in practice), and the domain of the distributions to be compact Euclidean. In this paper, we study a simple, analytically computable, ridge regression-based alternative to distribution regression, where we embed the distributions to a reproducing kernel Hilbert space, and learn the regressor from the embeddings to the outputs. Our main contribution is to prove that this scheme is consistent in the two-stage sampled setup under mild conditions (on separable topological domains enriched with kernels): we present an exact computational-statistical efficiency trade-off analysis showing that our estimator is able to match the one-stage sampled minimax optimal rate [Caponnetto and De Vito, 2007; , 2009]. This result answers a 17-year-old open question, establishing the consistency of the classical set kernel [Haussler, 1999; Gaertner et. al, 2002] in regression. We also cover consistency for more recent kernels on distributions, including those due to [Christmann and Steinwart, 2010].",
"We propose a framework for analyzing and comparing distributions, which we use to construct statistical tests to determine if two samples are drawn from different distributions. Our test statistic is the largest difference in expectations over functions in the unit ball of a reproducing kernel Hilbert space (RKHS), and is called the maximum mean discrepancy (MMD).We present two distribution free tests based on large deviation bounds for the MMD, and a third test based on the asymptotic distribution of this statistic. The MMD can be computed in quadratic time, although efficient linear time approximations are available. Our statistic is an instance of an integral probability metric, and various classical metrics on distributions are obtained when alternative function classes are used in place of an RKHS. We apply our two-sample tests to a variety of problems, including attribute matching for databases using the Hungarian marriage method, where they perform strongly. Excellent performance is also obtained when comparing distributions over graphs, for which these are the first such tests.",
"Randomized neural networks are immortalized in this AI Koan: In the days when Sussman was a novice, Minsky once came to him as he sat hacking at the PDP-6. \"What are you doing?\" asked Minsky. \"I am training a randomly wired neural net to play tic-tac-toe,\" Sussman replied. \"Why is the net wired randomly?\" asked Minsky. Sussman replied, \"I do not want it to have any preconceptions of how to play.\" Minsky then shut his eyes. \"Why do you close your eyes?\" Sussman asked his teacher. \"So that the room will be empty,\" replied Minsky. At that moment, Sussman was enlightened. We analyze shallow random networks with the help of concentration of measure inequalities. Specifically, we consider architectures that compute a weighted sum of their inputs after passing them through a bank of arbitrary randomized nonlinearities. We identify conditions under which these networks exhibit good classification performance, and bound their test error in terms of the size of the dataset and the number of random nonlinearities.",
"We propose an efficient nonparametric strategy for learning a message operator in expectation propagation (EP), which takes as input the set of incoming messages to a factor node, and produces an outgoing message as output. This learned operator replaces the multivariate integral required in classical EP, which may not have an analytic expression. We use kernel-based regression, which is trained on a set of probability distributions representing the incoming messages, and the associated outgoing messages. The kernel approach has two main advantages: first, it is fast, as it is implemented using a novel two-layer random feature representation of the input message distributions; second, it has principled uncertainty estimates, and can be cheaply updated online, meaning it can request and incorporate new training data when it encounters inputs on which it is uncertain. In experiments, our approach is able to solve learning problems where a single message operator is required for multiple, substantially different data sets (logistic regression for a variety of classification problems), where it is essential to accurately assess uncertainty and to efficiently and robustly update the message operator.",
"We study the problem of distribution to real-value regression, where one aims to regress a mapping @math that takes in a distribution input covariate @math (for a non-parametric family of distributions @math ) and outputs a real-valued response @math . This setting was recently studied, and a \"Kernel-Kernel\" estimator was introduced and shown to have a polynomial rate of convergence. However, evaluating a new prediction with the Kernel-Kernel estimator scales as @math . This causes the difficult situation where a large amount of data may be necessary for a low estimation risk, but the computation cost of estimation becomes infeasible when the data-set is too large. To this end, we propose the Double-Basis estimator, which looks to alleviate this big data problem in two ways: first, the Double-Basis estimator is shown to have a computation complexity that is independent of the number of of instances @math when evaluating new predictions after training; secondly, the Double-Basis estimator is shown to have a fast rate of convergence for a general class of mappings @math .",
"We propose a new measure of conditional dependence of random variables, based on normalized cross-covariance operators on reproducing kernel Hilbert spaces. Unlike previous kernel dependence measures, the proposed criterion does not depend on the choice of kernel in the limit of infinite data, for a wide class of kernels. At the same time, it has a straightforward empirical estimate with good convergence behaviour. We discuss the theoretical properties of the measure, and demonstrate its application in experiments.",
"",
"We consider training a deep neural network to generate samples from an unknown distribution given i.i.d. data. We frame learning as an optimization minimizing a two-sample test statistic---informally speaking, a good generator network produces samples that cause a two-sample test to fail to reject the null hypothesis. As our two-sample test statistic, we use an unbiased estimate of the maximum mean discrepancy, which is the centerpiece of the nonparametric kernel two-sample test proposed by (2012). We compare to the adversarial nets framework introduced by (2014), in which learning is a two-player game between a generator network and an adversarial discriminator network, both trained to outwit the other. From this perspective, the MMD statistic plays the role of the discriminator. In addition to empirical comparisons, we prove bounds on the generalization error incurred by optimizing the empirical MMD.",
"We consider the problem of learning deep generative models from data. We formulate a method that generates an independent sample via a single feedforward pass through a multilayer perceptron, as in the recently proposed generative adversarial networks (, 2014). Training a generative adversarial network, however, requires careful optimization of a difficult minimax program. Instead, we utilize a technique from statistical hypothesis testing known as maximum mean discrepancy (MMD), which leads to a simple objective that can be interpreted as matching all orders of statistics between a dataset and samples from the model, and can be trained by backpropagation. We further boost the performance of this approach by combining our generative network with an auto-encoder network, using MMD to learn to generate codes that can then be decoded to produce samples. We show that the combination of these techniques yields excellent generative models compared to baseline approaches as measured on MNIST and the Toronto Face Database.",
"We pose causal inference as the problem of learning to classify probability distributions. In particular, we assume access to a collection (Si, li) i=1, where each Si is a sample drawn from the probability distribution ofXi×Yi, and li is a binary label indicating whether “Xi → Yi” or “Xi ← Yi”. Given these data, we build a causal inference rule in two steps. First, we featurize each Si using the kernel mean embedding associated with some characteristic kernel. Second, we train a binary classifier on such embeddings to distinguish between causal directions. We present generalization bounds showing the statistical consistency and learning rates of the proposed approach, and provide a simple implementation that achieves state-of-the-art cause-effect inference. Furthermore, we extend our ideas to infer causal relationships between more than two variables."
]
} |
1511.04143 | 2253991908 | Recent work has shown that deep neural networks are capable of approximating both value functions and policies in reinforcement learning domains featuring continuous state and action spaces. However, to the best of our knowledge no previous work has succeeded at using deep neural networks in structured (parameterized) continuous action spaces. To fill this gap, this paper focuses on learning within the domain of simulated RoboCup soccer, which features a small set of discrete action types, each of which is parameterized with continuous variables. The best learned agent can score goals more reliably than the 2012 RoboCup champion agent. As such, this paper represents a successful extension of deep reinforcement learning to the class of parameterized action space MDPs. | Competitive RoboCup agents are primarily handcoded but may feature components that are learned or optimized. @cite_6 employed the layered-learning framework to incrementally learn a series of interdependent behaviors. Perhaps the best example of comprehensively integrating learning is the Brainstormers who, in competition, use a neural network to make a large portion of decisions spanning low level skills through high level strategy . However their work was done prior to the advent of deep reinforcement learning, and thus required more constrained, focused training environments for each of their skills. In contrast, our study learns to approach the ball, kick towards the goal, and score, all within the context of a single, monolithic policy. | {
"cite_N": [
"@cite_6"
],
"mid": [
"2127038482"
],
"abstract": [
"RoboCup soccer simulation features the challenges of a fully distributed multi-agent domain with continuous state and action spaces, partial observability, as well as noisy perception and action execution. While the application of machine learning techniques in this domain represents a promising idea in itself, the competitive character of RoboCup also evokes the desire to head for the development of learning algorithms that are more than just a proof of concept. In this paper, we report on our experiences and achievements in applying reinforcement learning (RL) methods in the scope of our Brainstormers competition team within the Simulation League of RoboCup during the past years"
]
} |
1511.04143 | 2253991908 | Recent work has shown that deep neural networks are capable of approximating both value functions and policies in reinforcement learning domains featuring continuous state and action spaces. However, to the best of our knowledge no previous work has succeeded at using deep neural networks in structured (parameterized) continuous action spaces. To fill this gap, this paper focuses on learning within the domain of simulated RoboCup soccer, which features a small set of discrete action types, each of which is parameterized with continuous variables. The best learned agent can score goals more reliably than the 2012 RoboCup champion agent. As such, this paper represents a successful extension of deep reinforcement learning to the class of parameterized action space MDPs. | Deep learning methods have proven useful in various control domains. As previously mentioned DQN and DDPG provide great starting points for learning in discrete and continuous action spaces. Additionally, @cite_4 demonstrates the ability of deep learning paired with guided policy search to learn manipulation policies on a physical robot. The high requirement for data (in the form of experience) is a hurdle for applying deep reinforcement learning directly onto robotic platforms. Our work differs by examining an action space with latent structure and parameterized-continuous actions. | {
"cite_N": [
"@cite_4"
],
"mid": [
"2964161785"
],
"abstract": [
"Policy search methods can allow robots to learn control policies for a wide range of tasks, but practical applications of policy search often require hand-engineered components for perception, state estimation, and low-level control. In this paper, we aim to answer the following question: does training the perception and control systems jointly end-to-end provide better performance than training each component separately? To this end, we develop a method that can be used to learn policies that map raw image observations directly to torques at the robot's motors. The policies are represented by deep convolutional neural networks (CNNs) with 92,000 parameters, and are trained using a guided policy search method, which transforms policy search into supervised learning, with supervision provided by a simple trajectory-centric reinforcement learning method. We evaluate our method on a range of real-world manipulation tasks that require close coordination between vision and control, such as screwing a cap onto a bottle, and present simulated comparisons to a range of prior policy search methods."
]
} |
1511.03885 | 2138140974 | The Unified Model (UM) code supports simulation of weather, climate and earth system processes. It is primarily developed by the UK Met Office, but in recent years a wider community of users and developers have grown around the code. Here we present results from the optimisation work carried out by the UK National Centre for Atmospheric Science (NCAS) for a high resolution configuration (N512 @math 25km) on the UK ARCHER supercomputer, a Cray XC-30. On ARCHER, we use Cray Performance Analysis Tools (CrayPAT) to analyse the performance of UM and then Cray Reveal to identify and parallelise serial loops using OpenMP directives. We compare performance of the optimised version at a range of scales, and with a range of optimisations, including altered MPI rank placement, and addition of OpenMP directives. It is seen that improvements in MPI configuration yield performance improvements of between 5 and 12 , and the added OpenMP directives yield an additional 5-16 speedup. We also identify further code optimisations which could yield yet greater improvement in performance. We note that speedup gained using addition of OpenMP directives does not result in improved performance on the IBM Power platform where much of the code has been developed. This suggests that performance gains on future heterogeneous architectures will be hard to port. Nonetheless, it is clear that the investment of months in analysis and optimisation has yielded performance gains that correspond to the saving of tens of millions of core-hours on current climate projects. | In @cite_21 , the authors prescribe that the MPI ranks should be reordered to match the application communication pattern with the underlying hardware. Also @cite_16 found MPI reordering to be a promising optimisation for unstructured CFD code. In @cite_32 , the researchers studied the performance of a simple shallow water model on a Cray XE6 machine (HECTOR). They suggest that MPI ranks should be mapped to physical cores such that the off-node data transfer volume in a nearest neighbour communication can be reduced. This is in agreement with the our findings based on the performance of the UM on ARCHER, a Cray XC30 machine. | {
"cite_N": [
"@cite_21",
"@cite_32",
"@cite_16"
],
"mid": [
"106431406",
"2023682886",
""
],
"abstract": [
"Modern hardware architectures featuring multicores and a complex memory hierarchy raise challenges that need to be addressed by parallel applications programmers. It is therefore tempting to adapt an application communication pattern to the characteristics of the underlying hardware. The MPI standard features several functions that allow the ranks of MPI processes to be reordered according to a graph attached to a newly created communicator. In this paper, we explain how theMPICH2 implementation of the MPI Dist graph create function was modified to reorder the MPI process ranks to create a match between the application communication pattern and the hardware topology. The experimental results on a multicore cluster show that improvements can be achieved as long as the application communication pattern is expressed by a relevant metric.",
"The complexity of current and emerging architectures provides users with options about how best to use the available resources, but makes predicting performance challenging. In this work a benchmark-driven model is developed for a simple shallow water code on a Cray XE6 system, to explore how deployment choices such as domain decomposition and core affinity affect performance. The resource sharing present in modern multi-core architectures adds various levels of heterogeneity to the system. Shared resources often includes cache, memory, network controllers and in some cases floating point units (as in the AMD Bulldozer), which mean that the access time depends on the mapping of application tasks, and the core's location within the system. Heterogeneity further increases with the use of hardware-accelerators such as GPUs and the Intel Xeon Phi, where many specialist cores are attached to general-purpose cores. This trend for shared resources and non-uniform cores is expected to continue into the exascale era. The complexity of these systems means that various runtime scenarios are possible, and it has been found that under-populating nodes, altering the domain decomposition and non-standard task to core mappings can dramatically alter performance. To find this out, however, is often a process of trial and error. To better inform this process, a performance model was developed for a simple regular grid-based kernel code, shallow. The code comprises two distinct types of work, loop-based array updates and nearest-neighbour halo-exchanges. Separate performance models were developed for each part, both based on a similar methodology. Application specific benchmarks were run to measure performance for different problem sizes under different execution scenarios. These results were then fed into a performance model that derives resource usage for a given deployment scenario, with interpolation between results as necessary.",
""
]
} |
1511.04119 | 2963750390 | MOTIVATION Attention based models have been shown to achieve promising results on several challenging tasks, including caption generation [9], machine translation [1], game-playing and tracking [4]. Attention based models can potentially infer the action happening in videos by focusing only on the relevant places in each video frame. Soft-attention models are deterministic and can be trained using backpropagation. We propose a soft-attention based model for action recognition in videos. We use multi-layered Recurrent Neural Networks (RNNs) with Long Short-Term Memory (LSTM). Our model tends to recognize important elements in video frames based on the activities it detects. | Most of the existing approaches also tend to have CNNs underlying the LSTMs and classify sequences directly or do temporal pooling of features prior to classification . LSTMs have also been used to learn an effective representation of videos in unsupervised settings by using them in an encoder-decoder framework. More recently, @cite_0 have proposed to use 3-D CNN features and an LSTM decoder in an encoder-decoder framework to generate video descriptions. Their model incorporates attention on a video level by defining a probability distribution over frames used to generate individual words. They, however, do not employ an attention mechanism on a frame level (i.e. within a single frame). | {
"cite_N": [
"@cite_0"
],
"mid": [
"2133564696"
],
"abstract": [
"Neural machine translation is a recently proposed approach to machine translation. Unlike the traditional statistical machine translation, the neural machine translation aims at building a single neural network that can be jointly tuned to maximize the translation performance. The models proposed recently for neural machine translation often belong to a family of encoder-decoders and consists of an encoder that encodes a source sentence into a fixed-length vector from which a decoder generates a translation. In this paper, we conjecture that the use of a fixed-length vector is a bottleneck in improving the performance of this basic encoder-decoder architecture, and propose to extend this by allowing a model to automatically (soft-)search for parts of a source sentence that are relevant to predicting a target word, without having to form these parts as a hard segment explicitly. With this new approach, we achieve a translation performance comparable to the existing state-of-the-art phrase-based system on the task of English-to-French translation. Furthermore, qualitative analysis reveals that the (soft-)alignments found by the model agree well with our intuition."
