aid stringlengths 9 15 | mid stringlengths 7 10 | abstract stringlengths 78 2.56k | related_work stringlengths 92 1.77k | ref_abstract dict |
|---|---|---|---|---|
1704.05004 | 2955107877 | Memory corruption vulnerabilities in C C++ applications enable attackers to execute code, change data, and leak information. Current memory sanitizers do no provide comprehensive coverage of a program's data. In particular, existing tools focus primarily on heap allocations with limited support for stack allocations and globals. Additionally, existing tools focus on the main executable with limited support for system libraries. Further, they suffer from both false positives and false negatives. We present Comprehensive User-Space Protection for C C++, CUP, an LLVM sanitizer that provides complete spatial and probabilistic temporal memory safety for C C++ program on 64-bit architectures (with a prototype implementation for x86_64). CUP uses a hybrid metadata scheme that supports all program data including globals, heap, or stack and maintains the ABI. Compared to existing approaches with the NIST Juliet test suite, CUP reduces false negatives by 10x (0.1 ) compared to the state of the art LLVM sanitizers, and produces no false positives. CUP instruments all user-space code, including libc and other system libraries, removing them from the trusted code base. | is required to prevent temporal memory safety violations (lifetime errors). This property requires remembering for every pointer whether the object to which it is assigned is still allocated. AddressSanitizer @cite_35 and Low-Fat Pointers @cite_3 make no attempt to do this (and their metadata does not support this property), while SoftBound+CETS @cite_13 enforces this property. AddressSanitizer and Low-Fat Pointers do not maintain metadata either per object or per pointer. requires either per objecct or per pointer metadata. Consequently, they fundamentally enforce temporal safety because their (current) metadata be object aware. | {
"cite_N": [
"@cite_35",
"@cite_13",
"@cite_3"
],
"mid": [
"1878544538",
"2122757982",
"2301548261"
],
"abstract": [
"Memory access bugs, including buffer overflows and uses of freed heap memory, remain a serious problem for programming languages like C and C++. Many memory error detectors exist, but most of them are either slow or detect a limited set of bugs, or both. This paper presents AddressSanitizer, a new memory error detector. Our tool finds out-of-bounds accesses to heap, stack, and global objects, as well as use-after-free bugs. It employs a specialized memory allocator and code instrumentation that is simple enough to be implemented in any compiler, binary translation system, or even in hardware. AddressSanitizer achieves efficiency without sacrificing comprehensiveness. Its average slowdown is just 73 yet it accurately detects bugs at the point of occurrence. It has found over 300 previously unknown bugs in the Chromium browser and many bugs in other software.",
"Temporal memory safety errors, such as dangling pointer dereferences and double frees, are a prevalent source of software bugs in unmanaged languages such as C. Existing schemes that attempt to retrofit temporal safety for such languages have high runtime overheads and or are incomplete, thereby limiting their effectiveness as debugging aids. This paper presents CETS, a compile-time transformation for detecting all violations of temporal safety in C programs. Inspired by existing approaches, CETS maintains a unique identifier with each object, associates this metadata with the pointers in a disjoint metadata space to retain memory layout compatibility, and checks that the object is still allocated on pointer dereferences. A formal proof shows that this is sufficient to provide temporal safety even in the presence of arbitrary casts if the program contains no spatial safety violations. Our CETS prototype employs both temporal check removal optimizations and traditional compiler optimizations to achieve a runtime overhead of just 48 on average. When combined with a spatial-checking system, the average overall overhead is 116 for complete memory safety",
"Heap buffer overflow (underflow) errors are a common source of security vulnerabilities. One prevention mechanism is to add object bounds meta information and to instrument the program with explicit bounds checks for all memory access. The so-called \"fat pointers\" approach is one method for maintaining and propagating the meta information where native machine pointers are replaced with \"fat\" objects that explicitly store object bounds. Another approach is \"low fat pointers\", which encodes meta information within a native pointer itself, eliminating space overheads and also code compatibility issues. This paper presents a new low-fat pointer encoding that is fully compatible with existing libraries (e.g. pre-compiled libraries unaware of the encoding) and standard hardware (e.g. x86_64). We show that our approach has very low memory overhead, and competitive with existing state-of-the-art bounds instrumentation solutions."