]
} |
1511.04190 | 2233633742 | We study the computational complexity of committee selection problem for several approval-based voting rules in the presence of outliers. Our first result shows that outlier consideration makes committee selection problem intractable for approval, net approval, and minisum approval voting rules. We then study parameterized complexity of this problem with five natural parameters, namely the target score, the size of the committee (and its dual parameter, the number of candidates outside the committee), the number of outliers (and its dual parameter, the number of non-outliers). For net approval and minisum approval voting rules, we provide a dichotomous result, resolving the parameterized complexity of this problem for all subsets of five natural parameters considered (by showing either FPT or W[1]-hardness for all subsets of parameters). For the approval voting rule, we resolve the parameterized complexity of this problem for all subsets of parameters except one. We also study approximation algorithms for this problem. We show that there does not exist any alpha(.) factor approximation algorithm for approval and net approval voting rules, for any computable function alpha(.), unless P=NP. For the minisum voting rule, we provide a pseudopolynomial (1+eps) factor approximation algorithm. | The notion of outliers is quite prominent in the literature pertaining to closest strings, where the usual setting is that we are given @math strings over some alphabet @math , and the task is to find a string @math that minimizes either the maximum Hamming distance from any string, or the sum of its Hamming distances from all strings. In the context of social choice theory, the notion of outliers are intimately related to manipulation and control of election by removing voters and candidates @cite_0 @cite_43 @cite_42 @cite_12 @cite_27 @cite_38 @cite_17 @cite_8 @cite_14 @cite_5 @cite_2 @cite_24 . | {
"cite_N": [
"@cite_38",
"@cite_14",
"@cite_8",
"@cite_42",
"@cite_0",
"@cite_43",
"@cite_27",
"@cite_24",
"@cite_2",
"@cite_5",
"@cite_12",
"@cite_17"
],
"mid": [
"119722335",
"2126893170",
"2123942784",
"1972375916",
"2161392628",
"2003296459",
"2570387407",
"2167625605",
"2065691221",
"1529060712",
"236418988",
"1644945789"
],
"abstract": [
"3 Voting 17 3.1 Voting Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 3.1.1 Scoring Rules . . . . . . . . . . . . . . . . . . . . . . . . 18 3.1.2 Condorcet Extensions . . . . . . . . . . . . . . . . . . . 19 3.1.3 Other Rules . . . . . . . . . . . . . . . . . . . . . . . . . 22 3.2 Manipulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 3.2.1 The Gibbard-Satterthwaite Impossibility . . . . . . . . . . 24 3.2.2 Restricted Domains of Preferences . . . . . . . . . . . . . 24 3.2.3 Computational Hardness of Manipulation . . . . . . . . . 26 3.2.4 Probabilistic Voting Rules . . . . . . . . . . . . . . . . . 29 3.2.5 Irresolute Voting Rules . . . . . . . . . . . . . . . . . . . 30 3.3 Possible and Necessary Winners . . . . . . . . . . . . . . . . . . 31",
"Although recent years have seen a surge of interest in the computational aspects of social choice, no specific attention has previously been devoted to elections with multiple winners, e.g., elections of an assembly or committee. In this paper, we characterize the worst-case complexity of manipulation and control in the context of four prominent multiwinner voting systems, under different formulations of the strategic agent's goal.",
"In 1992, Bartholdi, Tovey, and Trick opened the study of control attacks on elections-- attempts to improve the election outcome by such actions as adding deleting candidates or voters. That work has led to many results on how algorithms can be used to find attacks on elections and how complexity-theoretic hardness results can be used as shields against attacks. However, all the work in this line has assumed that the attacker employs just a single type of attack. In this paper, we model and study the case in which the attacker launches a multipronged (i.e., multimode) attack. We do so to more realistically capture the richness of real-life settings. For example, an attacker might simultaneously try to suppress some voters, attract new voters into the election, and introduce a spoiler candidate. Our model provides a unified framework for such varied attacks. By constructing polynomialtime multiprong attack algorithms we prove that for various election systems even such concerted, flexible attacks can be perfectly planned in deterministic polynomial time.",
"In multiagent settings where the agents have different preferences, preference aggregation is a central issue. Voting is a general method for preference aggregation, but seminal results have shown that all general voting protocols are manipulable. One could try to avoid manipulation by using protocols where determining a beneficial manipulation is hard. Especially among computational agents, it is reasonable to measure this hardness by computational complexity. Some earlier work has been done in this area, but it was assumed that the number of voters and candidates is unbounded. Such hardness results lose relevance when the number of candidates is small, because manipulation algorithms that are exponential only in the number of candidates (and only slightly so) might be available. We give such an algorithm for an individual agent to manipulate the Single Transferable Vote (STV) protocol, which has been shown hard to manipulate in the above sense. This motivates the core of this article, which derives hardness results for realistic elections where the number of candidates is a small constant (but the number of voters can be large). The main manipulation question we study is that of coalitional manipulation by weighted voters. (We show that for simpler manipulation problems, manipulation cannot be hard with few candidates.) We study both constructive manipulation (making a given candidate win) and destructive manipulation (making a given candidate not win). We characterize the exact number of candidates for which manipulation becomes hard for the plurality, Borda, STV, Copeland, maximin, veto, plurality with runoff, regular cup, and randomized cup protocols. We also show that hardness of manipulation in this setting implies hardness of manipulation by an individual in unweighted settings when there is uncertainty about the others' votes (but not vice-versa). To our knowledge, these are the first results on the hardness of manipulation when there is uncertainty about the others' votes.",
"Some voting schemes that are in principle susceptible to control are nevertheless resistant in practice due to excessive computational costs; others are vulnerable. We illustrate this in detail for plurality voting and for Condorcet voting.",
"We show how computational complexity might protect the integrity of social choice. We exhibit a voting rule that efficiently computes winners but is computationally resistant to strategic manipulation. It is NP-complete for a manipulative voter to determine how to exploit knowledge of the preferences of others. In contrast, many standard voting schemes can be manipulated with only polynomial computational effort.",
"Preference aggregation in a multiagent setting is a central issue in both human and computer contexts. In this paper, we study in terms of complexity the vulnerability of preference aggregation to destructive control. In particular, we study the ability of an election's chair to, through such mechanisms as voter candidate addition suppression partition, ensure that a particular candidate (equivalently, alternative) does not win. And we study the extent to which election systems can make it impossible, or computationally costly (NP-complete), for the chair to execute such control. Among the systems we study-plurality, Condorcet, and approval voting-we find cases where systems immune or computationally resistant to a chair choosing the winner nonetheless are vulnerable to the chair blocking a victory. Beyond that, we see that among our studied systems no one system offers the best protection against destructive control. Rather, the choice of a preference aggregation system will depend closely on which types of control one wishes to be protected against. We also find concrete cases where the complexity of or susceptibility to control varies dramatically based on the choice among natural tie-handling rules.",
"There are different ways for an external agent to influence the outcome of an election. We concentrate on \"control\" by adding or deleting candidates of an election. Our main focus is to investigate the parameterized complexity of various control problems for different voting systems. To this end, we introduce natural digraph problems that may be of independent interest. They help in determining the parameterized complexity of control for different voting systems including Llull, Copeland, and plurality votings. Devising several parameterized reductions, we provide a parameterized complexity overview of the digraph and control problems with respect to natural parameters.",
"The concept of distance rationalizability has several applications within social choice. In the context of voting, it allows one to define (\"rationalize\") voting rules via a consensus class (roughly, a set of elections in which it is obvious who should win) and a distance function: namely, a candidate is said to be an election winner if it is ranked first in one of the nearest (with respect to the given distance) consensus elections. It is known that many classic voting rules can be represented in this manner. In this paper, we provide new results on distance rationalizability of several well-known voting rules such as all scoring rules, Approval, Young's rule and Maximin. We also show that a previously published proof of distance rationalizability of Young's rule is incorrect: the consensus notion and the distance function used in that proof give rise to a voting rule that is similar to---but distinct from---the Young's rule. Finally, we demonstrate that some voting rules cannot be rationalized via certain notions of consensus. To the best of our knowledge, these are the first non-distance-rationalizability results for voting rules.",
"Although recent years have seen a surge of interest in the computational aspects of social choice, no attention has previously been devoted to elections with multiple winners, e.g., elections of an assembly or committee. In this paper, we fully characterize the worst-case complexity of manipulation and control in the context of four prominent multi-winner voting systems. Additionally, we show that several tailor-made multi-winner voting schemes are impractical, as it is NP-hard to select the winners in these schemes.",
"",
"We provide an overview of some recent progress on the complexity of election systems. The issues studied include the complexity of the winner, manipulation, bribery, and control problems."
]
} |
1511.04190 | 2233633742 | We study the computational complexity of committee selection problem for several approval-based voting rules in the presence of outliers. Our first result shows that outlier consideration makes committee selection problem intractable for approval, net approval, and minisum approval voting rules. We then study parameterized complexity of this problem with five natural parameters, namely the target score, the size of the committee (and its dual parameter, the number of candidates outside the committee), the number of outliers (and its dual parameter, the number of non-outliers). For net approval and minisum approval voting rules, we provide a dichotomous result, resolving the parameterized complexity of this problem for all subsets of five natural parameters considered (by showing either FPT or W[1]-hardness for all subsets of parameters). For the approval voting rule, we resolve the parameterized complexity of this problem for all subsets of parameters except one. We also study approximation algorithms for this problem. We show that there does not exist any alpha(.) factor approximation algorithm for approval and net approval voting rules, for any computable function alpha(.), unless P=NP. For the minisum voting rule, we provide a pseudopolynomial (1+eps) factor approximation algorithm. | In the minimax notion of distance, the problem of finding a closest string in the presence of outliers was initiated in @cite_29 , and was further refined in @cite_30 . The work in @cite_30 demonstrates hardness of approximation for variations involving outliers. With the minisum notion of distance, the study of finding the best string with respect to outliers was studied in @cite_7 . In contrast with the minimax rule, this work shows a PTAS for . | {
"cite_N": [
"@cite_30",
"@cite_29",
"@cite_7"
],
"mid": [
"2180828828",
"2030393641",
"2049696330"
],
"abstract": [
"Many problems in bioinformatics are about finding strings that approximately represent a collection of given strings. We look at more general problems where some input strings can be classified as outliers. The Close to Most Strings problem is, given a set S of the same-length strings, and a parameter d, find a string x that maximizes the number of ''non-outliers'' within Hamming distance d of x. We prove that this problem has no polynomial-time approximation scheme (PTAS) unless NP has randomized polynomial-time algorithms, correcting a decade-old erroneous proof made previously in the literature. The Most Strings with Few Bad Columns problem is to find a maximum-size subset of input strings so that the number of non-identical positions is at most k; we show it has no PTAS unless P=NP. We also observe Closest to k Strings has no efficient PTAS (EPTAS) unless a parameterized complexity hierarchy collapses. In sum, outliers help model problems associated with using biological data, but we show the problem of finding an approximate solution is computationally difficult.",
"Background Given n strings s1, …, sn each of length l and a nonnegative integer d, the CLOSEST STRING problem asks to find a center string s such that none of the input strings has Hamming distance greater than d from s. Finding a common pattern in many – but not necessarily all – input strings is an important task that plays a role in many applications in bioinformatics.",
"RNA splicing is a cellular process driven by the interaction between numerous regulatory sequences and binding sites, however, such interactions have been primarily explored by laboratory methods since computational tools largely ignore the relationship between different splicing elements. Current computational methods identify either splice sites or other regulatory sequences, such as enhancers and silencers. We present a novel approach for characterizing co-occurring relationships between splice site motifs and splicing enhancers. Our approach relies on an efficient algorithm for approximately solving Consensus Sequence with Outliers, an NP-complete string clustering problem. In particular, we give an algorithm for this problem that outputs near-optimal solutions in polynomial time. To our knowledge, this is the first formulation and computational attempt for detecting co-occurring sequence elements in RNA sequence data. Further, we demonstrate that SeeSite is capable of showing that certain ESEs are preferentially associated with weaker splice sites, and that there exists a co-occurrence relationship with splice site motifs."
]
} |
1511.04190 | 2233633742 | We study the computational complexity of committee selection problem for several approval-based voting rules in the presence of outliers. Our first result shows that outlier consideration makes committee selection problem intractable for approval, net approval, and minisum approval voting rules. We then study parameterized complexity of this problem with five natural parameters, namely the target score, the size of the committee (and its dual parameter, the number of candidates outside the committee), the number of outliers (and its dual parameter, the number of non-outliers). For net approval and minisum approval voting rules, we provide a dichotomous result, resolving the parameterized complexity of this problem for all subsets of five natural parameters considered (by showing either FPT or W[1]-hardness for all subsets of parameters). For the approval voting rule, we resolve the parameterized complexity of this problem for all subsets of parameters except one. We also study approximation algorithms for this problem. We show that there does not exist any alpha(.) factor approximation algorithm for approval and net approval voting rules, for any computable function alpha(.), unless P=NP. For the minisum voting rule, we provide a pseudopolynomial (1+eps) factor approximation algorithm. | The fact that the closest string problem has similarities to the setup of approval ballots has been explored in the past, see, for instance @cite_26 . More specifically, and again in the spirit of observing similarities with the closest string family of problems, the minimax approval rule has been studied from the perspective of outliers @cite_32 . To the best of our knowledge, our work is the first comprehensive study of the complexity behavior of approval ballots in the presence of outliers. | {
"cite_N": [
"@cite_26",
"@cite_32"
],
"mid": [
"1859600792",
"2181533063"
],
"abstract": [
"We consider Approval Voting systems where each voter decides on a subset of candidates he she approves. We focus on the optimization problem of finding the committee of fixed size k, minimizing the maximal Hamming distance from a vote. In this paper we give a PTAS for this problem and hence resolve the open question raised by [AAAI’10]. The result is obtained by adapting the techniques developed by [JACM’02] originally used for the less constrained Closest String problem. The technique relies on extracting information and structural properties of constant size subsets of votes.",
"In this work, we initiate a detailed study of the parameterized complexity of Minimax Approval Voting. We demonstrate that the problem is W[2]-hard when parameterized by the size of the committee to be chosen, but does admit a FPT algorithm when parameterized by the number of strings that is more efficient than the previous ILP-based approaches for the problem. We also consider several combinations of parameters and provide a detailed landscape of the parameterized and kernelization complexity of the problem. We also study the version of the problem where we permit outliers, that is, where the chosen committee is required to satisfy a large number of voters (instead of all of them). In this context, we strengthen an APX-hardness result in the literature, and also show a simple but strong W-hardness result."
]
} |
1511.03759 | 2952494918 | Recently, there is a surge of social recommendation, which leverages social relations among users to improve recommendation performance. However, in many applications, social relations are absent or very sparse. Meanwhile, the attribute information of users or items may be rich. It is a big challenge to exploit these attribute information for the improvement of recommendation performance. In this paper, we organize objects and relations in recommendation system as a heterogeneous information network, and introduce meta path based similarity measure to evaluate the similarity of users or items. Furthermore, a matrix factorization based dual regularization framework SimMF is proposed to flexibly integrate different types of information through adopting the similarity of users and items as regularization on latent factors of users and items. Extensive experiments not only validate the effectiveness of SimMF but also reveal some interesting findings. We find that attribute information of users and items can significantly improve recommendation accuracy, and their contribution seems more important than that of social relations. The experiments also reveal that different regularization models have obviously different impact on users and items. | With the prevalence of social media, more and more research study social recommender systems which utilize social relations among users. Many researchers utilized trust information among users. @cite_7 fused user-item matrix with users' social trust networks by sharing a common latent low-dimensional user feature matrix. Furthermore, the authors in @cite_14 coined the social trust ensemble to represent the formulation of the social trust restrictions. More and more researches began to use friend relation among users. In @cite_27 , the additional social regularization term ensures that the distance of latent feature vectors of two friends with similar tastes to be closer. @cite_29 inferred category-specific social trust circles from available rating data combined with friend relations. Recently, many studies have begun to utilize other types of informations. For example, @cite_16 made use of user and item profiles defined in terms of weighted lists of social tags for top N recommendation. Furthermore, they presented a comparative study on the influence that different types of information available in social systems have on item recommendation @cite_12 . | {
"cite_N": [
"@cite_14",
"@cite_7",
"@cite_29",
"@cite_27",
"@cite_16",
"@cite_12"
],
"mid": [
"2093219534",
"2119825970",
"2126539515",
"2144487656",
"2018426558",
"2092902234"
],
"abstract": [
"As an indispensable technique in the field of Information Filtering, Recommender System has been well studied and developed both in academia and in industry recently. However, most of current recommender systems suffer the following problems: (1) The large-scale and sparse data of the user-item matrix seriously affect the recommendation quality. As a result, most of the recommender systems cannot easily deal with users who have made very few ratings. (2) The traditional recommender systems assume that all the users are independent and identically distributed; this assumption ignores the connections among users, which is not consistent with the real world recommendations. Aiming at modeling recommender systems more accurately and realistically, we propose a novel probabilistic factor analysis framework, which naturally fuses the users' tastes and their trusted friends' favors together. In this framework, we coin the term Social Trust Ensemble to represent the formulation of the social trust restrictions on the recommender systems. The complexity analysis indicates that our approach can be applied to very large datasets since it scales linearly with the number of observations, while the experimental results show that our method performs better than the state-of-the-art approaches.",
"Data sparsity, scalability and prediction quality have been recognized as the three most crucial challenges that every collaborative filtering algorithm or recommender system confronts. Many existing approaches to recommender systems can neither handle very large datasets nor easily deal with users who have made very few ratings or even none at all. Moreover, traditional recommender systems assume that all the users are independent and identically distributed; this assumption ignores the social interactions or connections among users. In view of the exponential growth of information generated by online social networks, social network analysis is becoming important for many Web applications. Following the intuition that a person's social network will affect personal behaviors on the Web, this paper proposes a factor analysis approach based on probabilistic matrix factorization to solve the data sparsity and poor prediction accuracy problems by employing both users' social network information and rating records. The complexity analysis indicates that our approach can be applied to very large datasets since it scales linearly with the number of observations, while the experimental results shows that our method performs much better than the state-of-the-art approaches, especially in the circumstance that users have made few or no ratings.",
"Online social network information promises to increase recommendation accuracy beyond the capabilities of purely rating feedback-driven recommender systems (RS). As to better serve users' activities across different domains, many online social networks now support a new feature of \"Friends Circles\", which refines the domain-oblivious \"Friends\" concept. RS should also benefit from domain-specific \"Trust Circles\". Intuitively, a user may trust different subsets of friends regarding different domains. Unfortunately, in most existing multi-category rating datasets, a user's social connections from all categories are mixed together. This paper presents an effort to develop circle-based RS. We focus on inferring category-specific social trust circles from available rating data combined with social network data. We outline several variants of weighting friends within circles based on their inferred expertise levels. Through experiments on publicly available data, we demonstrate that the proposed circle-based recommendation models can better utilize user's social trust information, resulting in increased recommendation accuracy.",
"Although Recommender Systems have been comprehensively analyzed in the past decade, the study of social-based recommender systems just started. In this paper, aiming at providing a general method for improving recommender systems by incorporating social network information, we propose a matrix factorization framework with social regularization. The contributions of this paper are four-fold: (1) We elaborate how social network information can benefit recommender systems; (2) We interpret the differences between social-based recommender systems and trust-aware recommender systems; (3) We coin the term Social Regularization to represent the social constraints on recommender systems, and we systematically illustrate how to design a matrix factorization objective function with social regularization; and (4) The proposed method is quite general, which can be easily extended to incorporate other contextual information, like social tags, etc. The empirical analysis on two large datasets demonstrates that our approaches outperform other state-of-the-art methods.",
"We present and evaluate various content-based recommendation models that make use of user and item profiles defined in terms of weighted lists of social tags. The studied approaches are adaptations of the Vector Space and Okapi BM25 information retrieval models. We empirically compare the recommenders using two datasets obtained from Delicious and Last.fm social systems, in order to analyse the performance of the approaches in scenarios with different domains and tagging behaviours.",
"While recommendation approaches exploiting different input sources have started to proliferate in the literature, an explicit study of the effect of the combination of heterogeneous inputs is still missing. On the other hand, in this context there are sides to recommendation quality requiring further characterisation and methodological research - a gap that is acknowledged in the field. We present a comparative study on the influence that different types of information available in social systems have on item recommendation. Aiming to identify which sources of user interest evidence - tags, social contacts, and user-item interaction data - are more effective to achieve useful recommendations, and in what aspect, we evaluate a number of content-based, collaborative filtering, and social recommenders on three datasets obtained from Delicious, Last.fm, and MovieLens. Aiming to determine whether and how combining such information sources may enhance over individual recommendation approaches, we extend the common accuracy-oriented evaluation practice with various metrics to measure further recommendation quality dimensions, namely coverage, diversity, novelty, overlap, and relative diversity between ranked item recommendations. We report empiric observations showing that exploiting tagging information by content-based recommenders provides high coverage and novelty, and combining social networking and collaborative filtering information by hybrid recommenders results in high accuracy and diversity. This, along with the fact that recommendation lists from the evaluated approaches had low overlap and relative diversity values between them, gives insights that meta-hybrid recommenders combining the above strategies may provide valuable, balanced item suggestions in terms of performance and non-performance metrics."
]
} |
1511.03759 | 2952494918 | Recently, there is a surge of social recommendation, which leverages social relations among users to improve recommendation performance. However, in many applications, social relations are absent or very sparse. Meanwhile, the attribute information of users or items may be rich. It is a big challenge to exploit these attribute information for the improvement of recommendation performance. In this paper, we organize objects and relations in recommendation system as a heterogeneous information network, and introduce meta path based similarity measure to evaluate the similarity of users or items. Furthermore, a matrix factorization based dual regularization framework SimMF is proposed to flexibly integrate different types of information through adopting the similarity of users and items as regularization on latent factors of users and items. Extensive experiments not only validate the effectiveness of SimMF but also reveal some interesting findings. We find that attribute information of users and items can significantly improve recommendation accuracy, and their contribution seems more important than that of social relations. The experiments also reveal that different regularization models have obviously different impact on users and items. | The proposed SimMF belongs to heterogeneous information based methods. Compared to feedback based and social relation based methods, SimMF can flexibly integrate various heterogeneous information. And SimMF is also different from existing heterogeneous information based models in several aspects. Contemporary methods usually consider one or two types of heterogeneous information. For example, HETEROMF focuses on different interactions of users. The method proposed by only considers attributes of items @cite_20 , and it is an item recommendation @cite_19 @cite_28 model. SimMF considers all kinds of information and flexibly integrates them together. Moreover, we intensively investigate the impact of this heterogeneous information which is seldom explored before. WHyLDR considers heterogeneous information as SimMF does. However, while WHyLDR focuses on component selection and component combination and is for item recommendation rather than rating prediction. The method proposed in @cite_4 and SemRec @cite_18 are both meta-path based model, while SimMF should be considered as a matrix factorization based model. The proposed work is similar to Hete-CF, but Hete-CF only applies one type of matrix factorization constraint called individual regularization on users and items, SimMF considers two types of regularization and exploits their different impacts on recommendation performance. | {
"cite_N": [
"@cite_18",
"@cite_4",
"@cite_28",
"@cite_19",
"@cite_20"
],
"mid": [
"2047729491",
"2408419220",
"2070700141",
"2010187764",
""
],
"abstract": [
"Recently heterogeneous information network (HIN) analysis has attracted a lot of attention, and many data mining tasks have been exploited on HIN. As an important data mining task, recommender system includes a lot of object types (e.g., users, movies, actors, and interest groups in movie recommendation) and the rich relations among object types, which naturally constitute a HIN. The comprehensive information integration and rich semantic information of HIN make it promising to generate better recommendations. However, conventional HINs do not consider the attribute values on links, and the widely used meta path in HIN may fail to accurately capture semantic relations among objects, due to the existence of rating scores (usually ranging from 1 to 5) between users and items in recommender system. In this paper, we are the first to propose the weighted HIN and weighted meta path concepts to subtly depict the path semantics through distinguishing different link attribute values. Furthermore, we propose a semantic path based personalized recommendation method SemRec to predict the rating scores of users on items. Through setting meta paths, SemRec not only flexibly integrates heterogeneous information but also obtains prioritized and personalized weights representing user preferences on paths. Experiments on two real datasets illustrate that SemRec achieves better recommendation performance through flexibly integrating information with the help of weighted meta paths.",
"In the midst of vast amounts of available fashion items, consumers today require more efficient recommendation services. A system that sorts out items that form a stylish ensemble with already selected or possessed items would provide them with greater convenience. In this paper, we propose a fashion item recommendation method that learns the way the fashion items are matched from a large ensemble database. We empirically show that the proposed method can explain factors that affect item matching and recommend the most suitable items to the given set of items.",
"Recent studies suggest that by using additional user or item relationship information when building hybrid recommender systems, the recommendation quality can be largely improved. However, most such studies only consider a single type of relationship, e.g., social network. Notice that in many applications, the recommendation problem exists in an attribute-rich heterogeneous information network environment. In this paper, we study the entity recommendation problem in heterogeneous information networks. We propose to combine various relationship information from the network with user feedback to provide high quality recommendation results. The major challenge of building recommender systems in heterogeneous information networks is to systematically define features to represent the different types of relationships between entities, and learn the importance of each relationship type. In the proposed framework, we first use meta-path-based latent features to represent the connectivity between users and items along different paths in the related information network. We then define a recommendation model with such latent features and use Bayesian ranking optimization techniques to estimate the model. Empirical studies show that our approach outperforms several widely employed implicit feedback entity recommendation techniques.",
"Among different hybrid recommendation techniques, network-based entity recommendation methods, which utilize user or item relationship information, are beginning to attract increasing attention recently. Most of the previous studies in this category only consider a single relationship type, such as friendships in a social network. In many scenarios, the entity recommendation problem exists in a heterogeneous information network environment. Different types of relationships can be potentially used to improve the recommendation quality. In this paper, we study the entity recommendation problem in heterogeneous information networks. Specifically, we propose to combine heterogeneous relationship information for each user differently and aim to provide high-quality personalized recommendation results using user implicit feedback data and personalized recommendation models. In order to take full advantage of the relationship heterogeneity in information networks, we first introduce meta-path-based latent features to represent the connectivity between users and items along different types of paths. We then define recommendation models at both global and personalized levels and use Bayesian ranking optimization techniques to estimate the proposed models. Empirical studies show that our approaches outperform several widely employed or the state-of-the-art entity recommendation techniques.",
""
]
} |
1511.04003 | 2255499165 | This paper presents Pinterest Related Pins, an item-to-item recommendation system that combines collaborative filtering with content-based ranking. We demonstrate that signals derived from user curation, the activity of users organizing content, are highly effective when used in conjunction with content-based ranking. This paper also demonstrates the effectiveness of visual features, such as image or object representations learned from convnets, in improving the user engagement rate of our item-to-item recommendation system. | Visual features are widely used in both content-based recommendation systems and image search systems @cite_0 @cite_11 , and the the learning-to-rank framework used in this paper from @cite_18 has been widely used in industry @cite_20 @cite_4 @cite_15 @cite_16 . To our best knowledge this work contains the first published empirical results on how the latest convolutional neural network (CNN) based visual features (e.g. VGG @cite_12 ) and large-scale object detection using Faster R-CNN @cite_1 can improve commercial recommendation systems. Visual features are computed using a distributed process described in our previous work @cite_2 . | {
"cite_N": [
"@cite_18",
"@cite_4",
"@cite_1",
"@cite_0",
"@cite_2",
"@cite_15",
"@cite_16",
"@cite_12",
"@cite_20",
"@cite_11"
],
"mid": [
"2047221353",
"",
"2953106684",
"2167407098",
"2952846590",
"",
"2121672615",
"1994002998",
"2171278097",
"2118509786"
],
"abstract": [
"This paper presents an approach to automatically optimizing the retrieval quality of search engines using clickthrough data. Intuitively, a good information retrieval system should present relevant documents high in the ranking, with less relevant documents following below. While previous approaches to learning retrieval functions from examples exist, they typically require training data generated from relevance judgments by experts. This makes them difficult and expensive to apply. The goal of this paper is to develop a method that utilizes clickthrough data for training, namely the query-log of the search engine in connection with the log of links the users clicked on in the presented ranking. Such clickthrough data is available in abundance and can be recorded at very low cost. Taking a Support Vector Machine (SVM) approach, this paper presents a method for learning retrieval functions. From a theoretical perspective, this method is shown to be well-founded in a risk minimization framework. Furthermore, it is shown to be feasible even for large sets of queries and features. The theoretical results are verified in a controlled experiment. It shows that the method can effectively adapt the retrieval function of a meta-search engine to a particular group of users, outperforming Google in terms of retrieval quality after only a couple of hundred training examples.",
"",
"State-of-the-art object detection networks depend on region proposal algorithms to hypothesize object locations. Advances like SPPnet and Fast R-CNN have reduced the running time of these detection networks, exposing region proposal computation as a bottleneck. In this work, we introduce a Region Proposal Network (RPN) that shares full-image convolutional features with the detection network, thus enabling nearly cost-free region proposals. An RPN is a fully convolutional network that simultaneously predicts object bounds and objectness scores at each position. The RPN is trained end-to-end to generate high-quality region proposals, which are used by Fast R-CNN for detection. We further merge RPN and Fast R-CNN into a single network by sharing their convolutional features---using the recently popular terminology of neural networks with 'attention' mechanisms, the RPN component tells the unified network where to look. For the very deep VGG-16 model, our detection system has a frame rate of 5fps (including all steps) on a GPU, while achieving state-of-the-art object detection accuracy on PASCAL VOC 2007, 2012, and MS COCO datasets with only 300 proposals per image. In ILSVRC and COCO 2015 competitions, Faster R-CNN and RPN are the foundations of the 1st-place winning entries in several tracks. Code has been made publicly available.",
"The last decade has witnessed great interest in research on content-based image retrieval. This has paved the way for a large number of new techniques and systems, and a growing interest in associated fields to support such systems. Likewise, digital imagery has expanded its horizon in many directions, resulting in an explosion in the volume of image data required to be organized. In this paper, we discuss some of the key contributions in the current decade related to image retrieval and automated image annotation, spanning 120 references. We also discuss some of the key challenges involved in the adaptation of existing image retrieval techniques to build useful systems that can handle real-world data. We conclude with a study on the trends in volume and impact of publications in the field with respect to venues journals and sub-topics.",
"We demonstrate that, with the availability of distributed computation platforms such as Amazon Web Services and open-source tools, it is possible for a small engineering team to build, launch and maintain a cost-effective, large-scale visual search system with widely available tools. We also demonstrate, through a comprehensive set of live experiments at Pinterest, that content recommendation powered by visual search improve user engagement. By sharing our implementation details and the experiences learned from launching a commercial visual search engines from scratch, we hope visual search are more widely incorporated into today's commercial applications.",
"",
"Since the publication of Brin and Page's paper on PageRank, many in the Web community have depended on PageRank for the static (query-independent) ordering of Web pages. We show that we can significantly outperform PageRank using features that are independent of the link structure of the Web. We gain a further boost in accuracy by using data on the frequency at which users visit Web pages. We use RankNet, a ranking machine learning algorithm, to combine these and other static features based on anchor text and domain characteristics. The resulting model achieves a static ranking pairwise accuracy of 67.3 (vs. 56.7 for PageRank or 50 for random).",
"The latest generation of Convolutional Neural Networks (CNN) have achieved impressive results in challenging benchmarks on image recognition and object detection, significantly raising the interest of the community in these methods. Nevertheless, it is still unclear how different CNN methods compare with each other and with previous state-of-the-art shallow representations such as the Bag-of-Visual-Words and the Improved Fisher Vector. This paper conducts a rigorous evaluation of these new techniques, exploring different deep architectures and comparing them on a common ground, identifying and disclosing important implementation details. We identify several useful properties of CNN-based representations, including the fact that the dimensionality of the CNN output layer can be reduced significantly without having an adverse effect on performance. We also identify aspects of deep and shallow methods that can be successfully shared. In particular, we show that the data augmentation techniques commonly applied to CNN-based methods can also be applied to shallow methods, and result in an analogous performance boost. Source code and models to reproduce the experiments in the paper is made publicly available.",
"IBM Research undertook a challenge to build a computer system that could compete at the human champion level in real time on the American TV Quiz show, Jeopardy! The extent of the challenge includes fielding a real-time automatic contestant on the show, not merely a laboratory exercise. The Jeopardy! Challenge helped us address requirements that led to the design of the DeepQA architecture and the implementation of Watson. After 3 years of intense research and development by a core team of about 20 researches, Watson is performing at human expert-levels in terms of precision, confidence and speed at the Jeopardy! Quiz show. Our results strongly suggest that DeepQA is an effective and extensible architecture that may be used as a foundation for combining, deploying, evaluating and advancing a wide range of algorithmic techniques to rapidly advance the field of QA.",
"Because of the relative ease in understanding and processing text, commercial image-search systems often rely on techniques that are largely indistinguishable from text search. Recently, academic studies have demonstrated the effectiveness of employing image-based features to provide either alternative or additional signals to use in this process. However, it remains uncertain whether such techniques will generalize to a large number of popular Web queries and whether the potential improvement to search quality warrants the additional computational cost. In this work, we cast the image-ranking problem into the task of identifying \"authority\" nodes on an inferred visual similarity graph and propose VisualRank to analyze the visual link structures among images. The images found to be \"authorities\" are chosen as those that answer the image-queries well. To understand the performance of such an approach in a real system, we conducted a series of large-scale experiments based on the task of retrieving images for 2,000 of the most popular products queries. Our experimental results show significant improvement, in terms of user satisfaction and relevancy, in comparison to the most recent Google image search results. Maintaining modest computational cost is vital to ensuring that this procedure can be used in practice; we describe the techniques required to make this system practical for large-scale deployment in commercial search engines."
]
} |
1511.03406 | 2259799777 | PEGs are a formal grammar foundation for describing syntax, and are not hard to generate parsers with a plain recursive decent parsing. However, the large amount of C-stack consumption in the recursive parsing is not acceptable especially in resource-restricted embedded systems. Alternatively, we have attempted the machine virtualization approach to PEG-based parsing. MiniNez, our implemented virtual machine, is presented in this paper with several downsizing techniques, including instruction specialization, inline expansion and static flow analysis. As a result, the MiniNez machine achieves both a very small footprint and competitive performance to generated C parsers. We have demonstrated the experimental results by comparing on two major embedded platforms: Cortex-A7 and Intel Atom processor. | Parsing and syntactic analysis is ubiquitous and an integral part of text processing. Since the development of Lex Yacc, the parser generator has increasingly adopted as a standard approach to formal grammar engineering. In contexts of data parsing, binpac @cite_0 and XMLScreamer @cite_10 notably extend Yacc to produce efficient protocol parsers and XML parsers, respectively. However, Yacc and its extended tools do not guarantee its availability on resource-restricted embedded systems, while they produce a portable parser code. To our knowledge, the parser generation that is specialized for embedded systems has been unexplored, and then is an important open problem to be addressed. | {
"cite_N": [
"@cite_0",
"@cite_10"
],
"mid": [
"2141950720",
"2140106368"
],
"abstract": [
"A key step in the semantic analysis of network traffic is to parse the traffic stream according to the high-level protocols it contains. This process transforms raw bytes into structured, typed, and semantically meaningful data fields that provide a high-level representation of the traffic. However, constructing protocol parsers by hand is a tedious and error-prone affair due to the complexity and sheer number of application protocols.This paper presents binpac, a declarative language and compiler designed to simplify the task of constructing robust and efficient semantic analyzers for complex network protocols. We discuss the design of the binpac language and a range of issues in generating efficient parsers from high-level specifications. We have used binpac to build several protocol parsers for the \"Bro\" network intrusion detection system, replacing some of its existing analyzers (handcrafted in C++), and supplementing its operation with analyzers for new protocols. We can then use Bro's powerful scripting language to express application-level analysis of network traffic in high-level terms that are both concise and expressive. binpac is now part of the open-source Bro distribution.",
"This paper describes an experimental system in which customized high performance XML parsers are prepared using parser generation and compilation techniques. Parsing is integrated with Schema-based validation and deserialization, and the resulting validating processors are shown to be as fast as or in many cases significantly faster than traditional nonvalidating parsers. High performance is achieved by integration across layers of software that are traditionally separate, by avoiding unnecessary data copying and transformation, and by careful attention to detail in the generated code. The effect of API design on XML performance is also briefly discussed.."
]
} |
1511.03406 | 2259799777 | PEGs are a formal grammar foundation for describing syntax, and are not hard to generate parsers with a plain recursive decent parsing. However, the large amount of C-stack consumption in the recursive parsing is not acceptable especially in resource-restricted embedded systems. Alternatively, we have attempted the machine virtualization approach to PEG-based parsing. MiniNez, our implemented virtual machine, is presented in this paper with several downsizing techniques, including instruction specialization, inline expansion and static flow analysis. As a result, the MiniNez machine achieves both a very small footprint and competitive performance to generated C parsers. We have demonstrated the experimental results by comparing on two major embedded platforms: Cortex-A7 and Intel Atom processor. | PEGs @cite_12 are a relatively newer formalism than Context Free Grammars, and have received much popularity among many parser researchers. There are many PEG-based parser tools such as Pappy @cite_14 , Rats ! @cite_1 , Mouse @cite_6 , Leg Peg @cite_9 , PEGjs @cite_13 , LPeg @cite_16 , Waxeye @cite_11 and so on. Most of these tools are based on a parser generator as with in Lex Yacc. An important exception is LPeg, a pattern matching library for Lua @cite_16 @cite_17 . As with MiniNez, LPeg has adopted a virtual parsing machine @cite_7 instead of parser generations. MiniNez and LPeg significantly differ in terms of backtracking handling: local static jump vs. global dynamic jump. Since the failure jump address is not pushed on the stack, MiniNez achieves a fewer stack consumption. In addition, we add several specialized downsizing to the MiniNez design and implementation. LPeg, on the other hand, does not work standalone and requires a complementary Lua runtime, resulting in hard comparison on the same condition. | {
"cite_N": [
"@cite_14",
"@cite_7",
"@cite_9",
"@cite_1",
"@cite_17",
"@cite_6",
"@cite_16",
"@cite_13",
"@cite_12",
"@cite_11"
],
"mid": [
"1971135355",
"2044735303",
"",
"2098396599",
"2091731831",
"1542200249",
"2046801051",
"",
"2018045485",
""
],
"abstract": [
"Packrat parsing is a novel technique for implementing parsers in a lazy functional programming language. A packrat parser provides the power and flexibility of top-down parsing with backtracking and unlimited lookahead, but nevertheless guarantees linear parse time. Any language defined by an LL(k) or LR(k) grammar can be recognized by a packrat parser, in addition to many languages that conventional linear-time algorithms do not support. This additional power simplifies the handling of common syntactic idioms such as the widespread but troublesome longest-match rule, enables the use of sophisticated disambiguation strategies such as syntactic and semantic predicates, provides better grammar composition properties, and allows lexical analysis to be integrated seamlessly into parsing. Yet despite its power, packrat parsing shares the same simplicity and elegance as recursive descent parsing; in fact converting a backtracking recursive descent parser into a linear-time packrat parser often involves only a fairly straightforward structural change. This paper describes packrat parsing informally with emphasis on its use in practical applications, and explores its advantages and disadvantages with respect to the more conventional alternatives.",
"Parsing Expression Grammar (PEG) is a recognition-based foundation for describing syntax that renewed interest in top-down parsing approaches. Generally, the implementation of PEGs is based on a recursive-descent parser, or uses a memoization algorithm. We present a new approach for implementing PEGs, based on a virtual parsing machine, which is more suitable for pattern matching. Each PEG has a corresponding program that is executed by the parsing machine, and new programs are dynamically created and composed. The virtual machine is embedded in a scripting language and used by a patternmatching tool. We give an operational semantics of PEGs used for pattern matching, then describe our parsing machine and its semantics. We show how to transform PEGs to parsing machine programs, and give a correctness proof of our transformation.",
"",
"We explore how to make the benefits of modularity available for syntactic specifications and present Rats!, a parser generator for Java that supports easily extensible syntax. Our parser generator builds on recent research on parsing expression grammars (PEGs), which, by being closed under composition, prioritizing choices, supporting unlimited lookahead, and integrating lexing and parsing, offer an attractive alternative to context-free grammars. PEGs are implemented by so-called packrat parsers, which are recursive descent parsers that memoize all intermediate results (hence their name). Memoization ensures linear-time performance in the presence of unlimited lookahead, but also results in an essentially lazy, functional parsing technique. In this paper, we explore how to leverage PEGs and packrat parsers as the foundation for extensible syntax. In particular, we show how make packrat parsing more widely applicable by implementing this lazy, functional technique in a strict, imperative language, while also generating better performing parsers through aggressive optimizations. Next, we develop a module system for organizing, modifying, and composing large-scale syntactic specifications. Finally, we describe a new technique for managing (global) parsing state in functional parsers. Our experimental evaluation demonstrates that the resulting parser generator succeeds at providing extensible syntax. In particular, Rats! enables other grammar writers to realize real-world language extensions in little time and code, and it generates parsers that consistently out-perform parsers created by two GLR parser generators.",
"We report on the birth and evolution of Lua and discuss how it moved from a simple configuration language to a versatile, widely used language that supports extensible semantics, anonymous functions, full lexical scoping, proper tail calls, and coroutines.",
"Two recent developments in the field of formal languages are Parsing Expression Grammar (PEG) and packrat parsing. The PEG formalism is similar to BNF, but defines syntax in terms of recognizing strings, rather than constructing them. It is, in fact, precise specification of a backtracking recursive-descent parser. Packrat parsing is a general method to handle backtracking in recursive-descent parsers. It ensures linear working time, at a huge memory cost. This paper reports an experiment that consisted of defining the syntax of Java 1.5 in PEG formalism, and literally transcribing the PEG definitions into parsing procedures (accidentally, also in Java). The resulting primitive parser shows an acceptable behavior, indicating that packrat parsing might be an overkill for practical languages. The exercise with defining the Java syntax suggests thatmore work is needed on PEG as a language specification tool.",