]
} |
1704.05004 | 2955107877 | Memory corruption vulnerabilities in C C++ applications enable attackers to execute code, change data, and leak information. Current memory sanitizers do no provide comprehensive coverage of a program's data. In particular, existing tools focus primarily on heap allocations with limited support for stack allocations and globals. Additionally, existing tools focus on the main executable with limited support for system libraries. Further, they suffer from both false positives and false negatives. We present Comprehensive User-Space Protection for C C++, CUP, an LLVM sanitizer that provides complete spatial and probabilistic temporal memory safety for C C++ program on 64-bit architectures (with a prototype implementation for x86_64). CUP uses a hybrid metadata scheme that supports all program data including globals, heap, or stack and maintains the ABI. Compared to existing approaches with the NIST Juliet test suite, CUP reduces false negatives by 10x (0.1 ) compared to the state of the art LLVM sanitizers, and produces no false positives. CUP instruments all user-space code, including libc and other system libraries, removing them from the trusted code base. | The state-of-the-art C C++ pointer-based memory safety scheme is SoftBound+CETS @cite_32 . Other pointer-based schemes include CCured @cite_40 and Cyclone @cite_11 . CCured uses a fat pointer to store metadata, as well as programmer annotations for indicating safe casts. Unfortunately, fat pointers break the ABI, and programmer annotations can significantly increase developer time. Even with annotations, CCured fails to handle structure changes. Cyclone also uses a fat pointer scheme, but does not guarantee full memory safety. We refer to @cite_32 for a survey of other related work on memory safety and a discussion on different pointer metadata schemes. | {
"cite_N": [
"@cite_40",
"@cite_32",
"@cite_11"
],
"mid": [
"",
"2126790595",
"1825457006"
],
"abstract": [
"",
"Lack of memory safety in C C++ has resulted in numerous security vulnerabilities and serious bugs in large software systems. This paper highlights the challenges in enforcing memory safety for C C++ programs and progress made as part of the SoftBoundCETS project. We have been exploring memory safety enforcement at various levels ‐ in hardware, in the compiler, and as a hardware-compiler hybrid ‐ in this project. Our research has identified that maintaining metadata with pointers in a disjoint metadata space and performing bounds and use-after-free checking can provide comprehensive memory safety. We describe the rationale behind the design decisions and its ramifications on various dimensions, our experience with the various variants that we explored in this project, and the lessons learned in the process. We also describe and analyze the forthcoming Intel Memory Protection Extensions (MPX) that provides hardware acceleration for disjoint metadata and pointer checking in mainstream hardware, which is expected to be available later this year. 1998 ACM Subject Classification D.3.4 Processors, D.2.5 Testing and Debugging",
"Cyclone is a safe dialect of C. It has been designed from the ground up to prevent the buffer overflows, format string attacks, and memory management errors that are common in C programs, while retaining C’s syntax and semantics. This paper examines safety violations enabled by C’s design, and shows how Cyclone avoids them, without giving up C’s hallmark control over low-level details such as data representation and memory management."
]
} |
1704.04722 | 2616123428 | There have been numerous studies on the problem of flocking control for multiagent systems whose simplified models are presented in terms of point-mass elements. Meanwhile, full dynamic models pose some challenging problems in addressing the flocking control problem of mobile robots due to their nonholonomic dynamic properties. Taking practical constraints into consideration, we propose a novel approach to distributed flocking control of nonholonomic mobile robots by bounded feedback. The flocking control objectives consist of velocity consensus, collision avoidance, and cohesion maintenance among mobile robots. A flocking control protocol that is based on the information of neighbor mobile robots is constructed. The theoretical analysis is conducted with the help of a Lyapunov-like function and graph theory. Simulation results are shown to demonstrate the efficacy of the proposed distributed flocking control scheme. | In @cite_5 , a distributed flocking control approach was proposed but no constraints on the control inputs are imposed. The work in @cite_45 @cite_51 considers the bounded feedback flocking control problem for nonholonomic mobile robots without a flocking desired heading angle. The problem of interest in this chapter is to address bounded control of nonholonomic dynamic mobile vehicles, which also achieves CMCA and obstacle avoidance. We also consider a flocking desired heading angle, which reveals a collective flocking behaviour. | {
"cite_N": [
"@cite_5",
"@cite_45",
"@cite_51"
],
"mid": [
"2061949197",
"",
"2152556309"
],
"abstract": [