"Current text pattern-matching tools are based on regular expressions. However, pure regular expressions have proven too weak a formalism for the task: many interesting patterns either are difficult to describe or cannot be described by regular expressions. Moreover, the inherent non-determinism of regular expressions does not fit the need to capture specific parts of a match. Motivated by these reasons, most scripting languages nowadays use pattern-matching tools that extend the original regular-expression formalism with a set of ad hoc features, such as greedy repetitions, lazy repetitions, possessive repetitions, ‘longest-match rule,’ lookahead, etc. These ad hoc extensions bring their own set of problems, such as lack of a formal foundation and complex implementations. In this paper, we propose the use of Parsing Expression Grammars (PEGs) as a basis for pattern matching. Following this proposal, we present LPEG, a pattern-matching tool based on PEGs for the Lua scripting language. LPEG unifies the ease of use of pattern-matching tools with the full expressive power of PEGs. Because of this expressive power, it can avoid the myriad of ad hoc constructions present in several current pattern-matching tools. We also present a Parsing Machine that allows a small and efficient implementation of PEGs for pattern matching. Copyright © 2008 John Wiley & Sons, Ltd.",
"",
"For decades we have been using Chomsky's generative system of grammars, particularly context-free grammars (CFGs) and regular expressions (REs), to express the syntax of programming languages and protocols. The power of generative grammars to express ambiguity is crucial to their original purpose of modelling natural languages, but this very power makes it unnecessarily difficult both to express and to parse machine-oriented languages using CFGs. Parsing Expression Grammars (PEGs) provide an alternative, recognition-based formal foundation for describing machine-oriented syntax, which solves the ambiguity problem by not introducing ambiguity in the first place. Where CFGs express nondeterministic choice between alternatives, PEGs instead use prioritized choice. PEGs address frequently felt expressiveness limitations of CFGs and REs, simplifying syntax definitions and making it unnecessary to separate their lexical and hierarchical components. A linear-time parser can be built for any PEG, avoiding both the complexity and fickleness of LR parsers and the inefficiency of generalized CFG parsing. While PEGs provide a rich set of operators for constructing grammars, they are reducible to two minimal recognition schemas developed around 1970, TS TDPL and gTS GTDPL, which are here proven equivalent in effective recognition power.",
""
]
} |
1511.03371 | 2210522918 | Most social network analysis works at the level of interactions between users. But the vast growth in size and complexity of social networks enables us to examine interactions at larger scale. In this work we use a dataset of 76M submissions to the social network Reddit, which is organized into distinct sub-communities called subreddits. We measure the similarity between entire subreddits both in terms of user similarity and topical similarity. Our goal is to find community pairs with similar userbases, but dissimilar content; we refer to this type of relationship as a "latent interest." Detection of latent interests not only provides a perspective on individual users as they shift between roles (student, sports fan, political activist) but also gives insight into the dynamics of Reddit as a whole. Latent interest detection also has potential applications for recommendation systems and for researchers examining community evolution. | While we are aware of no work that multiple definitions of similarity to detect new types of relationships, previous work has explored the interplay of topic and social structure. For instance, in @cite_13 , the authors examine the capacity of Twitter hashtags to predict underlying social structures. In a similar vein, @cite_25 explore the extent to which Twitter acts as a news source and, separately, as a social network. Also, @cite_2 explore clustering through content and social structures using social tags on Instagram. | {
"cite_N": [
"@cite_13",
"@cite_25",
"@cite_2"
],
"mid": [
"1806085624",
"2101196063",
""
],
"abstract": [
"People's interests and people's social relationships are intuitively connected, but understanding their interplay and whether they can help predict each other has remained an open question. We examine the interface of two decisive structures forming the backbone of online social media: the graph structure of social networks - who connects with whom - and the set structure of topical affiliations - who is interested in what. In studying this interface, we identify key relationships whereby each of these structures can be understood in terms of the other. The context for our analysis is Twitter, a complex social network of both follower relationships and communication relationships. On Twitter, \"hashtags\" are used to label conversation topics, and we examine hashtag usage alongside these social structures. We find that the hashtags that users adopt can predict their social relationships, and also that the social relationships between the initial adopters of a hashtag can predict the future popularity of that hashtag. By studying weighted social relationships, we observe that while strong reciprocated ties are the easiest to predict from hashtag structure, they are also much less useful than weak directed ties for predicting hashtag popularity. Importantly, we show that computationally simple structural determinants can provide remarkable performance in both tasks. While our analyses focus on Twitter, we view our findings as broadly applicable to topical affiliations and social relationships in a host of diverse contexts, including the movies people watch, the brands people like, or the locations people frequent.",
"Twitter, a microblogging service less than three years old, commands more than 41 million users as of July 2009 and is growing fast. Twitter users tweet about any topic within the 140-character limit and follow others to receive their tweets. The goal of this paper is to study the topological characteristics of Twitter and its power as a new medium of information sharing. We have crawled the entire Twitter site and obtained 41.7 million user profiles, 1.47 billion social relations, 4,262 trending topics, and 106 million tweets. In its follower-following topology analysis we have found a non-power-law follower distribution, a short effective diameter, and low reciprocity, which all mark a deviation from known characteristics of human social networks [28]. In order to identify influentials on Twitter, we have ranked users by the number of followers and by PageRank and found two rankings to be similar. Ranking by retweets differs from the previous two rankings, indicating a gap in influence inferred from the number of followers and that from the popularity of one's tweets. We have analyzed the tweets of top trending topics and reported on their temporal behavior and user participation. We have classified the trending topics based on the active period and the tweets and show that the majority (over 85 ) of topics are headline news or persistent news in nature. A closer look at retweets reveals that any retweeted tweet is to reach an average of 1,000 users no matter what the number of followers is of the original tweet. Once retweeted, a tweet gets retweeted almost instantly on next hops, signifying fast diffusion of information after the 1st retweet. To the best of our knowledge this work is the first quantitative study on the entire Twittersphere and information diffusion on it.",
""
]
} |
1511.03371 | 2210522918 | Most social network analysis works at the level of interactions between users. But the vast growth in size and complexity of social networks enables us to examine interactions at larger scale. In this work we use a dataset of 76M submissions to the social network Reddit, which is organized into distinct sub-communities called subreddits. We measure the similarity between entire subreddits both in terms of user similarity and topical similarity. Our goal is to find community pairs with similar userbases, but dissimilar content; we refer to this type of relationship as a "latent interest." Detection of latent interests not only provides a perspective on individual users as they shift between roles (student, sports fan, political activist) but also gives insight into the dynamics of Reddit as a whole. Latent interest detection also has potential applications for recommendation systems and for researchers examining community evolution. | There also exist several topic models for uncovering latent network structure that take content and social structure into account. For instance, Topic-link LDA @cite_6 , Pairwise Link-LDA @cite_8 , and Relational Topic Models @cite_15 jointly model social structures and user content. Reddit has been specifically examined recently using backbone networks @cite_10 but without looking deeply at textual content. | {
"cite_N": [
"@cite_15",
"@cite_10",
"@cite_6",
"@cite_8"
],
"mid": [
"2123549998",
"1515908375",
"2130978632",
"2165636119"
],
"abstract": [
"We develop the relational topic model (RTM), a model of documents and the links between them. For each pair of documents, the RTM models their link as a binary random variable that is conditioned on their contents. The model can be used to summarize a network of documents, predict links between them, and predict words within them. We derive efficient inference and learning algorithms based on variational methods and evaluate the predictive performance of the RTM for large networks of scientific abstracts and web documents.",
"",
"Given a large-scale linked document collection, such as a collection of blog posts or a research literature archive, there are two fundamental problems that have generated a lot of interest in the research community. One is to identify a set of high-level topics covered by the documents in the collection; the other is to uncover and analyze the social network of the authors of the documents. So far these problems have been viewed as separate problems and considered independently from each other. In this paper we argue that these two problems are in fact inter-dependent and should be addressed together. We develop a Bayesian hierarchical approach that performs topic modeling and author community discovery in one unified framework. The effectiveness of our model is demonstrated on two blog data sets in different domains and one research paper citation data from CiteSeer.",
"In this work, we address the problem of joint modeling of text and citations in the topic modeling framework. We present two different models called the Pairwise-Link-LDA and the Link-PLSA-LDA models. The Pairwise-Link-LDA model combines the ideas of LDA [4] and Mixed Membership Block Stochastic Models [1] and allows modeling arbitrary link structure. However, the model is computationally expensive, since it involves modeling the presence or absence of a citation (link) between every pair of documents. The second model solves this problem by assuming that the link structure is a bipartite graph. As the name indicates, Link-PLSA-LDA model combines the LDA and PLSA models into a single graphical model. Our experiments on a subset of Citeseer data show that both these models are able to predict unseen data better than the baseline model of Erosheva and Lafferty [8], by capturing the notion of topical similarity between the contents of the cited and citing documents. Our experiments on two different data sets on the link prediction task show that the Link-PLSA-LDA model performs the best on the citation prediction task, while also remaining highly scalable. In addition, we also present some interesting visualizations generated by each of the models."
]
} |
1511.03416 | 2136462581 | We have seen great progress in basic perceptual tasks such as object recognition and detection. However, AI models still fail to match humans in high-level vision tasks due to the lack of capacities for deeper reasoning. Recently the new task of visual question answering (QA) has been proposed to evaluate a model’s capacity for deep image understanding. Previous works have established a loose, global association between QA sentences and images. However, many questions and answers, in practice, relate to local regions in the images. We establish a semantic link between textual descriptions and image regions by object-level grounding. It enables a new type of QA with visual answers, in addition to textual answers used in previous work. We study the visual QA tasks in a grounded setting with a large collection of 7W multiple-choice QA pairs. Furthermore, we evaluate human performance and several baseline models on the QA tasks. Finally, we propose a novel LSTM model with spatial attention to tackle the 7W QA tasks. | There have been years of effort in connecting the visual and textual information for joint learning @cite_36 @cite_38 @cite_8 @cite_40 @cite_33 @cite_12 . Image and video captioning has become a popular task in the past year @cite_50 @cite_32 @cite_25 @cite_20 @cite_51 @cite_43 . The goal is to generate text snippets to describe the images and regions instead of just predicting a few labels. Visual question answering is a natural extension to the captioning tasks, but is more interactive and has a stronger connection to real-world applications @cite_16 . | {
"cite_N": [
"@cite_38",
"@cite_33",
"@cite_8",
"@cite_36",
"@cite_32",
"@cite_16",
"@cite_43",
"@cite_40",
"@cite_50",
"@cite_20",
"@cite_51",
"@cite_25",
"@cite_12"
],
"mid": [
"2048343491",
"2149557440",
"46519926",
"2137471889",
"2951183276",
"2058556535",
"2950178297",
"",
"1895989618",
"2110933980",
"1895577753",
"2951805548",
"2125436662"
],
"abstract": [
"In this paper we exploit natural sentential descriptions of RGB-D scenes in order to improve 3D semantic parsing. Importantly, in doing so, we reason about which particular object each noun pronoun is referring to in the image. This allows us to utilize visual information in order to disambiguate the so-called coreference resolution problem that arises in text. Towards this goal, we propose a structure prediction model that exploits potentials computed from text and RGB-D imagery to reason about the class of the 3D objects, the scene type, as well as to align the nouns pronouns with the referred visual objects. We demonstrate the effectiveness of our approach on the challenging NYU-RGBD v2 dataset, which we enrich with natural lingual descriptions. We show that our approach significantly improves 3D detection and scene classification accuracy, and is able to reliably estimate the text-to-image alignment. Furthermore, by using textual and visual information, we are also able to successfully deal with coreference in text, improving upon the state-of-the-art Stanford coreference system [15].",
"Previous work on Recursive Neural Networks (RNNs) shows that these models can produce compositional feature vectors for accurately representing and classifying sentences or images. However, the sentence vectors of previous models cannot accurately represent visually grounded meaning. We introduce the DT-RNN model which uses dependency trees to embed sentences into a vector space in order to retrieve images that are described by those sentences. Unlike previous RNN-based models which use constituency trees, DT-RNNs naturally focus on the action and agents in a sentence. They are better able to abstract from the details of word order and syntactic expression. DT-RNNs outperform other recursive and recurrent neural networks, kernelized CCA and a bag-of-words baseline on the tasks of finding an image that fits a sentence description and vice versa. They also give more similar representations to sentences that describe the same image.",
"Abstract : Humans have the remarkable capability to infer the motivations of other people's actions, likely due to cognitive skills known in psychophysics as the theory of mind. In this paper, we strive to build a computational model that predicts the motivation behind the actions of people from images. To our knowledge, this challenging problem has not yet been extensively explored in computer vision. We present a novel learning based framework that uses high-level visual recognition to infer why people are performing an actions in images. However, the information in an image alone may not be sufficient to automatically solve this task. Since humans can rely on their own experiences to infer motivation, we propose to give computer vision systems access to some of these experiences by using recently developed natural language models to mine knowledge stored in massive amounts of text. While we are still far away from automatically inferring motivation, our results suggest that transferring knowledge from language into vision can help machines understand why a person might be performing an action in an image.",
"We present a new approach for modeling multi-modal data sets, focusing on the specific case of segmented images with associated text. Learning the joint distribution of image regions and words has many applications. We consider in detail predicting words associated with whole images (auto-annotation) and corresponding to particular image regions (region naming). Auto-annotation might help organize and access large collections of images. Region naming is a model of object recognition as a process of translating image regions to words, much as one might translate from one language to another. Learning the relationships between image regions and semantic correlates (words) is an interesting example of multi-modal data mining, particularly because it is typically hard to apply data mining techniques to collections of images. We develop a number of models for the joint distribution of image regions and words, including several which explicitly learn the correspondence between regions and words. We study multi-modal and correspondence extensions to Hofmann's hierarchical clustering aspect model, a translation model adapted from statistical machine translation (), and a multi-modal extension to mixture of latent Dirichlet allocation (MoM-LDA). All models are assessed using a large collection of annotated images of real scenes. We study in depth the difficult problem of measuring performance. For the annotation task, we look at prediction performance on held out data. We present three alternative measures, oriented toward different types of task. Measuring the performance of correspondence methods is harder, because one must determine whether a word has been placed on the right region of an image. We can use annotation performance as a proxy measure, but accurate measurement requires hand labeled data, and thus must occur on a smaller scale. We show results using both an annotation proxy, and manually labeled data.",
"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.",
"Visual information pervades our environment. Vision is used to decide everything from what we want to eat at a restaurant and which bus route to take to whether our clothes match and how long until the milk expires. Individually, the inability to interpret such visual information is a nuisance for blind people who often have effective, if inefficient, work-arounds to overcome them. Collectively, however, they can make blind people less independent. Specialized technology addresses some problems in this space, but automatic approaches cannot yet answer the vast majority of visual questions that blind people may have. VizWiz addresses this shortcoming by using the Internet connections and cameras on existing smartphones to connect blind people and their questions to remote paid workers' answers. VizWiz is designed to have low latency and low cost, making it both competitive with expensive automatic solutions and much more versatile.",
"Inspired by recent work in machine translation and object detection, we introduce an attention based model that automatically learns to describe the content of images. We describe how we can train this model in a deterministic manner using standard backpropagation techniques and stochastically by maximizing a variational lower bound. We also show through visualization how the model is able to automatically learn to fix its gaze on salient objects while generating the corresponding words in the output sequence. We validate the use of attention with state-of-the-art performance on three benchmark datasets: Flickr8k, Flickr30k and MS COCO.",
"",
"In this paper we explore the bi-directional mapping between images and their sentence-based descriptions. Critical to our approach is a recurrent neural network that attempts to dynamically build a visual representation of the scene as a caption is being generated or read. The representation automatically learns to remember long-term visual concepts. Our model is capable of both generating novel captions given an image, and reconstructing visual features given an image description. We evaluate our approach on several tasks. These include sentence generation, sentence retrieval and image retrieval. State-of-the-art results are shown for the task of generating novel image descriptions. When compared to human generated captions, our automatically generated captions are equal to or preferred by humans 21.0 of the time. Results are better than or comparable to state-of-the-art results on the image and sentence retrieval tasks for methods using similar visual features.",
"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.",
"Automatically describing the content of an image is a fundamental problem in artificial intelligence that connects computer vision and natural language processing. In this paper, we present a generative model based on a deep recurrent architecture that combines recent advances in computer vision and machine translation and that can be used to generate natural sentences describing an image. The model is trained to maximize the likelihood of the target description sentence given the training image. Experiments on several datasets show the accuracy of the model and the fluency of the language it learns solely from image descriptions. Our model is often quite accurate, which we verify both qualitatively and quantitatively. For instance, while the current state-of-the-art BLEU-1 score (the higher the better) on the Pascal dataset is 25, our approach yields 59, to be compared to human performance around 69. We also show BLEU-1 score improvements on Flickr30k, from 56 to 66, and on SBU, from 19 to 28. Lastly, on the newly released COCO dataset, we achieve a BLEU-4 of 27.7, which is the current state-of-the-art.",
"We present a model that generates natural language descriptions of images and their regions. Our approach leverages datasets of images and their sentence descriptions to learn about the inter-modal correspondences between language and visual data. Our alignment model is based on a novel combination of Convolutional Neural Networks over image regions, bidirectional Recurrent Neural Networks over sentences, and a structured objective that aligns the two modalities through a multimodal embedding. We then describe a Multimodal Recurrent Neural Network architecture that uses the inferred alignments to learn to generate novel descriptions of image regions. We demonstrate that our alignment model produces state of the art results in retrieval experiments on Flickr8K, Flickr30K and MSCOCO datasets. We then show that the generated descriptions significantly outperform retrieval baselines on both full images and on a new dataset of region-level annotations.",
"Sentences that describe visual scenes contain a wide variety of information pertaining to the presence of objects, their attributes and their spatial relations. In this paper we learn the visual features that correspond to semantic phrases derived from sentences. Specifically, we extract predicate tuples that contain two nouns and a relation. The relation may take several forms, such as a verb, preposition, adjective or their combination. We model a scene using a Conditional Random Field (CRF) formulation where each node corresponds to an object, and the edges to their relations. We determine the potentials of the CRF using the tuples extracted from the sentences. We generate novel scenes depicting the sentences' visual meaning by sampling from the CRF. The CRF is also used to score a set of scenes for a text-based image retrieval task. Our results show we can generate (retrieve) scenes that convey the desired semantic meaning, even when scenes (queries) are described by multiple sentences. Significant improvement is found over several baseline approaches."
]
} |
1511.03416 | 2136462581 | We have seen great progress in basic perceptual tasks such as object recognition and detection. However, AI models still fail to match humans in high-level vision tasks due to the lack of capacities for deeper reasoning. Recently the new task of visual question answering (QA) has been proposed to evaluate a model’s capacity for deep image understanding. Previous works have established a loose, global association between QA sentences and images. However, many questions and answers, in practice, relate to local regions in the images. We establish a semantic link between textual descriptions and image regions by object-level grounding. It enables a new type of QA with visual answers, in addition to textual answers used in previous work. We study the visual QA tasks in a grounded setting with a large collection of 7W multiple-choice QA pairs. Furthermore, we evaluate human performance and several baseline models on the QA tasks. Finally, we propose a novel LSTM model with spatial attention to tackle the 7W QA tasks. | Question answering in NLP has been a well-established problem. Successful applications can be seen in voice assistants in mobile devices, search engines and game shows (e.g., IBM Waston). Traditional question answering system relies on an elaborate pipeline of models involving natural language parsing, knowledge base querying, and answer generation @cite_47 . Recent neural network models attempt to learn end-to-end directly from questions and answers @cite_26 @cite_44 . | {
"cite_N": [
"@cite_44",
"@cite_47",
"@cite_26"
],
"mid": [
"1525961042",
"2171278097",
"2130237711"
],
"abstract": [
"One long-term goal of machine learning research is to produce methods that are applicable to reasoning and natural language, in particular building an intelligent dialogue agent. To measure progress towards that goal, we argue for the usefulness of a set of proxy tasks that evaluate reading comprehension via question answering. Our tasks measure understanding in several ways: whether a system is able to answer questions via chaining facts, simple induction, deduction and many more. The tasks are designed to be prerequisites for any system that aims to be capable of conversing with a human. We believe many existing learning systems can currently not solve them, and hence our aim is to classify these tasks into skill sets, so that researchers can identify (and then rectify) the failings of their systems. We also extend and improve the recently introduced Memory Networks model, and show it is able to solve some, but not all, of the tasks.",
"IBM Research undertook a challenge to build a computer system that could compete at the human champion level in real time on the American TV Quiz show, Jeopardy! The extent of the challenge includes fielding a real-time automatic contestant on the show, not merely a laboratory exercise. The Jeopardy! Challenge helped us address requirements that led to the design of the DeepQA architecture and the implementation of Watson. After 3 years of intense research and development by a core team of about 20 researches, Watson is performing at human expert-levels in terms of precision, confidence and speed at the Jeopardy! Quiz show. Our results strongly suggest that DeepQA is an effective and extensible architecture that may be used as a foundation for combining, deploying, evaluating and advancing a wide range of algorithmic techniques to rapidly advance the field of QA.",
"Text classification methods for tasks like factoid question answering typically use manually defined string matching rules or bag of words representations. These methods are ineective when question text contains very few individual words (e.g., named entities) that are indicative of the answer. We introduce a recursive neural network (rnn) model that can reason over such input by modeling textual compositionality. We apply our model, qanta, to a dataset of questions from a trivia competition called quiz bowl. Unlike previous rnn models, qanta learns word and phrase-level representations that combine across sentences to reason about entities. The model outperforms multiple baselines and, when combined with information retrieval methods, rivals the best human players."
]
} |
1511.03416 | 2136462581 | We have seen great progress in basic perceptual tasks such as object recognition and detection. However, AI models still fail to match humans in high-level vision tasks due to the lack of capacities for deeper reasoning. Recently the new task of visual question answering (QA) has been proposed to evaluate a model’s capacity for deep image understanding. Previous works have established a loose, global association between QA sentences and images. However, many questions and answers, in practice, relate to local regions in the images. We establish a semantic link between textual descriptions and image regions by object-level grounding. It enables a new type of QA with visual answers, in addition to textual answers used in previous work. We study the visual QA tasks in a grounded setting with a large collection of 7W multiple-choice QA pairs. Furthermore, we evaluate human performance and several baseline models on the QA tasks. Finally, we propose a novel LSTM model with spatial attention to tackle the 7W QA tasks. | @cite_42 proposed a restricted visual Turing test to evaluate visual understanding. The DAQUAR dataset is the first toy-sized QA benchmark built upon indoor scene RGB-D images. Most of the other datasets @cite_3 @cite_15 @cite_13 @cite_14 collected QA pairs on Microsoft COCO images @cite_31 , either generated automatically by NLP tools @cite_13 or written by human workers @cite_3 @cite_15 @cite_14 . Following these datasets, an array of models has been proposed to tackle the visual QA tasks. The proposed models range from probabilistic inference @cite_6 @cite_22 @cite_0 and recurrent neural networks @cite_3 @cite_15 @cite_5 @cite_13 to convolutional networks @cite_48 . Previous visual QA datasets evaluate textual answers on images while omitting the links between the object mentions and their visual appearances. Inspired by @cite_42 , we establish the link by grounding objects in the images and perform experiments in the grounded QA setting. | {
"cite_N": [
"@cite_31",
"@cite_14",
"@cite_22",
"@cite_48",
"@cite_42",
"@cite_3",
"@cite_6",
"@cite_0",
"@cite_5",
"@cite_15",
"@cite_13"
],
"mid": [
"1861492603",
"627986001",
"1895641373",
"1606748815",
"1983927101",
"2950761309",
"2151498684",
"1891689858",
"2952246170",
"1488163396",
"2949218037"
],
"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 introduce a new dataset consisting of 360,001 focused natural language descriptions for 10,738 images. This dataset, the Visual Madlibs dataset, is collected using automatically produced fill-in-the-blank templates designed to gather targeted descriptions about: people and objects, their appearances, activities, and interactions, as well as inferences about the general scene or its broader context. We provide several analyses of the Visual Madlibs dataset and demonstrate its applicability to two new description generation tasks: focused description generation, and multiple-choice question-answering for images. Experiments using joint-embedding and deep learning methods show promising results on these tasks.",
"This article proposes a multimedia analysis framework to process video and text jointly for understanding events and answering user queries. The framework produces a parse graph that represents the compositional structures of spatial information (objects and scenes), temporal information (actions and events), and causal information (causalities between events and fluents) in the video and text. The knowledge representation of the framework is based on a spatial-temporal-causal AND-OR graph (S T C-AOG), which jointly models possible hierarchical compositions of objects, scenes, and events as well as their interactions and mutual contexts, and specifies the prior probabilistic distribution of the parse graphs. The authors present a probabilistic generative model for joint parsing that captures the relations between the input video text, their corresponding parse graphs, and the joint parse graph. Based on the probabilistic model, the authors propose a joint parsing system consisting of three modules: video parsing, text parsing, and joint inference. Video parsing and text parsing produce two parse graphs from the input video and text, respectively. The joint inference module produces a joint parse graph by performing matching, deduction, and revision on the video and text parse graphs. The proposed framework has the following objectives: to provide deep semantic parsing of video and text that goes beyond the traditional bag-of-words approaches; to perform parsing and reasoning across the spatial, temporal, and causal dimensions based on the joint S T C-AOG representation; and to show that deep joint parsing facilitates subsequent applications such as generating narrative text descriptions and answering queries in the forms of who, what, when, where, and why. The authors empirically evaluated the system based on comparison against ground-truth as well as accuracy of query answering and obtained satisfactory results.",
"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.",
"Today, computer vision systems are tested by their accuracy in detecting and localizing instances of objects. As an alternative, and motivated by the ability of humans to provide far richer descriptions and even tell a story about an image, we construct a “visual Turing test”: an operator-assisted device that produces a stochastic sequence of binary questions from a given test image. The query engine proposes a question; the operator either provides the correct answer or rejects the question as ambiguous; the engine proposes the next question (“just-in-time truthing”). The test is then administered to the computer-vision system, one question at a time. After the system’s answer is recorded, the system is provided the correct answer and the next question. Parsing is trivial and deterministic; the system being tested requires no natural language processing. The query engine employs statistical constraints, learned from a training set, to produce questions with essentially unpredictable answers—the answer to a question, given the history of questions and their correct answers, is nearly equally likely to be positive or negative. In this sense, the test is only about vision. The system is designed to produce streams of questions that follow natural story lines, from the instantiation of a unique object, through an exploration of its properties, and on to its relationships with other uniquely instantiated objects.",
"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 propose a method for automatically answering questions about images by bringing together recent advances from natural language processing and computer vision. We combine discrete reasoning with uncertain predictions by a multi-world approach that represents uncertainty about the perceived world in a bayesian framework. Our approach can handle human questions of high complexity about realistic scenes and replies with range of answer like counts, object classes, instances and lists of them. The system is directly trained from question-answer pairs. We establish a first benchmark for this task that can be seen as a modern attempt at a visual turing test.",
"Reasoning about objects and their affordances is a fundamental problem for visual intelligence. Most of the previous work casts this problem as a classification task where separate classifiers are trained to label objects, recognize attributes, or assign affordances. In this work, we consider the problem of object affordance reasoning using a knowledge base representation. Diverse information of objects are first harvested from images and other meta-data sources. We then learn a knowledge base (KB) using a Markov Logic Network (MLN). Given the learned KB, we show that a diverse set of visual inference tasks can be done in this unified framework without training separate classifiers, including zero-shot affordance prediction and object recognition given human poses.",
"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.",
"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",
"This work aims to address the problem of image-based question-answering (QA) with new models and datasets. In our work, we propose to use neural networks and visual semantic embeddings, without intermediate stages such as object detection and image segmentation, to predict answers to simple questions about images. Our model performs 1.8 times better than the only published results on an existing image QA dataset. We also present a question generation algorithm that converts image descriptions, which are widely available, into QA form. We used this algorithm to produce an order-of-magnitude larger dataset, with more evenly distributed answers. A suite of baseline results on this new dataset are also presented."
]
} |
1511.03351 | 2152268600 | Media sharing is an extremely popular paradigm of social interaction in online social networks (OSNs) nowadays. The scalable media access control is essential to perform information sharing among users with various access privileges. In this paper, we present a multi-dimensional scalable media access control (MD-SMAC) system based on the proposed scalable ciphertext policy attribute-based encryption (SCP-ABE) algorithm. In the proposed MD-SMAC system, fine-grained access control can be performed on the media contents encoded in a multi-dimensional scalable manner based on data consumers' diverse attributes. Through security analysis, we show that the proposed MC-SMAC system is able to resist collusion attacks. Additionally, we conduct experiments to evaluate the efficiency performance of the proposed system, especially on mobile devices. | Unfortunately, such method may result in user collusion when the media layers are organized following the multi-dimensional scalable format. That is two users may collude with each other to generate a valid but illegal access key for unauthorized higher level access. The resistance of such collusion attack requires increasing the number of key segments. From the results shown in @cite_9 , the best performance of a collusion secure scheme needs @math key segments under a @math -by- @math -by- @math ( @math ) data structure, but at the same time results in @math more hash calculations. As a result, the communication cost saved by segmenting access keys is less significant, while the resulted computation cost is more than expected. Therefore, the aim to reduce the cost of key distribution is not sufficiently meaningful. | {
"cite_N": [
"@cite_9"
],
"mid": [
"2081768367"
],
"abstract": [
"This paper proposes a novel key management scheme for multi-dimensional scalable multimedia access control. We build up a collusion-attack model and prove that the proposed scheme is indeed resilient to collusion attack. The lower bound of number of the segment keys is also given in the analysis. We consider the key management problem on partially order set and propose a general approach consists of a poset decomposition step and an element projection step. Under the proposed framework, a novel hierarchical 3D poset decomposition approach is further developed. The proposed scheme is the first provable collusion resilient scheme in 3-dimension (and higher) scenario. It can be easily adapted to other scalable dependence structure and higher dimension. Analysis on the proposed scheme proves its collusion-resilient property."
]
} |
1511.03690 | 2137010615 | In this paper, we present a model which takes as input a corpus of images with relevant spoken captions and finds a correspondence between the two modalities. We employ a pair of convolutional neural networks to model visual objects and speech signals at the word level, and tie the networks together with an embedding and alignment model which learns a joint semantic space over both modalities. We evaluate our model using image search and annotation tasks on the Flickr8k dataset, which we augmented by collecting a corpus of 40,000 spoken captions using Amazon Mechanical Turk. | Multimodal modeling of images and text has been an extremely popular pursuit in the machine learning field during the past decade, with many approaches focusing on accurately annotating objects and regions within images. For example, @cite_5 relied on pre-segmented and labelled images to estimate joint distributions over words and objects, while Socher @cite_2 learned a latent meaning space covering images and words learned on non-parallel data. While these approaches focused on improving the identification of visual objects from a pool of predefined classes, other research has studied the problem of aligning text to the images or videos they describe. For example, @cite_21 took visual scenes with high level captions, parsed the text, detected visual objects, and then aligned the two modalities with a Markov random field. @cite_11 aligned semantic graphs over text queries to relational graphs over objects in videos to perform natural language video search. @cite_10 employed separate classifiers over text and visual objects that shared the same label sets. | {
"cite_N": [
"@cite_21",
"@cite_2",
"@cite_5",
"@cite_10",
"@cite_11"
],
"mid": [
"2048343491",
"2062955551",
"2137471889",
"1923162067",
"2086842362"
],
"abstract": [
"In this paper we exploit natural sentential descriptions of RGB-D scenes in order to improve 3D semantic parsing. Importantly, in doing so, we reason about which particular object each noun pronoun is referring to in the image. This allows us to utilize visual information in order to disambiguate the so-called coreference resolution problem that arises in text. Towards this goal, we propose a structure prediction model that exploits potentials computed from text and RGB-D imagery to reason about the class of the 3D objects, the scene type, as well as to align the nouns pronouns with the referred visual objects. We demonstrate the effectiveness of our approach on the challenging NYU-RGBD v2 dataset, which we enrich with natural lingual descriptions. We show that our approach significantly improves 3D detection and scene classification accuracy, and is able to reliably estimate the text-to-image alignment. Furthermore, by using textual and visual information, we are also able to successfully deal with coreference in text, improving upon the state-of-the-art Stanford coreference system [15].",
"We propose a semi-supervised model which segments and annotates images using very few labeled images and a large unaligned text corpus to relate image regions to text labels. Given photos of a sports event, all that is necessary to provide a pixel-level labeling of objects and background is a set of newspaper articles about this sport and one to five labeled images. Our model is motivated by the observation that words in text corpora share certain context and feature similarities with visual objects. We describe images using visual words, a new region-based representation. The proposed model is based on kernelized canonical correlation analysis which finds a mapping between visual and textual words by projecting them into a latent meaning space. Kernels are derived from context and adjective features inside the respective visual and textual domains. We apply our method to a challenging dataset and rely on articles of the New York Times for textual features. Our model outperforms the state-of-the-art in annotation. In segmentation it compares favorably with other methods that use significantly more labeled training data.",
"We present a new approach for modeling multi-modal data sets, focusing on the specific case of segmented images with associated text. Learning the joint distribution of image regions and words has many applications. We consider in detail predicting words associated with whole images (auto-annotation) and corresponding to particular image regions (region naming). Auto-annotation might help organize and access large collections of images. Region naming is a model of object recognition as a process of translating image regions to words, much as one might translate from one language to another. Learning the relationships between image regions and semantic correlates (words) is an interesting example of multi-modal data mining, particularly because it is typically hard to apply data mining techniques to collections of images. We develop a number of models for the joint distribution of image regions and words, including several which explicitly learn the correspondence between regions and words. We study multi-modal and correspondence extensions to Hofmann's hierarchical clustering aspect model, a translation model adapted from statistical machine translation (), and a multi-modal extension to mixture of latent Dirichlet allocation (MoM-LDA). All models are assessed using a large collection of annotated images of real scenes. We study in depth the difficult problem of measuring performance. For the annotation task, we look at prediction performance on held out data. We present three alternative measures, oriented toward different types of task. Measuring the performance of correspondence methods is harder, because one must determine whether a word has been placed on the right region of an image. We can use annotation performance as a proxy measure, but accurate measurement requires hand labeled data, and thus must occur on a smaller scale. We show results using both an annotation proxy, and manually labeled data.",
"As robots become more ubiquitous and capable, it becomes ever more important for untrained users to easily interact with them. Recently, this has led to study of the language grounding problem, where the goal is to extract representations of the meanings of natural language tied to the physical world. We present an approach for joint learning of language and perception models for grounded attribute induction. The perception model includes classifiers for physical characteristics and a language model based on a probabilistic categorial grammar that enables the construction of compositional meaning representations. We evaluate on the task of interpreting sentences that describe sets of objects in a physical workspace, and demonstrate accurate task performance and effective latent-variable concept induction in physical grounded scenes.",
"In this paper, we tackle the problem of retrieving videos using complex natural language queries. Towards this goal, we first parse the sentential descriptions into a semantic graph, which is then matched to visual concepts using a generalized bipartite matching algorithm. Our approach exploits object appearance, motion and spatial relations, and learns the importance of each term using structure prediction. We demonstrate the effectiveness of our approach on a new dataset designed for semantic search in the context of autonomous driving, which exhibits complex and highly dynamic scenes with many objects. We show that our approach is able to locate a major portion of the objects described in the query with high accuracy, and improve the relevance in video retrieval."
]
} |
1511.03483 | 2559471018 | An important question in evolutionary computation is how good solutions evolutionary algorithms can produce. This paper aims to provide an analytic analysis of solution quality in terms of the relative approximation error, which is defined by the error between 1 and the approximation ratio of the solution found by an evolutionary algorithm. Since evolutionary algorithms are iterative methods, the relative approximation error is a function of generations. With the help of matrix analysis, it is possible to obtain an exact expression of such a function. In this paper, an analytic expression for calculating the relative approximation error is presented for a class of evolutionary algorithms, that is, (1+1) strictly elitist evolution algorithms. Furthermore, analytic expressions of the fitness value and the average convergence rate in each generation are also derived for this class of evolutionary algorithms. The approach is promising, and it can be extended to non-elitist or population-based algorithms too. | The relative approximation error belongs to the convergence rate study of EAs, which can be traced back to 1990s @cite_22 @cite_13 @cite_16 . This paper only investigates EAs for discrete optimisation, although the converegnce rate of EAs for continuous optimization @cite_18 @cite_11 @cite_21 is also important. EAs belong to iterative methods. A fundamental question in iterative methods is the convergence rate, which can be formalised as follow @cite_16 . Since the @math th generation solution is a random variable, we let @math be a vector representing its probability distribution over the search space, @math a vector such that @math for the optimal solution set @math and @math for the non-optimal solution set @math . The convergence rate problem asks the question how fast @math converges to @math . The goal is to obtain a bound @math such that @math , where @math is a norm. There are various ways to assign the norm. For example, if the @math th generation solution is a binary string @math and the optimal solution is @math , the norm is set to be the Hamming distance: In the current paper, the norm is set to be the relative approximation error @math . | {
"cite_N": [
"@cite_18",
"@cite_22",
"@cite_21",
"@cite_16",
"@cite_13",
"@cite_11"
],
"mid": [
"2123552742",
"2167325880",
"2122267698",
"2053135339",
"2003186787",
""
],
"abstract": [
"The standard choice for mutating an individual of an evolutionary algorithm with continuous variables is the normal distribution; however other distributions, especially some versions of the multivariate Cauchy distribution, have recently gained increased popularity in practical applications. Here the extent to which Cauchy mutation distributions may affect the local convergence behavior of evolutionary algorithms is analyzed. The results show that the order of local convergence is identical for Gaussian and spherical Cauchy distributions, whereas nonspherical Cauchy mutations lead to slower local convergence. As a by-product of the analysis, some recommendations for the parametrization of the self-adaptive step size control mechanism can be derived.",
"This paper addresses a Markov chain analysis of genetic algorithms (GAs), in particular for a variety called a modified elitist strategy. The modified elitist strategy generates the current population of M individuals by reserving the individual with the highest fitness value from the previous generation and generating M-1 individuals through a generation change. The author's analysis is based on a Markov chain: by assuming a simple GA in which the genetic operation in the generation changes is restricted to selection, crossover, and mutation, and by evaluating the eigenvalues of the transition matrix of the Markov chain, the convergence rate of the GAs is computed in terms of a mutation probability spl mu . In this way, the authors show the probability that the population includes the individual with the highest fitness value is lower-bounded by 1-O(| spl lambda *| sup n ), | spl lambda *| >",
"The best achievable convergence rates of mutation-based evolutionary algorithms are known for various characteristic test problems. Most results are available for convex quadratic functions with Hessians of full rank. Here, we prove that linear convergence rates are achievable for convex quadratic functions even though the Hessians are rank-deficient. This result has immediate implications for recent convergence results for certain evolutionary algorithms for bi-objective optimization problems.",
"This paper discusses the convergence rates of genetic algorithms by using the minorization condition in the Markov chain theory. We classify genetic algorithms into two kinds: one with time-invariant genetic operators, another with time-variant genetic operators. For the former case, we have obtained the bound on its convergence rate on the general state space; for the later case, we have bounded its convergence rate on the finite state space.",
"This paper shows a theoretical property on the Markov chain of genetic algorithms: the stationary distribution focuses on the uniform population with the optimal solution as mutation and crossover probabilities go to zero and some selective pressure defined in this paper goes to infinity. Moreover, as a result, a sufficient condition for ergodicity is derived when a simulated annealing-like strategy is considered. Additionally, the uniform crossover counterpart of the Vose-Liepins formula is derived using the Markov chain model.",
""
]
} |
1511.03483 | 2559471018 | An important question in evolutionary computation is how good solutions evolutionary algorithms can produce. This paper aims to provide an analytic analysis of solution quality in terms of the relative approximation error, which is defined by the error between 1 and the approximation ratio of the solution found by an evolutionary algorithm. Since evolutionary algorithms are iterative methods, the relative approximation error is a function of generations. With the help of matrix analysis, it is possible to obtain an exact expression of such a function. In this paper, an analytic expression for calculating the relative approximation error is presented for a class of evolutionary algorithms, that is, (1+1) strictly elitist evolution algorithms. Furthermore, analytic expressions of the fitness value and the average convergence rate in each generation are also derived for this class of evolutionary algorithms. The approach is promising, and it can be extended to non-elitist or population-based algorithms too. | The research in this paper is also linked to fixed budge analysis. Jansen and Zarges @cite_12 @cite_4 proposed fixed budget analysis. It aims to find lower and upper bounds @math and @math such that @math usually for a fixed budget @math . They investigated two algorithms: random local search and the @math EA and obtained such bounds on several pseudo-Boolean optimization problems. | {
"cite_N": [
"@cite_4",
"@cite_12"
],
"mid": [
"2111575001",
"2041706554"
],
"abstract": [
"When for a difficult real-world optimisation problem no good problem-specific algorithm is available often randomised search heuristics are used. They are hoped to deliver good solutions in acceptable time. The theoretical analysis usually concentrates on the average time needed to find an optimal or approximately optimal solution. This matches neither the application in practice nor the empirical analysis since usually optimal solutions are not known and even if found cannot be recognised. More often the algorithms are stopped after some time. This motivates a theoretical analysis to concentrate on the quality of the best solution obtained after a pre-specified number of function evaluations called budget. Using this perspective two simple randomised search heuristics, random local search and the (1+1) evolutionary algorithm, are analysed on some well-known example problems. Upper and lower bounds on the expected quality of a solution for a fixed budget of function evaluations are proven. The analysis shows novel and challenging problems in the study of randomised search heuristics. It demonstrates the potential of this shift in perspective from expected run time to expected solution quality.",
"Randomised search heuristics are used in practice to solve difficult problems where no good problem-specific algorithm is known. They deliver a solution of acceptable quality in reasonable time in many cases. When theoretically analysing the performance of randomised search heuristics one usually considers the average time needed to find an optimal solution or one of a pre-specified approximation quality. This is very different from practice where usually the algorithm is stopped after some time. For a theoretical analysis this corresponds to investigating the quality of the solution obtained after a pre-specified number of function evaluations called budget. Such a perspective is taken here and two simple randomised search heuristics, random local search and the (1+1) evolutionary algorithm, are analysed on simple and well-known example functions. If the budget is significantly smaller than the expected time needed for optimisation the behaviour of the algorithms can be very different depending on the problem at hand. Precise analytical results are proven. They demonstrate novel and interesting challenges in the analysis of randomised search heuristics. The potential of this different perspective to provide a more practically useful theory is shown."