"In this study, a decoupled controller, consisting of a force controller and a torque controller, is designed to achieve a smooth translational and rotational motion control of a group of nonholonomic mobile robots. The proposed controller also solves the problem of obstacle avoidance, where obstacles with arbitrary boundary shapes are taken into account. Since the tangential direction of obstacle boundary is adopted as the guiding direction of a robot, the proposed controller allows a mobile robot to escape from a concave obstacle, while the robot could be trapped with most of the conventional obstacle avoidance algorithms.Copyright © 2005 by ASME",
"",
"This paper proposes a design methodology to stabilize isolated relative equilibria in a model of all-to-all coupled identical particles moving in the plane at unit speed. Isolated relative equilibria correspond to either parallel motion of all particles with fixed relative spacing or to circular motion of all particles with fixed relative phases. The stabilizing feedbacks derive from Lyapunov functions that prove exponential stability and suggest almost global convergence properties. The results of the paper provide a low-order parametric family of stabilizable collectives that offer a set of primitives for the design of higher-level tasks at the group level"
]
} |
1704.04905 | 2605877100 | A number of formal methods exist for capturing stimulus-response requirements in a declarative form. Someone yet needs to translate the resulting declarative statements into imperative programs. The present article describes a method for specification and verification of stimulus-response requirements in the form of imperative program routines with conditionals and assertions. A program prover then checks a candidate program directly against the stated requirements. The article illustrates the approach by applying it to an ASM model of the Landing Gear System, a widely used realistic example proposed for evaluating specification and verification techniques. | Modeling of real-time computation and related requirements is a well-investigated matter @cite_9 . Representation of real-time requirements, expressed in general or specific form, is a challenging task that has been attacked by the use of several formalisms both in sequential and concurrent settings, and in a broad set of application domains. The difficulty (or impossibility) to fully represents general real-time requirements other than in natural language, or making use of excessively complicated formalisms (unsuitable for software developers), has been recognized. | {
"cite_N": [
"@cite_9"
],
"mid": [
"2168723754"
],
"abstract": [
"As an attempt to investigate a general theory of real-time computability in digital computers, a subclass of Turing machines is formally introduced together with some classes of functions that are computable by them in real time. Then the existence is established of a class of recursive functions that are not computable in real time by use of a class of machines, no matter how general we make the machines subject to a given constraint."
]
} |
1704.04865 | 2607491080 | Traditional generative adversarial networks (GAN) and many of its variants are trained by minimizing the KL or JS-divergence loss that measures how close the generated data distribution is from the true data distribution. A recent advance called the WGAN based on Wasserstein distance can improve on the KL and JS-divergence based GANs, and alleviate the gradient vanishing, instability, and mode collapse issues that are common in the GAN training. In this work, we aim at improving on the WGAN by first generalizing its discriminator loss to a margin-based one, which leads to a better discriminator, and in turn a better generator, and then carrying out a progressive training paradigm involving multiple GANs to contribute to the maximum margin ranking loss so that the GAN at later stages will improve upon early stages. We call this method Gang of GANs (GoGAN). We have shown theoretically that the proposed GoGAN can reduce the gap between the true data distribution and the generated data distribution by at least half in an optimally trained WGAN. We have also proposed a new way of measuring GAN quality which is based on image completion tasks. We have evaluated our method on four visual datasets: CelebA, LSUN Bedroom, CIFAR-10, and 50K-SSFF, and have seen both visual and quantitative improvement over baseline WGAN. | Other applications include cross-domain image generation @cite_27 through a domain transfer network (DTN) which employs a compound of loss functions including a multi-class GAN loss, an @math -constancy component, and a regularization component that encourages the generator to map samples from the target domain to themselves. The image-to-image translation approach @cite_14 is based on conditional GAN, and learns a conditional generative model for generating a corresponding output image at a different domain, conditioned on an input image. The image super-resolution GAN (SRGAN) @cite_34 combines both the image content loss and the adversarial loss for recovering high-resolution counterpart of the low-resolution input image. The plug and play generative networks (PPGN) @cite_25 is able to produce high quality images at higher resolution for all 1000 ImageNet categories. It is composed of a generator that is capable of drawing a wide range of image types, and a replaceable condition network that tells the generator what to draw, hence plug and play. | {
"cite_N": [
"@cite_27",
"@cite_14",
"@cite_25",
"@cite_34"
],
"mid": [
"2553897675",
"2552465644",
"2951140085",
"2523714292"
],
"abstract": [
"We study the problem of transferring a sample in one domain to an analog sample in another domain. Given two related domains, S and T, we would like to learn a generative function G that maps an input sample from S to the domain T, such that the output of a given function f, which accepts inputs in either domains, would remain unchanged. Other than the function f, the training data is unsupervised and consist of a set of samples from each domain. The Domain Transfer Network (DTN) we present employs a compound loss function that includes a multiclass GAN loss, an f-constancy component, and a regularizing component that encourages G to map samples from T to themselves. We apply our method to visual domains including digits and face images and demonstrate its ability to generate convincing novel images of previously unseen entities, while preserving their identity.",
"We investigate conditional adversarial networks as a general-purpose solution to image-to-image translation problems. These networks not only learn the mapping from input image to output image, but also learn a loss function to train this mapping. This makes it possible to apply the same generic approach to problems that traditionally would require very different loss formulations. We demonstrate that this approach is effective at synthesizing photos from label maps, reconstructing objects from edge maps, and colorizing images, among other tasks. Indeed, since the release of the pix2pix software associated with this paper, a large number of internet users (many of them artists) have posted their own experiments with our system, further demonstrating its wide applicability and ease of adoption without the need for parameter tweaking. As a community, we no longer hand-engineer our mapping functions, and this work suggests we can achieve reasonable results without hand-engineering our loss functions either.",
"Generating high-resolution, photo-realistic images has been a long-standing goal in machine learning. Recently, (2016) showed one interesting way to synthesize novel images by performing gradient ascent in the latent space of a generator network to maximize the activations of one or multiple neurons in a separate classifier network. In this paper we extend this method by introducing an additional prior on the latent code, improving both sample quality and sample diversity, leading to a state-of-the-art generative model that produces high quality images at higher resolutions (227x227) than previous generative models, and does so for all 1000 ImageNet categories. In addition, we provide a unified probabilistic interpretation of related activation maximization methods and call the general class of models \"Plug and Play Generative Networks\". PPGNs are composed of 1) a generator network G that is capable of drawing a wide range of image types and 2) a replaceable \"condition\" network C that tells the generator what to draw. We demonstrate the generation of images conditioned on a class (when C is an ImageNet or MIT Places classification network) and also conditioned on a caption (when C is an image captioning network). Our method also improves the state of the art of Multifaceted Feature Visualization, which generates the set of synthetic inputs that activate a neuron in order to better understand how deep neural networks operate. Finally, we show that our model performs reasonably well at the task of image inpainting. While image models are used in this paper, the approach is modality-agnostic and can be applied to many types of data.",
"Despite the breakthroughs in accuracy and speed of single image super-resolution using faster and deeper convolutional neural networks, one central problem remains largely unsolved: how do we recover the finer texture details when we super-resolve at large upscaling factors? The behavior of optimization-based super-resolution methods is principally driven by the choice of the objective function. Recent work has largely focused on minimizing the mean squared reconstruction error. The resulting estimates have high peak signal-to-noise ratios, but they are often lacking high-frequency details and are perceptually unsatisfying in the sense that they fail to match the fidelity expected at the higher resolution. In this paper, we present SRGAN, a generative adversarial network (GAN) for image super-resolution (SR). To our knowledge, it is the first framework capable of inferring photo-realistic natural images for 4x upscaling factors. To achieve this, we propose a perceptual loss function which consists of an adversarial loss and a content loss. The adversarial loss pushes our solution to the natural image manifold using a discriminator network that is trained to differentiate between the super-resolved images and original photo-realistic images. In addition, we use a content loss motivated by perceptual similarity instead of similarity in pixel space. Our deep residual network is able to recover photo-realistic textures from heavily downsampled images on public benchmarks. An extensive mean-opinion-score (MOS) test shows hugely significant gains in perceptual quality using SRGAN. The MOS scores obtained with SRGAN are closer to those of the original high-resolution images than to those obtained with any state-of-the-art method."