]
} |
1511.03483 | 2559471018 | An important question in evolutionary computation is how good solutions evolutionary algorithms can produce. This paper aims to provide an analytic analysis of solution quality in terms of the relative approximation error, which is defined by the error between 1 and the approximation ratio of the solution found by an evolutionary algorithm. Since evolutionary algorithms are iterative methods, the relative approximation error is a function of generations. With the help of matrix analysis, it is possible to obtain an exact expression of such a function. In this paper, an analytic expression for calculating the relative approximation error is presented for a class of evolutionary algorithms, that is, (1+1) strictly elitist evolution algorithms. Furthermore, analytic expressions of the fitness value and the average convergence rate in each generation are also derived for this class of evolutionary algorithms. The approach is promising, and it can be extended to non-elitist or population-based algorithms too. | The convergence rate study can implement the same task as fixed budget analysis does. Provided that @math , it is trivial to derive @math . Nevertheless, there are significant differences between the convergence rate study and fixed budget analysis. The convergence rate study has existed in EAs for more than two decades @cite_22 @cite_13 @cite_16 . Fixed budget analysis was recently proposed by Jansen and Zarges @cite_4 . | {
"cite_N": [
"@cite_16",
"@cite_13",
"@cite_22",
"@cite_4"
],
"mid": [
"2053135339",
"2003186787",
"2167325880",
"2111575001"
],
"abstract": [
"This paper discusses the convergence rates of genetic algorithms by using the minorization condition in the Markov chain theory. We classify genetic algorithms into two kinds: one with time-invariant genetic operators, another with time-variant genetic operators. For the former case, we have obtained the bound on its convergence rate on the general state space; for the later case, we have bounded its convergence rate on the finite state space.",
"This paper shows a theoretical property on the Markov chain of genetic algorithms: the stationary distribution focuses on the uniform population with the optimal solution as mutation and crossover probabilities go to zero and some selective pressure defined in this paper goes to infinity. Moreover, as a result, a sufficient condition for ergodicity is derived when a simulated annealing-like strategy is considered. Additionally, the uniform crossover counterpart of the Vose-Liepins formula is derived using the Markov chain model.",
"This paper addresses a Markov chain analysis of genetic algorithms (GAs), in particular for a variety called a modified elitist strategy. The modified elitist strategy generates the current population of M individuals by reserving the individual with the highest fitness value from the previous generation and generating M-1 individuals through a generation change. The author's analysis is based on a Markov chain: by assuming a simple GA in which the genetic operation in the generation changes is restricted to selection, crossover, and mutation, and by evaluating the eigenvalues of the transition matrix of the Markov chain, the convergence rate of the GAs is computed in terms of a mutation probability spl mu . In this way, the authors show the probability that the population includes the individual with the highest fitness value is lower-bounded by 1-O(| spl lambda *| sup n ), | spl lambda *| >",
"When for a difficult real-world optimisation problem no good problem-specific algorithm is available often randomised search heuristics are used. They are hoped to deliver good solutions in acceptable time. The theoretical analysis usually concentrates on the average time needed to find an optimal or approximately optimal solution. This matches neither the application in practice nor the empirical analysis since usually optimal solutions are not known and even if found cannot be recognised. More often the algorithms are stopped after some time. This motivates a theoretical analysis to concentrate on the quality of the best solution obtained after a pre-specified number of function evaluations called budget. Using this perspective two simple randomised search heuristics, random local search and the (1+1) evolutionary algorithm, are analysed on some well-known example problems. Upper and lower bounds on the expected quality of a solution for a fixed budget of function evaluations are proven. The analysis shows novel and challenging problems in the study of randomised search heuristics. It demonstrates the potential of this shift in perspective from expected run time to expected solution quality."
]
} |
1511.03483 | 2559471018 | An important question in evolutionary computation is how good solutions evolutionary algorithms can produce. This paper aims to provide an analytic analysis of solution quality in terms of the relative approximation error, which is defined by the error between 1 and the approximation ratio of the solution found by an evolutionary algorithm. Since evolutionary algorithms are iterative methods, the relative approximation error is a function of generations. With the help of matrix analysis, it is possible to obtain an exact expression of such a function. In this paper, an analytic expression for calculating the relative approximation error is presented for a class of evolutionary algorithms, that is, (1+1) strictly elitist evolution algorithms. Furthermore, analytic expressions of the fitness value and the average convergence rate in each generation are also derived for this class of evolutionary algorithms. The approach is promising, and it can be extended to non-elitist or population-based algorithms too. | The convergence rate study focuses on estimating the error @math , where the norm @math can be chosen as the absolute error @math , relative error @math or Hamming distance. Fixed budget analysis aims at bounding @math for a fixed budget @math (that is a fixed number of generations) @cite_4 . | {
"cite_N": [
"@cite_4"
],
"mid": [
"2111575001"
],
"abstract": [
"When for a difficult real-world optimisation problem no good problem-specific algorithm is available often randomised search heuristics are used. They are hoped to deliver good solutions in acceptable time. The theoretical analysis usually concentrates on the average time needed to find an optimal or approximately optimal solution. This matches neither the application in practice nor the empirical analysis since usually optimal solutions are not known and even if found cannot be recognised. More often the algorithms are stopped after some time. This motivates a theoretical analysis to concentrate on the quality of the best solution obtained after a pre-specified number of function evaluations called budget. Using this perspective two simple randomised search heuristics, random local search and the (1+1) evolutionary algorithm, are analysed on some well-known example problems. Upper and lower bounds on the expected quality of a solution for a fixed budget of function evaluations are proven. The analysis shows novel and challenging problems in the study of randomised search heuristics. It demonstrates the potential of this shift in perspective from expected run time to expected solution quality."
]
} |
1511.03483 | 2559471018 | An important question in evolutionary computation is how good solutions evolutionary algorithms can produce. This paper aims to provide an analytic analysis of solution quality in terms of the relative approximation error, which is defined by the error between 1 and the approximation ratio of the solution found by an evolutionary algorithm. Since evolutionary algorithms are iterative methods, the relative approximation error is a function of generations. With the help of matrix analysis, it is possible to obtain an exact expression of such a function. In this paper, an analytic expression for calculating the relative approximation error is presented for a class of evolutionary algorithms, that is, (1+1) strictly elitist evolution algorithms. Furthermore, analytic expressions of the fitness value and the average convergence rate in each generation are also derived for this class of evolutionary algorithms. The approach is promising, and it can be extended to non-elitist or population-based algorithms too. | In the convergence rate study, the upper bound @math on @math usually is an exponential function or combination of linear functions of @math @cite_16 . In fixed budget analysis, the bound @math on @math may not be an exponential function of @math @cite_4 . | {
"cite_N": [
"@cite_16",
"@cite_4"
],
"mid": [
"2053135339",
"2111575001"
],
"abstract": [
"This paper discusses the convergence rates of genetic algorithms by using the minorization condition in the Markov chain theory. We classify genetic algorithms into two kinds: one with time-invariant genetic operators, another with time-variant genetic operators. For the former case, we have obtained the bound on its convergence rate on the general state space; for the later case, we have bounded its convergence rate on the finite state space.",
"When for a difficult real-world optimisation problem no good problem-specific algorithm is available often randomised search heuristics are used. They are hoped to deliver good solutions in acceptable time. The theoretical analysis usually concentrates on the average time needed to find an optimal or approximately optimal solution. This matches neither the application in practice nor the empirical analysis since usually optimal solutions are not known and even if found cannot be recognised. More often the algorithms are stopped after some time. This motivates a theoretical analysis to concentrate on the quality of the best solution obtained after a pre-specified number of function evaluations called budget. Using this perspective two simple randomised search heuristics, random local search and the (1+1) evolutionary algorithm, are analysed on some well-known example problems. Upper and lower bounds on the expected quality of a solution for a fixed budget of function evaluations are proven. The analysis shows novel and challenging problems in the study of randomised search heuristics. It demonstrates the potential of this shift in perspective from expected run time to expected solution quality."
]
} |
1511.03483 | 2559471018 | An important question in evolutionary computation is how good solutions evolutionary algorithms can produce. This paper aims to provide an analytic analysis of solution quality in terms of the relative approximation error, which is defined by the error between 1 and the approximation ratio of the solution found by an evolutionary algorithm. Since evolutionary algorithms are iterative methods, the relative approximation error is a function of generations. With the help of matrix analysis, it is possible to obtain an exact expression of such a function. In this paper, an analytic expression for calculating the relative approximation error is presented for a class of evolutionary algorithms, that is, (1+1) strictly elitist evolution algorithms. Furthermore, analytic expressions of the fitness value and the average convergence rate in each generation are also derived for this class of evolutionary algorithms. The approach is promising, and it can be extended to non-elitist or population-based algorithms too. | In the convergence rate, the bound on @math holds for all @math . But in fixed budget analysis, the bound on @math often is estimated for a fixed budget @math @cite_4 . | {
"cite_N": [
"@cite_4"
],
"mid": [
"2111575001"
],
"abstract": [
"When for a difficult real-world optimisation problem no good problem-specific algorithm is available often randomised search heuristics are used. They are hoped to deliver good solutions in acceptable time. The theoretical analysis usually concentrates on the average time needed to find an optimal or approximately optimal solution. This matches neither the application in practice nor the empirical analysis since usually optimal solutions are not known and even if found cannot be recognised. More often the algorithms are stopped after some time. This motivates a theoretical analysis to concentrate on the quality of the best solution obtained after a pre-specified number of function evaluations called budget. Using this perspective two simple randomised search heuristics, random local search and the (1+1) evolutionary algorithm, are analysed on some well-known example problems. Upper and lower bounds on the expected quality of a solution for a fixed budget of function evaluations are proven. The analysis shows novel and challenging problems in the study of randomised search heuristics. It demonstrates the potential of this shift in perspective from expected run time to expected solution quality."
]
} |
1511.03292 | 2226480163 | In this paper we propose the construction of linguistic descriptions of images. This is achieved through the extraction of scene description graphs (SDGs) from visual scenes using an automatically constructed knowledge base. SDGs are constructed using both vision and reasoning. Specifically, commonsense reasoning is applied on (a) detections obtained from existing perception methods on given images, (b) a "commonsense" knowledge base constructed using natural language processing of image annotations and (c) lexical ontological knowledge from resources such as WordNet. Amazon Mechanical Turk(AMT)-based evaluations on Flickr8k, Flickr30k and MS-COCO datasets show that in most cases, sentences auto-constructed from SDGs obtained by our method give a more relevant and thorough description of an image than a recent state-of-the-art image caption based approach. Our Image-Sentence Alignment Evaluation results are also comparable to that of the recent state-of-the art approaches. | Our work is influenced by various lines of work where researchers have proposed approaches to extract meaningful information from images and videos. As @cite_14 suggests, such works can be categorized into 1) dense image annotations, 2) generating textual descriptions, 3) grounding natural language in images and 4) neural networks in visual and language domains. | {
"cite_N": [
"@cite_14"
],
"mid": [
"2951805548"
],
"abstract": [
"We present a model that generates natural language descriptions of images and their regions. Our approach leverages datasets of images and their sentence descriptions to learn about the inter-modal correspondences between language and visual data. Our alignment model is based on a novel combination of Convolutional Neural Networks over image regions, bidirectional Recurrent Neural Networks over sentences, and a structured objective that aligns the two modalities through a multimodal embedding. We then describe a Multimodal Recurrent Neural Network architecture that uses the inferred alignments to learn to generate novel descriptions of image regions. We demonstrate that our alignment model produces state of the art results in retrieval experiments on Flickr8K, Flickr30K and MSCOCO datasets. We then show that the generated descriptions significantly outperform retrieval baselines on both full images and on a new dataset of region-level annotations."