]
} |
1704.04859 | 2608068008 | Previous work has modeled the compositionality of words by creating character-level models of meaning, reducing problems of sparsity for rare words. However, in many writing systems compositionality has an effect even on the character-level: the meaning of a character is derived by the sum of its parts. In this paper, we model this effect by creating embeddings for characters based on their visual characteristics, creating an image for the character and running it through a convolutional neural network to produce a visual character embedding. Experiments on a text classification task demonstrate that such model allows for better processing of instances with rare characters in languages such as Chinese, Japanese, and Korean. Additionally, qualitative analyses demonstrate that our proposed model learns to focus on the parts of characters that carry semantic content, resulting in embeddings that are coherent in visual space. | There is also work done in understanding mathematical expressions with a convolutional network for text and layout recognition by using an attention-based neural machine translation system @cite_20 . They tested on real-world rendered mathematical expressions paired with LaTeX markup and show the system is effective at generating accurate markup. Other than that, there are several works that combine visual information with text in improving machine translation @cite_13 , visual question answering, caption generation @cite_33 , etc. These works extract image representations from a pre-trained CNN @cite_9 @cite_34 . | {
"cite_N": [
"@cite_33",
"@cite_9",
"@cite_34",
"@cite_13",
"@cite_20"
],
"mid": [
"2950178297",
"2136462581",
"",
"2949888546",
"2521665229"
],
"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.",
"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.",
"",
"Deep Neural Networks (DNNs) are powerful models that have achieved excellent performance on difficult learning tasks. Although DNNs work well whenever large labeled training sets are available, they cannot be used to map sequences to sequences. In this paper, we present a general end-to-end approach to sequence learning that makes minimal assumptions on the sequence structure. Our method uses a multilayered Long Short-Term Memory (LSTM) to map the input sequence to a vector of a fixed dimensionality, and then another deep LSTM to decode the target sequence from the vector. Our main result is that on an English to French translation task from the WMT'14 dataset, the translations produced by the LSTM achieve a BLEU score of 34.8 on the entire test set, where the LSTM's BLEU score was penalized on out-of-vocabulary words. Additionally, the LSTM did not have difficulty on long sentences. For comparison, a phrase-based SMT system achieves a BLEU score of 33.3 on the same dataset. When we used the LSTM to rerank the 1000 hypotheses produced by the aforementioned SMT system, its BLEU score increases to 36.5, which is close to the previous best result on this task. The LSTM also learned sensible phrase and sentence representations that are sensitive to word order and are relatively invariant to the active and the passive voice. Finally, we found that reversing the order of the words in all source sentences (but not target sentences) improved the LSTM's performance markedly, because doing so introduced many short term dependencies between the source and the target sentence which made the optimization problem easier.",
"Building on recent advances in image caption generation and optical character recognition (OCR), we present a general-purpose, deep learning-based system to decompile an image into presentational markup. While this task is a well-studied problem in OCR, our method takes an inherently different, data-driven approach. Our model does not require any knowledge of the underlying markup language, and is simply trained end-to-end on real-world example data. The model employs a convolutional network for text and layout recognition in tandem with an attention-based neural machine translation system. To train and evaluate the model, we introduce a new dataset of real-world rendered mathematical expressions paired with LaTeX markup, as well as a synthetic dataset of web pages paired with HTML snippets. Experimental results show that the system is surprisingly effective at generating accurate markup for both datasets. While a standard domain-specific LaTeX OCR system achieves around 25 accuracy, our model reproduces the exact rendered image on 75 of examples."
]
} |
1704.04859 | 2608068008 | Previous work has modeled the compositionality of words by creating character-level models of meaning, reducing problems of sparsity for rare words. However, in many writing systems compositionality has an effect even on the character-level: the meaning of a character is derived by the sum of its parts. In this paper, we model this effect by creating embeddings for characters based on their visual characteristics, creating an image for the character and running it through a convolutional neural network to produce a visual character embedding. Experiments on a text classification task demonstrate that such model allows for better processing of instances with rare characters in languages such as Chinese, Japanese, and Korean. Additionally, qualitative analyses demonstrate that our proposed model learns to focus on the parts of characters that carry semantic content, resulting in embeddings that are coherent in visual space. | Several techniques for reducing the rare words effects have been introduced in the literature, including spelling expansion @cite_7 , dictionary term expansion @cite_7 , proper name transliteration @cite_25 , treating words as a sequence of characters @cite_19 , subword units @cite_35 , and reading text as bytes @cite_38 . However, most of these techniques still have no mechanism for handling low frequency characters, which are the target of this work. | {
"cite_N": [
"@cite_38",
"@cite_35",
"@cite_7",
"@cite_19",
"@cite_25"
],
"mid": [
"2185720331",
"2962784628",
"2153903004",
"2339995566",
"2172268343"
],
"abstract": [
"We describe an LSTM-based model which we call Byte-to-Span (BTS) that reads text as bytes and outputs span annotations of the form [start, length, label] where start positions, lengths, and labels are separate entries in our vocabulary. Because we operate directly on unicode bytes rather than language-specific words or characters, we can analyze text in many languages with a single model. Due to the small vocabulary size, these multilingual models are very compact, but produce results similar to or better than the state-of- the-art in Part-of-Speech tagging and Named Entity Recognition that use only the provided training datasets (no external data sources). Our models are learning \"from scratch\" in that they do not rely on any elements of the standard pipeline in Natural Language Processing (including tokenization), and thus can run in standalone fashion on raw text.",
"Neural machine translation (NMT) models typically operate with a fixed vocabulary, but translation is an open-vocabulary problem. Previous work addresses the translation of out-of-vocabulary words by backing off to a dictionary. In this paper, we introduce a simpler and more effective approach, making the NMT model capable of open-vocabulary translation by encoding rare and unknown words as sequences of subword units. This is based on the intuition that various word classes are translatable via smaller units than words, for instance names (via character copying or transliteration), compounds (via compositional translation), and cognates and loanwords (via phonological and morphological transformations). We discuss the suitability of different word segmentation techniques, including simple character ngram models and a segmentation based on the byte pair encoding compression algorithm, and empirically show that subword models improve over a back-off dictionary baseline for the WMT 15 translation tasks English!German and English!Russian by up to 1.1 and 1.3 BLEU, respectively.",
"We present four techniques for online handling of Out-of-Vocabulary words in Phrase-based Statistical Machine Translation. The techniques use spelling expansion, morphological expansion, dictionary term expansion and proper name transliteration to reuse or extend a phrase table. We compare the performance of these techniques and combine them. Our results show a consistent improvement over a state-of-the-art baseline in terms of BLEU and a manual error analysis.",
"Nearly all previous work on neural machine translation (NMT) has used quite restricted vocabularies, perhaps with a subsequent method to patch in unknown words. This paper presents a novel word-character solution to achieving open vocabulary NMT. We build hybrid systems that translate mostly at the word level and consult the character components for rare words. Our character-level recurrent neural networks compute source word representations and recover unknown target words when needed. The twofold advantage of such a hybrid approach is that it is much faster and easier to train than character-based ones; at the same time, it never produces unknown words as in the case of word-based models. On the WMT'15 English to Czech translation task, this hybrid approach offers an addition boost of +2.1-11.4 BLEU points over models that already handle unknown words. Our best system achieves a new state-of-the-art result with 20.7 BLEU score. We demonstrate that our character models can successfully learn to not only generate well-formed words for Czech, a highly-inflected language with a very complex vocabulary, but also build correct representations for English source words.",
"We show that unseen words account for a large part of the translation error when moving to new domains. Using an extension of a recent approach to mining translations from comparable corpora (, 2008), we are able to find translations for otherwise OOV terms. We show several approaches to integrating such translations into a phrase-based translation system, yielding consistent improvements in translations quality (between 0.5 and 1.5 Bleu points) on four domains and two language pairs."
]
} |
1704.04859 | 2608068008 | Previous work has modeled the compositionality of words by creating character-level models of meaning, reducing problems of sparsity for rare words. However, in many writing systems compositionality has an effect even on the character-level: the meaning of a character is derived by the sum of its parts. In this paper, we model this effect by creating embeddings for characters based on their visual characteristics, creating an image for the character and running it through a convolutional neural network to produce a visual character embedding. Experiments on a text classification task demonstrate that such model allows for better processing of instances with rare characters in languages such as Chinese, Japanese, and Korean. Additionally, qualitative analyses demonstrate that our proposed model learns to focus on the parts of characters that carry semantic content, resulting in embeddings that are coherent in visual space. | Finally, there are works on improving embeddings with radicals, which explicitly splits Chinese characters into radicals based on a dictionary of what radicals are included in which characters @cite_22 @cite_28 @cite_37 . The motivation of this method is similar to ours, but is only applicable to Chinese, in contrast to the method in this paper, which works on any language for which we can render text. | {
"cite_N": [
"@cite_28",
"@cite_37",
"@cite_22"
],
"mid": [
"2251131401",
"2566150155",
"1912843035"
],
"abstract": [
"Languages using Chinese characters are mostly processed at word level. Inspired by recent success of deep learning, we delve deeper to character and radical levels for Chinese language processing. We propose a new deep learning technique, called “radical embedding”, with justifications based on Chinese linguistics, and validate its feasibility and utility through a set of three experiments: two in-house standard experiments on short-text categorization (STC) and Chinese word segmentation (CWS), and one in-field experiment on search ranking. We show that radical embedding achieves comparable, and sometimes even better, results than competing methods.",
"",
"Distributed word representations are very useful for capturing semantic information and have been successfully applied in a variety of NLP tasks, especially on English. In this work, we innovatively develop two component-enhanced Chinese character embedding models and their bigram extensions. Distinguished from English word embeddings, our models explore the compositions of Chinese characters, which often serve as semantic indictors inherently. The evaluations on both word similarity and text classification demonstrate the effectiveness of our models."