]
} |
1511.03292 | 2226480163 | In this paper we propose the construction of linguistic descriptions of images. This is achieved through the extraction of scene description graphs (SDGs) from visual scenes using an automatically constructed knowledge base. SDGs are constructed using both vision and reasoning. Specifically, commonsense reasoning is applied on (a) detections obtained from existing perception methods on given images, (b) a "commonsense" knowledge base constructed using natural language processing of image annotations and (c) lexical ontological knowledge from resources such as WordNet. Amazon Mechanical Turk(AMT)-based evaluations on Flickr8k, Flickr30k and MS-COCO datasets show that in most cases, sentences auto-constructed from SDGs obtained by our method give a more relevant and thorough description of an image than a recent state-of-the-art image caption based approach. Our Image-Sentence Alignment Evaluation results are also comparable to that of the recent state-of-the art approaches. | According to the above categorization, we share our roots with the works of generating textual descriptions. This includes the works that retrieves and ranks sentences from training sets given an image such as @cite_37 , @cite_44 , @cite_21 , @cite_7 . @cite_40 , @cite_16 , @cite_29 , @cite_18 , @cite_47 are some of the works that have generated descriptions by stitching together annotations or applying templates on detected image content. | {
"cite_N": [
"@cite_37",
"@cite_18",
"@cite_7",
"@cite_29",
"@cite_21",
"@cite_44",
"@cite_40",
"@cite_47",
"@cite_16"
],
"mid": [
"68733909",
"1858383477",
"2149557440",
"2149172860",
"2109586012",
"1897761818",
"2143449221",
"1987835821",
""
],
"abstract": [
"The ability to associate images with natural language sentences that describe what is depicted in them is a hallmark of image understanding, and a prerequisite for applications such as sentence-based image search. In analogy to image search, we propose to frame sentence-based image annotation as the task of ranking a given pool of captions. We introduce a new benchmark collection for sentence-based image description and search, consisting of 8,000 images that are each paired with five different captions which provide clear descriptions of the salient entities and events. We introduce a number of systems that perform quite well on this task, even though they are only based on features that can be obtained with minimal supervision. Our results clearly indicate the importance of training on multiple captions per image, and of capturing syntactic (word order-based) and semantic features of these captions. We also perform an in-depth comparison of human and automatic evaluation metrics for this task, and propose strategies for collecting human judgments cheaply and on a very large scale, allowing us to augment our collection with additional relevance judgments of which captions describe which image. Our analysis shows that metrics that consider the ranked list of results for each query image or sentence are significantly more robust than metrics that are based on a single response per query. Moreover, our study suggests that the evaluation of ranking-based image description systems may be fully automated.",
"We propose a sentence generation strategy that describes images by predicting the most likely nouns, verbs, scenes and prepositions that make up the core sentence structure. The input are initial noisy estimates of the objects and scenes detected in the image using state of the art trained detectors. As predicting actions from still images directly is unreliable, we use a language model trained from the English Gigaword corpus to obtain their estimates; together with probabilities of co-located nouns, scenes and prepositions. We use these estimates as parameters on a HMM that models the sentence generation process, with hidden nodes as sentence components and image detections as the emissions. Experimental results show that our strategy of combining vision and language produces readable and descriptive sentences compared to naive strategies that use vision alone.",
"Previous work on Recursive Neural Networks (RNNs) shows that these models can produce compositional feature vectors for accurately representing and classifying sentences or images. However, the sentence vectors of previous models cannot accurately represent visually grounded meaning. We introduce the DT-RNN model which uses dependency trees to embed sentences into a vector space in order to retrieve images that are described by those sentences. Unlike previous RNN-based models which use constituency trees, DT-RNNs naturally focus on the action and agents in a sentence. They are better able to abstract from the details of word order and syntactic expression. DT-RNNs outperform other recursive and recurrent neural networks, kernelized CCA and a bag-of-words baseline on the tasks of finding an image that fits a sentence description and vice versa. They also give more similar representations to sentences that describe the same image.",
"We present a holistic data-driven approach to image description generation, exploiting the vast amount of (noisy) parallel image data and associated natural language descriptions available on the web. More specifically, given a query image, we retrieve existing human-composed phrases used to describe visually similar images, then selectively combine those phrases to generate a novel description for the query image. We cast the generation process as constraint optimization problems, collectively incorporating multiple interconnected aspects of language composition for content planning, surface realization and discourse structure. Evaluation by human annotators indicates that our final system generates more semantically correct and linguistically appealing descriptions than two nontrivial baselines.",
"We develop and demonstrate automatic image description methods using a large captioned photo collection. One contribution is our technique for the automatic collection of this new dataset – performing a huge number of Flickr queries and then filtering the noisy results down to 1 million images with associated visually relevant captions. Such a collection allows us to approach the extremely challenging problem of description generation using relatively simple non-parametric methods and produces surprisingly effective results. We also develop methods incorporating many state of the art, but fairly noisy, estimates of image content to produce even more pleasing results. Finally we introduce a new objective performance measure for image captioning.",
"Humans can prepare concise descriptions of pictures, focusing on what they find important. We demonstrate that automatic methods can do so too. We describe a system that can compute a score linking an image to a sentence. This score can be used to attach a descriptive sentence to a given image, or to obtain images that illustrate a given sentence. The score is obtained by comparing an estimate of meaning obtained from the image to one obtained from the sentence. Each estimate of meaning comes from a discriminative procedure that is learned us-ingdata. We evaluate on a novel dataset consisting of human-annotated images. While our underlying estimate of meaning is impoverished, it is sufficient to produce very good quantitative results, evaluated with a novel score that can account for synecdoche.",
"Describing the main event of an image involves identifying the objects depicted and predicting the relationships between them. Previous approaches have represented images as unstructured bags of regions, which makes it difficult to accurately predict meaningful relationships between regions. In this paper, we introduce visual dependency representations to capture the relationships between the objects in an image, and hypothesize that this representation can improve image description. We test this hypothesis using a new data set of region-annotated images, associated with visual dependency representations and gold-standard descriptions. We describe two template-based description generation models that operate over visual dependency representations. In an image description task, we find that these models outperform approaches that rely on object proximity or corpus information to generate descriptions on both automatic measures and on human judgements.",
"In this paper, we present an image parsing to text description (I2T) framework that generates text descriptions of image and video content based on image understanding. The proposed I2T framework follows three steps: 1) input images (or video frames) are decomposed into their constituent visual patterns by an image parsing engine, in a spirit similar to parsing sentences in natural language; 2) the image parsing results are converted into semantic representation in the form of Web ontology language (OWL), which enables seamless integration with general knowledge bases; and 3) a text generation engine converts the results from previous steps into semantically meaningful, human readable, and query-able text reports. The centerpiece of the I2T framework is an and-or graph (AoG) visual knowledge representation, which provides a graphical representation serving as prior knowledge for representing diverse visual patterns and provides top-down hypotheses during the image parsing. The AoG embodies vocabularies of visual elements including primitives, parts, objects, scenes as well as a stochastic image grammar that specifies syntactic relations (i.e., compositional) and semantic relations (e.g., categorical, spatial, temporal, and functional) between these visual elements. Therefore, the AoG is a unified model of both categorical and symbolic representations of visual knowledge. The proposed I2T framework has two objectives. First, we use semiautomatic method to parse images from the Internet in order to build an AoG for visual knowledge representation. Our goal is to make the parsing process more and more automatic using the learned AoG model. Second, we use automatic methods to parse image video in specific domains and generate text reports that are useful for real-world applications. In the case studies at the end of this paper, we demonstrate two automatic I2T systems: a maritime and urban scene video surveillance system and a real-time automatic driving scene understanding system.",
""
]
} |
1511.03292 | 2226480163 | In this paper we propose the construction of linguistic descriptions of images. This is achieved through the extraction of scene description graphs (SDGs) from visual scenes using an automatically constructed knowledge base. SDGs are constructed using both vision and reasoning. Specifically, commonsense reasoning is applied on (a) detections obtained from existing perception methods on given images, (b) a "commonsense" knowledge base constructed using natural language processing of image annotations and (c) lexical ontological knowledge from resources such as WordNet. Amazon Mechanical Turk(AMT)-based evaluations on Flickr8k, Flickr30k and MS-COCO datasets show that in most cases, sentences auto-constructed from SDGs obtained by our method give a more relevant and thorough description of an image than a recent state-of-the-art image caption based approach. Our Image-Sentence Alignment Evaluation results are also comparable to that of the recent state-of-the art approaches. | Several works have shown promising efforts to acquire and apply commonsense in different aspects of Scene Analysis. @cite_49 uses abstraction to discover semantically similar images. @cite_8 proposes to learn all variations pertaining to all concepts and @cite_42 uses common-sense to learn actions. | {
"cite_N": [
"@cite_42",
"@cite_49",
"@cite_8"
],
"mid": [
"1496872269",
"1996418862",
"2081613070"
],
"abstract": [
"This work presents a novel approach for human action recognition based on the combination of computer vision techniques and common-sense knowledge and reasoning capabilities. The emphasis of this work is on how common sense has to be leveraged to a vision-based human action recognition so that nonsensical errors can be amended at the understanding stage. The proposed framework is to be deployed in a realistic environment in which humans behave rationally, that is, motivated by an aim or a reason.",
"Relating visual information to its linguistic semantic meaning remains an open and challenging area of research. The semantic meaning of images depends on the presence of objects, their attributes and their relations to other objects. But precisely characterizing this dependence requires extracting complex visual information from an image, which is in general a difficult and yet unsolved problem. In this paper, we propose studying semantic information in abstract images created from collections of clip art. Abstract images provide several advantages. They allow for the direct study of how to infer high-level semantic information, since they remove the reliance on noisy low-level object, attribute and relation detectors, or the tedious hand-labeling of images. Importantly, abstract images also allow the ability to generate sets of semantically similar scenes. Finding analogous sets of semantically similar real images would be nearly impossible. We create 1,002 sets of 10 semantically similar abstract scenes with corresponding written descriptions. We thoroughly analyze this dataset to discover semantically important features, the relations of words to visual features and methods for measuring semantic similarity.",
"Recognition is graduating from labs to real-world applications. While it is encouraging to see its potential being tapped, it brings forth a fundamental challenge to the vision researcher: scalability. How can we learn a model for any concept that exhaustively covers all its appearance variations, while requiring minimal or no human supervision for compiling the vocabulary of visual variance, gathering the training images and annotations, and learning the models? In this paper, we introduce a fully-automated approach for learning extensive models for a wide range of variations (e.g. actions, interactions, attributes and beyond) within any concept. Our approach leverages vast resources of online books to discover the vocabulary of variance, and intertwines the data collection and modeling steps to alleviate the need for explicit human supervision in training the models. Our approach organizes the visual knowledge about a concept in a convenient and useful way, enabling a variety of applications across vision and NLP. Our online system has been queried by users to learn models for several interesting concepts including breakfast, Gandhi, beautiful, etc. To date, our system has models available for over 50, 000 variations within 150 concepts, and has annotated more than 10 million images with bounding boxes."
]
} |
1511.03292 | 2226480163 | In this paper we propose the construction of linguistic descriptions of images. This is achieved through the extraction of scene description graphs (SDGs) from visual scenes using an automatically constructed knowledge base. SDGs are constructed using both vision and reasoning. Specifically, commonsense reasoning is applied on (a) detections obtained from existing perception methods on given images, (b) a "commonsense" knowledge base constructed using natural language processing of image annotations and (c) lexical ontological knowledge from resources such as WordNet. Amazon Mechanical Turk(AMT)-based evaluations on Flickr8k, Flickr30k and MS-COCO datasets show that in most cases, sentences auto-constructed from SDGs obtained by our method give a more relevant and thorough description of an image than a recent state-of-the-art image caption based approach. Our Image-Sentence Alignment Evaluation results are also comparable to that of the recent state-of-the art approaches. | Recently, @cite_20 introduced scene graphs to describe scenes and @cite_50 creates scene graphs from descriptions. However, we automatically construct the graph from an image, and we believe, due to the event-entity-attribute based representation and meaningful edge-labels (borrowed from KM-ontology @cite_4 ) , SDGs are more equipped to facilitate symbolic-level reasoning. | {
"cite_N": [
"@cite_50",
"@cite_4",
"@cite_20"
],
"mid": [
"2250378130",
"110554524",
"2077069816"
],
"abstract": [
"Semantically complex queries which include attributes of objects and relations between objects still pose a major challenge to image retrieval systems. Recent work in computer vision has shown that a graph-based semantic representation called a scene graph is an effective representation for very detailed image descriptions and for complex queries for retrieval. In this paper, we show that scene graphs can be effectively created automatically from a natural language scene description. We present a rule-based and a classifierbased scene graph parser whose output can be used for image retrieval. We show that including relations and attributes in the query graph outperforms a model that only considers objects and that using the output of our parsers is almost as effective as using human-constructed scene graphs (Recall@10 of 27.1 vs. 33.4 ). Additionally, we demonstrate the general usefulness of parsing to scene graphs by showing that the output can also be used to generate 3D scenes.",
"",
"This paper develops a novel framework for semantic image retrieval based on the notion of a scene graph. Our scene graphs represent objects (“man”, “boat”), attributes of objects (“boat is white”) and relationships between objects (“man standing on boat”). We use these scene graphs as queries to retrieve semantically related images. To this end, we design a conditional random field model that reasons about possible groundings of scene graphs to test images. The likelihoods of these groundings are used as ranking scores for retrieval. We introduce a novel dataset of 5,000 human-generated scene graphs grounded to images and use this dataset to evaluate our method for image retrieval. In particular, we evaluate retrieval using full scene graphs and small scene subgraphs, and show that our method outperforms retrieval methods that use only objects or low-level image features. In addition, we show that our full model can be used to improve object localization compared to baseline methods."
]
} |
1511.03036 | 2146958003 | Display Omitted Provides multiple semantic representations of a data source with data virtualization.Supports data extraction and conversion in an automated way on-demand.Each step in the data flow is semantically processed, following the semantic web stack.The approach is tested on large clinical data with sound performance. There is a growing need to semantically process and integrate clinical data from different sources for clinical research. This paper presents an approach to integrate EHRs from heterogeneous resources and generate integrated data in different data formats or semantics to support various clinical research applications. The proposed approach builds semantic data virtualization layers on top of data sources, which generate data in the requested semantics or formats on demand. This approach avoids upfront dumping to and synchronizing of the data with various representations. Data from different EHR systems are first mapped to RDF data with source semantics, and then converted to representations with harmonized domain semantics where domain ontologies and terminologies are used to improve reusability. It is also possible to further convert data to application semantics and store the converted results in clinical research databases, e.g. i2b2, OMOP, to support different clinical research settings. Semantic conversions between different representations are explicitly expressed using N3 rules and executed by an N3 Reasoner (EYE), which can also generate proofs of the conversion processes. The solution presented in this paper has been applied to real-world applications that process large scale EHR data. | To improve the interoperability of EHRs represented with different standards, mappings between different standards are developed and domain ontologies are created. @cite_17 developed source ontologies for openEHR and ISO 13606, as well as a domain ontology which bridges the two standards. Data transformation is carried out through syntactic mapping between the archetypes of openEHR, ISO 13606 and the archetype model of the domain ontology. @cite_38 build up a domain ontology of ECG and map schemas of three ECG standards directly to the domain ontology. They find it difficult to develop a domain ontology to which different standards can directly be mapped. | {
"cite_N": [
"@cite_38",
"@cite_17"
],
"mid": [
"2121556460",
"2134273606"
],
"abstract": [
"In this paper we test the hypothesis that a domain reference ontology of the electrocardiogram (ECG) can be employed in an effective manner to achieve semantic integration between ECG data standards. Several standardization initiatives, namely AHA MIT-BIH (Physionet), SCP-ECG and HL7 aECG, have led to heterogeneous conceptualizations of the ECG domain. We then argue that a shared anchor, the biomedical reality under scrutiny, can effectively support the semantic integration of these ECG standards into a coherent ECG representation for the sake of a unified Electronic Health Record (EHR) model. Our hypothesis is tested by means of an integration experiment that uses, on the one hand, an ECG Ontology and, on the other hand, elicited conceptual models of the ECG standards. As a conclusion, we attest the hypothesis and also provide an integration table depicting correspondence links between entities in the ECG Ontology and elements in the ECG standards.",
"The semantic interoperability between health information systems is a major challenge to improve the quality of clinical practice and patient safety. In recent years many projects have faced this problem and provided solutions based on specific standards and technologies in order to satisfy the needs of a particular scenario. Most of such solutions cannot be easily adapted to new scenarios, thus more global solutions are needed. In this work, we have focused on the semantic interoperability of electronic healthcare records standards based on the dual model architecture and we have developed a solution that has been applied to ISO 13606 and openEHR. The technological infrastructure combines reference models, archetypes and ontologies, with the support of Model-driven Engineering techniques. For this purpose, the interoperability infrastructure developed in previous work by our group has been reused and extended to cover the requirements of data transformation."
]
} |
1511.03183 | 2100523817 | A novel approach for the fusion of heterogeneous object detection methods is proposed. In order to effectively integrate the outputs of multiple detectors, the level of ambiguity in each individual detection score is estimated using the precision recall relationship of the corresponding detector. The main contribution of the proposed work is a novel fusion method, called Dynamic Belief Fusion (DBF), which dynamically assigns probabilities to hypotheses (target, non-target, intermediate state (target or non-target)) based on confidence levels in the detection results conditioned on the prior performance of individual detectors. In DBF, a joint basic probability assignment, optimally fusing information from all detectors, is determined by the Dempster's combination rule, and is easily reduced to a single fused detection score. Experiments on ARL and PASCAL VOC 07 datasets demonstrate that the detection accuracy of DBF is considerably greater than conventional fusion approaches as well as individual detectors used for the fusion. | Kwon and Lee proposed two approaches integrating multiple sample-based tracking approaches using an interactive Markov Chain Monte Carlo (iMCMC) framework @cite_20 and using sampling in tracker space modeled by Markov Chain Monte Carlo (MCMC) method @cite_22 , respectively. @cite_9 introduced an approach combining detectors of different modalities (concept, text, speech) by using relationships among the modes in the event detection. However, in general, modeling the dependencies in fusion among multiple approaches built on different principles is infeasible. | {
"cite_N": [
"@cite_9",
"@cite_22",
"@cite_20"
],
"mid": [
"",
"2105403175",
"2098854771"
],
"abstract": [
"",
"We propose a novel tracking framework called visual tracker sampler that tracks a target robustly by searching for the appropriate trackers in each frame. Since the real-world tracking environment varies severely over time, the trackers should be adapted or newly constructed depending on the current situation. To do this, our method obtains several samples of not only the states of the target but also the trackers themselves during the sampling process. The trackers are efficiently sampled using the Markov Chain Monte Carlo method from the predefined tracker space by proposing new appearance models, motion models, state representation types, and observation types, which are the basic important components of visual trackers. Then, the sampled trackers run in parallel and interact with each other while covering various target variations efficiently. The experiment demonstrates that our method tracks targets accurately and robustly in the real-world tracking environments and outperforms the state-of-the-art tracking methods.",
"We propose a novel tracking algorithm that can work robustly in a challenging scenario such that several kinds of appearance and motion changes of an object occur at the same time. Our algorithm is based on a visual tracking decomposition scheme for the efficient design of observation and motion models as well as trackers. In our scheme, the observation model is decomposed into multiple basic observation models that are constructed by sparse principal component analysis (SPCA) of a set of feature templates. Each basic observation model covers a specific appearance of the object. The motion model is also represented by the combination of multiple basic motion models, each of which covers a different type of motion. Then the multiple basic trackers are designed by associating the basic observation models and the basic motion models, so that each specific tracker takes charge of a certain change in the object. All basic trackers are then integrated into one compound tracker through an interactive Markov Chain Monte Carlo (IMCMC) framework in which the basic trackers communicate with one another interactively while run in parallel. By exchanging information with others, each tracker further improves its performance, which results in increasing the whole performance of tracking. Experimental results show that our method tracks the object accurately and reliably in realistic videos where the appearance and motion are drastically changing over time."
]
} |
1511.03183 | 2100523817 | A novel approach for the fusion of heterogeneous object detection methods is proposed. In order to effectively integrate the outputs of multiple detectors, the level of ambiguity in each individual detection score is estimated using the precision recall relationship of the corresponding detector. The main contribution of the proposed work is a novel fusion method, called Dynamic Belief Fusion (DBF), which dynamically assigns probabilities to hypotheses (target, non-target, intermediate state (target or non-target)) based on confidence levels in the detection results conditioned on the prior performance of individual detectors. In DBF, a joint basic probability assignment, optimally fusing information from all detectors, is determined by the Dempster's combination rule, and is easily reduced to a single fused detection score. Experiments on ARL and PASCAL VOC 07 datasets demonstrate that the detection accuracy of DBF is considerably greater than conventional fusion approaches as well as individual detectors used for the fusion. | In the case where modeling dependency among multiple approaches is not possible, fusion can be performed over their outputs (late fusion). @cite_18 introduced a fusion framework in the target tracking paradigm. They collected trajectories from multiple tracking algorithms and computed one fused trajectory in order to improve accuracy, trajectory continuity, and smoothness. However, since temporal information obtained from trajectory cannot be applicable in the object detection task, a different fusion framework is necessary for our problem. @cite_23 and @cite_8 used weighted-sum (WS) methods to fuse multiple types of data for object detection. Their WS method learns weights in a manner that estimates trust in multiple data sources. However, since weights optimization is usually performed to maximize distance between positive and negative samples, like Bayesian fusion, WS does not provide a way to indicate ambiguity between the positives and negatives, which degrades fusion performance, as previously mentioned. The works of Ma and Yuen @cite_5 and @cite_11 employing Bayesian fusion also shows limited performance. | {
"cite_N": [
"@cite_18",
"@cite_8",
"@cite_23",
"@cite_5",
"@cite_11"
],
"mid": [
"2102304169",
"",
"2153986527",
"178037058",
"2090518410"
],
"abstract": [
"General object tracking is a challenging problem, where each tracking algorithm performs well on different sequences. This is because each of them has different strengths and weaknesses. We show that this fact can be utilized to create a fusion approach that clearly outperforms the best tracking algorithms in tracking performance. Thanks to dynamic programming based trajectory optimization we cannot only outperform tracking algorithms in accuracy but also in other important aspects like trajectory continuity and smoothness. Our fusion approach is very generic as it only requires frame-based tracking results in form of the object’s bounding box as input and thus can work with arbitrary tracking algorithms. It is also suited for live tracking. We evaluated our approach using 29 different algorithms on 51 sequences and show the superiority of our approach compared to state-of-the-art tracking methods.",
"",
"Many state-of-the-art optical flow estimation algorithms optimize the data and regularization terms to solve ill-posed problems. In this paper, in contrast to the conventional optical flow framework that uses a single or fixed data model, we study a novel framework that employs locally varying data term that adaptively combines different multiple types of data models. The locally adaptive data term greatly reduces the matching ambiguity due to the complementary nature of the multiple data models. The optimal number of complementary data models is learnt by minimizing the redundancy among them under the minimum description length constraint (MDL). From these chosen data models, a new optical flow estimation energy model is designed with the weighted sum of the multiple data models, and a convex optimization-based highly effective and practical solution that finds the optical flow, as well as the weights is proposed. Comparative experimental results on the Middlebury optical flow benchmark show that the proposed method using the complementary data models outperforms the state-of-the art methods.",
"This paper addresses the independent assumption issue in classifier fusion process. In the last decade, dependency modeling techniques were developed under some specific assumptions which may not be valid in practical applications. In this paper, using analytical functions on posterior probabilities of each feature, we propose a new framework to model dependency without those assumptions. With the analytical dependency model (ADM), we give an equivalent condition to the independent assumption from the properties of marginal distributions, and show that the proposed ADM can model dependency. Since ADM may contain infinite number of undetermined coefficients, we further propose a reduced form of ADM, based on the convergent properties of analytical functions. Finally, under the regularized least square criterion, an optimal Reduced Analytical Dependency Model (RADM) is learned by approximating posterior probabilities such that all training samples are correctly classified. Experimental results show that the proposed RADM outperforms existing classifier fusion methods on Digit, Flower, Face and Human Action databases.",
"While image alignment has been studied in different areas of computer vision for decades, aligning images depicting different scenes remains a challenging problem. Analogous to optical flow, where an image is aligned to its temporally adjacent frame, we propose SIFT flow, a method to align an image to its nearest neighbors in a large image corpus containing a variety of scenes. The SIFT flow algorithm consists of matching densely sampled, pixelwise SIFT features between two images while preserving spatial discontinuities. The SIFT features allow robust matching across different scene object appearances, whereas the discontinuity-preserving spatial model allows matching of objects located at different parts of the scene. Experiments show that the proposed approach robustly aligns complex scene pairs containing significant spatial differences. Based on SIFT flow, we propose an alignment-based large database framework for image analysis and synthesis, where image information is transferred from the nearest neighbors to a query image according to the dense scene correspondence. This framework is demonstrated through concrete applications such as motion field prediction from a single image, motion synthesis via object transfer, satellite image registration, and face recognition."
]
} |
1511.03015 | 2260041796 | In this paper, we present a novel approach to automatic 3D Facial Expression Recognition (FER) based on deep representation of facial 3D geometric and 2D photometric attributes. A 3D face is firstly represented by its geometric and photometric attributes, including the geometry map, normal maps, normalized curvature map and texture map. These maps are then fed into a pre-trained deep convolutional neural network to generate the deep representation. Then the facial expression prediction is simplyachieved by training linear SVMs over the deep representation for different maps and fusing these SVM scores. The visualizations show that the deep representation provides a complete and highly discriminative coding scheme for 3D faces. Comprehensive experiments on the BU-3DFE database demonstrate that the proposed deep representation can outperform the widely used hand-crafted descriptors (i.e., LBP, SIFT, HOG, Gabor) and the state-of-art approaches under the same experimental protocols. | approaches generally extract local expression features around facial landmarks based on surface geometric attributes or differential quantities. For example, 3D landmark distances @cite_26 , @cite_38 , @cite_29 , @cite_11 , 3D curves @cite_33 , geometry and normal maps @cite_28 , conformal images @cite_36 , surface normal @cite_30 and curvatures @cite_30 , @cite_18 are some popular 3D shape features. approaches generally perform better than ones. However, their performances are largely depend on the accuracy of 3D facial landmarking, which is a challenging task @cite_27 . | {
"cite_N": [
"@cite_30",
"@cite_38",
"@cite_18",
"@cite_26",
"@cite_33",
"@cite_28",
"@cite_36",
"@cite_29",
"@cite_27",
"@cite_11"
],
"mid": [
"101526648",
"2151430047",
"2136063698",
"1557043124",
"2095399814",
"1997145183",
"2006943008",
"2534547303",
"2069190036",
"2166034132"
],
"abstract": [
"3D face models accurately capture facial surfaces, making it possible for precise description of facial activities. In this paper, we present a novel mesh-based method for 3D facial expression recognition using two local shape descriptors. To characterize shape information of the local neighborhood of facial landmarks, we calculate the weighted statistical distributions of surface differential quantities, including histogram of mesh gradient (HoG) and histogram of shape index (HoS). Normal cycle theory based curvature estimation method is employed on 3D face models along with the common cubic fitting curvature estimation method for the purpose of comparison. Based on the basic fact that different expressions involve different local shape deformations, the SVM classifier with both linear and RBF kernels outperforms the state of the art results on the subset of the BU-3DFE database with the same experimental setting.",
"This paper describes a pose invariant three-dimensional (3D) facial expression recognition method using distance vectors retrieved from 3D distributions of facial feature points to classify universal facial expressions. Probabilistic Neural Network architecture is employed as a classifier to recognize the facial expressions from a distance vector obtained from 3D facial feature locations. Facial expressions such as anger, sadness, surprise, joy, disgust, fear and neutral are successfully recognized with an average recognition rate of 87.8 .",
"The creation of facial range models by 3D imaging systems has led to extensive work on 3D face recognition [19]. However, little work has been done to study the usefulness of such data for recognizing and understanding facial expressions. Psychological research shows that the shape of a human face, a highly mobile facial surface, is critical to facial expression perception. In this paper, we investigate the importance and usefulness of 3D facial geometric shapes to represent and recognize facial expressions using 3D facial expression range data. We propose a novel approach to extract primitive 3D facial expression features, and then apply the feature distribution to classify the prototypic facial expressions. In order to validate our proposed approach, we have conducted experiments for person-independent facial expression recognition using our newly created 3D facial expression database. We also demonstrate the advantages of our 3D geometric based approach over 2D texture based approaches in terms of various head poses.",
"In this paper, we propose a novel approach for facial expression analysis and recognition. The proposed approach relies on the distance vectors retrieved from 3D distribution of facial feature points to classify universal facial expressions. Neural network architecture is employed as a classifier to recognize the facial expressions from a distance vector obtained from 3D facial feature locations. Facial expressions such as anger, sadness, surprise, joy, disgust, fear and neutral are successfully recognized with an average recognition rate of 91.3 . The highest recognition rate reaches to 98.3 in the recognition of surprise.",
"We investigate the problem of facial expression recognition using 3D face data. Our approach is based on local shape analysis of several relevant regions of a given face scan. These regions or patches from facial surfaces are extracted and represented by sets of closed curves. A Riemannian framework is used to derive the shape analysis of the extracted patches. The applied framework permits to calculate a similarity (or dissimilarity) distances between patches, and to compute the optimal deformation between them. Once calculated, these measures are employed as inputs to a commonly used classification techniques such as AdaBoost and Support Vector Machines (SVM). A quantitative evaluation of our novel approach is conducted on a subset of the publicly available BU-3DFE database.",
"We present a semi-automatic 3D Facial Expression Recognition system based on geometric facial information. In this approach, the 3D facial meshes are first fitted to an Annotated Face Model (AFM). Then, the Expressive Maps are computed, which indicate the parts of the face that are most expressive according to a particular geometric feature (e.g., vertex coordinates, normals, and local curvature). The Expressive Maps provide a way to analyze the geometric features in terms of their discriminative information and their distribution along the face and allow the reduction of the dimensionality of the input space to 2:5 of the original size. Using the selected features a simple linear classifier was trained and yielded a very competitive average recognition rate of 90:4 when evaluated using ten-fold cross validation on the publicly available BU-3DFE database.",
"We propose a general and fully automatic framework for 3D facial expression recognition by modeling sparse representation of conformal images. According to Riemann Geometry theory, a 3D facial surface S embedded in ℝ3, which is a topological disk, can be conformally mapped to a 2D unit disk D through the discrete surface Ricci Flow algorithm. Such a conformal mapping induces a unique and intrinsic surface conformal representation denoted by a pair of functions defined on D, called conformal factor image (CFI) and mean curvature image (MCI). As facial expression features, CFI captures the local area distortion of S induced by the conformal mapping; MCI characterizes the geometry information of S. To model sparse representation of conformal images for expression classification, both CFI and MCI are further normalized by a Mobius transformation. This transformation is defined by the three main facial landmarks (i.e. nose tip, left and right inner eye corners) which can be detected automatically and precisely. Expression recognition is carried out by the minimal sparse expression-class-dependent reconstruction error over the conformal image based expression dictionary. Extensive experimental results on the BU-3DFER dataset demonstrate the effectiveness and generalization of the proposed framework.",
"The 3D facial geometry contains ample information about human facial expressions. Such information is invariant to pose and lighting conditions, which have imposed serious hurdles on many 2D facial analysis problems. In this paper, we perform person and gender independent facial expression recognition based on properties of the line segments connecting certain 3D facial feature points. The normalized distances and slopes of these line segments comprise a set of 96 distinguishing features for recognizing six universal facial expressions, namely anger, disgust, fear, happiness, sadness, and surprise. Using a multi-class support vector machine (SVM) classifier, an 87.1 average recognition rate is achieved on the publicly available 3D facial expression database BU-3DFE. The highest average recognition rate obtained in our experiments is 99.2 for the recognition of surprise. Our result outperforms the result reported in the prior work, which uses elaborately extracted primitive facial surface features and an LDA classifier and which yields an average recognition rate of 83.6 on the same database.",
"This survey focuses on discrete expression classification and facial action unit recognition performed using 3D face data, possibly including a corresponding 2D texture image. Research trends to date are summarized and the limitations of current methods are discussed. The challenges towards the development of more accurate and automated 3D facial expression recognition methods are identified. We also call for standardized experimental protocols in order to draw fair and meaningful comparisons between different systems.",
"In this paper, the problem of person-independent facial expression recognition from 3D facial shapes is investigated. We propose a novel automatic feature selection method based on maximizing the average relative entropy of marginalized class-conditional feature distributions and apply it to a complete pool of candidate features composed of normalized Euclidean distances between 83 facial feature points in the 3D space. Using a regularized multi-class AdaBoost classification algorithm, we achieve a 95.1 average recognition rate for six universal facial expressions on the publicly available 3D facial expression database BU-3DFE [1], with a highest average recognition rate of 99.2 for the recognition of surprise. We compare these results with the results based on a set of manually devised features and demonstrate that the auto features yield better results than the manual features. Our results outperform the results presented in the previous work [2] and [3], namely average recognition rates of 83.6 and 91.3 on the same database, respectively."
]
} |
1511.02580 | 2268466221 | We propose ways to improve the performance of fully connected networks. We found that two approaches in particular have a strong effect on performance: linear bottleneck layers and unsupervised pre-training using autoencoders without hidden unit biases. We show how both approaches can be related to improving gradient flow and reducing sparsity in the network. We show that a fully connected network can yield approximately 70 classification accuracy on the permutation-invariant CIFAR-10 task, which is much higher than the current state-of-the-art. By adding deformations to the training data, the fully connected network achieves 78 accuracy, which is just 10 short of a decent convolutional network. | The idea of a linear bottleneck layer is very old and has been used as early as the @math 's in the context of autoencoders (eg., @cite_3 ). More recently, @cite_5 used a linear bottleneck layer to factorize a single-layer network and showed that it helped speed up learning. An application of a linear bottleneck layer in the last layer of a neural network for dealing with high-dimensional outputs is described in @cite_6 and @cite_2 . In contrast to our work, in none of these methods is the goal to alleviate vanishing gradients and deal with sparsity, and (accordingly) they use just a single bottleneck layer in the network. | {
"cite_N": [
"@cite_5",
"@cite_6",
"@cite_3",
"@cite_2"
],
"mid": [
"2952881492",
"2058641082",
"2078626246",
"2294543795"
],
"abstract": [
"Currently, deep neural networks are the state of the art on problems such as speech recognition and computer vision. In this extended abstract, we show that shallow feed-forward networks can learn the complex functions previously learned by deep nets and achieve accuracies previously only achievable with deep models. Moreover, in some cases the shallow neural nets can learn these deep functions using a total number of parameters similar to the original deep model. We evaluate our method on the TIMIT phoneme recognition task and are able to train shallow fully-connected nets that perform similarly to complex, well-engineered, deep convolutional architectures. Our success in training shallow neural nets to mimic deeper models suggests that there probably exist better algorithms for training shallow feed-forward nets than those currently available.",
"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.",
"Abstract We consider the problem of learning from examples in layered linear feed-forward neural networks using optimization methods, such as back propagation, with respect to the usual quadratic error function E of the connection weights. Our main result is a complete description of the landscape attached to E in terms of principal component analysis. We show that E has a unique minimum corresponding to the projection onto the subspace generated by the first principal vectors of a covariance matrix associated with the training patterns. All the additional critical points of E are saddle points (corresponding to projections onto subspaces generated by higher order vectors). The auto-associative case is examined in detail. Extensions and implications for the learning algorithms are discussed.",
"Recently proposed deep neural network (DNN) obtains significant accuracy improvements in many large vocabulary continuous speech recognition (LVCSR) tasks. However, DNN requires much more parameters than traditional systems, which brings huge cost during online evaluation, and also limits the application of DNN in a lot of scenarios. In this paper we present our new effort on DNN aiming at reducing the model size while keeping the accuracy improvements. We apply singular value decomposition (SVD) on the weight matrices in DNN, and then restructure the model based on the inherent sparseness of the original matrices. After restructuring we can reduce the DNN model size significantly with negligible accuracy loss. We also fine-tune the restructured model using the regular back-propagation method to get the accuracy back when reducing the DNN model size heavily. The proposed method has been evaluated on two LVCSR tasks, with context-dependent DNN hidden Markov model (CD-DNN-HMM). Experimental results show that the proposed approach dramatically reduces the DNN model size by more than 80 without losing any accuracy. Index Terms: deep neural network, singular value decomposition, model restructuring"
]
} |
1511.02667 | 2279745846 | Hypergraph matching has recently become a popular approach for solving correspondence problems in computer vision as it allows the use of higher-order geometric information. Hypergraph matching can be formulated as a third-order optimization problem subject to assignment constraints which turns out to be NP-hard. In recent work, we have proposed an algorithm for hypergraph matching which first lifts the third-order problem to a fourth-order problem and then solves the fourth-order problem via optimization of the corresponding multilinear form. This leads to a tensor block coordinate ascent scheme which has the guarantee of providing monotonic ascent in the original matching score function and leads to state-of-the-art performance both in terms of achieved matching score and accuracy. In this paper we show that the lifting step to a fourth-order problem can be avoided yielding a third-order scheme with the same guarantees and performance but being two times faster. Moreover, we introduce a homotopy type method which further improves the performance. | The graph resp. hypergraph matching problem is known to be NP-hard except for special cases where polynomial-time algorithms exist for planar graphs @cite_33 . In order to make the problem computationally tractable, a myriad of approximate algorithms have been proposed over the last three decades aiming at an acceptable trade-off between the complexity of the algorithm and matching accuracy. They can be categorized from different perspectives and we refer to @cite_26 @cite_10 for an extensive review. In this section, we review only those approximate algorithms that use Lawler's formulations for both graph and hypergraph matching as they are closely related to our work. In particular, the graph matching problem can be formulated as a quadratic assignment problem (QAP) @math while the hypergraph matching problem is formulated as a higher-order assignment problem (HAP) @math In these formulations, @math and @math refer to the affinity matrix and affinity tensor respectively, and @math is some matching constraint set depending on specific applications. Also, depending on whether the problem is graph matching or hypergraph matching, the corresponding algorithms will be called second-order methods or higher-order methods. | {
"cite_N": [
"@cite_26",
"@cite_10",
"@cite_33"
],
"mid": [
"2109294083",
"",
"2046021036"
],
"abstract": [
"A recent paper posed the question: \"Graph Matching: What are we really talking about?\". Far from providing a definite answer to that question, in this paper we will try to characterize the role that graphs play within the Pattern Recognition field. To this aim two taxonomies are presented and discussed. The first includes almost all the graph matching algorithms proposed from the late seventies, and describes the different classes of algorithms. The second taxonomy considers the types of common applications of graph-based techniques in the Pattern Recognition and Machine Vision field.",
"",
"The isomorphism problem for graphs G 1 and G 2 is to determine if there exists a one-to-one mapping of the vertices of G 1 onto the vertices of G 2 such that two vertices of G 1 are adjacent if and only if their images in G 2 are adjacent. In addition to determining the existence of such an isomorphism, it is useful to be able to produce an isomorphism-inducing mapping in the case where one exists. The isomorphism problem for triconnected planar graphs is particularly simple since a triconnected planar graph has a unique embedding on a sphere [6]. Weinberg [5] exploited this fact in developing an algorithm for testing isomorphism of triconnected planar graphs in O(|V| 2 ) time where V is the set consisting of the vertices of both graphs. The result has been extended to arbitrary planar graphs and improved to O(|V|log|V|) steps by Hopcroft and Tarjan [2,3]. In this paper, the time bound for planar graph isomorphism is improved to O(|V|). In addition to determining the isomorphism of two planar graphs, the algorithm can be easily extended to partition a set of planar graphs into equivalence classes of isomorphic graphs in time linear in the total number of vertices in all graphs in the set. A random access model of computation (see Cook [1]) is assumed. Although the proposed algorithm has a linear asymptotic growth rate, at the present stage of development it appears to be inefficient on account of a rather large constant. This paper is intended only to establish the existence of a linear algorithm which subsequent work might make truly efficient."
]
} |
1511.02669 | 2244358074 | We propose a system for automated essay grading using ontologies and textual entailment. The process of textual entailment is guided by hypotheses, which are extracted from a domain ontology. Textual entailment checks if the truth of the hypothesis follows from a given text. We enact textual entailment to compare students answer to a model answer obtained from ontology. We validated the solution against various essays written by students in the chemistry domain. | The current methods of essay scoring can be categorized into two classes: holistic scoring and rubric-based scoring. In holistic scoring, the essay is assessed and a single score selected from a predefined score range is assigned as an overall score @cite_2 . In the analytical or rubric-based scoring method, essays are assessed on the basis of a certain set of well-defined features @cite_2 . Each feature has a scale associated with it and the final score awarded to the essay is the sum of scores of all the essay rubrics features. | {
"cite_N": [
"@cite_2"
],
"mid": [
"1990278592"
],
"abstract": [
"The main purpose of the study was to investigate the distinctness and reliability of analytic (or multi-trait) rating dimensions and their relationships to holistic scores and e-rater® essay feature variables in the context of the TOEFL® computer-based test (TOEFL CBT) writing assessment. Data analyzed in the study were holistic and multi-trait essay scores provided by human raters and essay feature variable scores computed by e-rater® (version 2.0) for two TOEFL CBT writing prompts. It was found that (i) all of the six multi-trait scores were not only correlated among themselves but also correlated with the holistic score, (ii) high correlations obtained among holistic and multi-trait scores were largely attributable to the impact of essay length on both holistic and multi-trait scoring, and (iii) some strong associations were confirmed between several e-rater variables and multi-trait rating dimensions. Implications are discussed for improving the multi-trait scoring of essays, refining e-rater essay feature variables, and validating automated essay scores."
]
} |
1511.02669 | 2244358074 | We propose a system for automated essay grading using ontologies and textual entailment. The process of textual entailment is guided by hypotheses, which are extracted from a domain ontology. Textual entailment checks if the truth of the hypothesis follows from a given text. We enact textual entailment to compare students answer to a model answer obtained from ontology. We validated the solution against various essays written by students in the chemistry domain. | The National Assessment Program Literacy and Numeracy (NAPLAN) @cite_1 rubric of persuasive essay grading lists a set of criteria for marking persuasive writing. The Spelling Mark algorithm is developed to formalise the NAPLAN rubric for marking spelling based on common heuristics and rules of the English language. The first step is to obtain the total number of words in the essay and the number of spelling errors in the essay. Then, each word is categorized based on the difficulty level into one of four classes: simple, common, difficult or challenging, while the number of correct and incorrect words in each category is counted. The final step in the algorithm is to assign the spelling mark according to the set of rules @cite_1 . | {
"cite_N": [
"@cite_1"
],
"mid": [
"1972609793"
],
"abstract": [
"Abstract Automated Essay Grading (AEG) is defined as a computer technology that evaluates and scores written prose. A number of AEG systems have been developed since the 1960s and in most of them, an ad-hoc or generalized approach is used to grade spelling even though it is an important element of an essay-scoring rubric. Existing approaches do not therefore give an accurate representation or measure of spelling in essays. According to the rubric-based scoring method used in the National Assessment Program “Literacy and Numeracy (NAPLAN) in Australia, spelling is marked in three steps” first, by identifying the correct and incorrect words in the essay; second, by categorizing each word based on the difficulty level into one of four classes: Simple, Common, Difficult or Challenging, and counting the number of correct and incorrect words in each category; finally, by using the pre-defined NAPLAN rubric scale to assign the mark. Only a small number of existing AEG systems can be used for rubric-based scoring, and none can be used to grade spelling according to the NAPLAN rubric. In this paper, we address this shortcoming in the existing literature and present an innovative approach to automatically mark spelling using rubric based scoring. We develop two algorithms based on the rules and heuristics of the English language to formulize the rubric for spelling and then implement these algorithms in Java language and perform a series of evaluations of our system using an essay dataset. Our results are very promising, even though it is the first system of this kind."
]
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.