]
} |
1704.04574 | 2956100892 | Unmanned aerial vehicles (UAVs) have gained much attention in recent years for both commercial and military applications. The progress in this field has gained much popularity and the research has encompassed various fields of scientific domain. Cyber securing a UAV communication has been one of the active research field since the attack on Predator UAV video stream hijacking in 2009. Since UAVs rely heavily on on-board autopilot to function, it is important to develop an autopilot system that is robust to possible cyber attacks. In this work, we present a biometric system to encrypt the UAV communication by generating a key which is derived from Beta component of the EEG signal of a user. We have developed a safety mechanism that would be activated in case the communication of the UAV from the ground control station gets attacked. This system has been validated on a commercial UAV under malicious attack conditions during which we implement a procedure where the UAV return safely to a "home" position. | Most UAV systems are moving their infrastructure towards more network-centric command and control, where all of the components are interconnected through sophisticated mesh networks @cite_1 . This enables fast communication and constant environmental and asset awareness, but introduces security drawbacks. Some military UAV systems, such as the Global Hawk, already employ this type of infrastructure. Public safety and disaster management UAVs are also moving to a similar network architecture for planning and communication @cite_6 . When the components of the system are interconnected through such a network, a compromise of one component can cause a propagation of failures or malicious behavior can occur throughout the whole system. | {
"cite_N": [
"@cite_1",
"@cite_6"
],
"mid": [
"2152203736",
"2534600276"
],
"abstract": [
"The purpose of this investigative study is to develop a model for a system of systems to improve situation awareness and targeting through a real-time aerially deployed wireless sensor network. The main hypothesis of this work is that scalable, affordable, real-time wireless network can contribute to the common operating picture. This network can be used to overcome targeting discrepancies and counter asymmetric threats in the new warfare paradigm that pertains to opaque environments. Surveys identified requirements for a system of systems to meet requirements for situation assessment. This paper describes 1. a model to support the hypothesis that this network can to improve response to asymmetric threats, and 2. a system architecture based on commercial-off-the-shelf technology for military operations in this new combat paradigm.",
"Developed societies have a high level of preparedness for natural or man-made disasters. But such incidents cannot be completely prevented, and when an incident like an earthquake or an accident in a chemical or nuclear plant hits a populated area, rescue teams need to be employed. In such situations it is a necessity for rescue teams to get a quick overview of the situation in order to identify possible locations of victims that need to be rescued and dangerous locations that need to be secured. Rescue forces must operate quickly in order to save lives, and they often need to operate in dangerous environments. Hence, robot-supported systems are increasingly used to support and accelerate search operations. The objective of the SENEKA concept is to network the various robots and sensor systems used by first responders in order to make the search for victims and survivors more quick and efficient. SENEKA targets the integration of the robot-sensor network into the operation procedures of the rescue teams. The aim of this paper is to inform on the goals and first research results of the ongoing joint research project SENEKA."
]
} |
1704.04574 | 2956100892 | Unmanned aerial vehicles (UAVs) have gained much attention in recent years for both commercial and military applications. The progress in this field has gained much popularity and the research has encompassed various fields of scientific domain. Cyber securing a UAV communication has been one of the active research field since the attack on Predator UAV video stream hijacking in 2009. Since UAVs rely heavily on on-board autopilot to function, it is important to develop an autopilot system that is robust to possible cyber attacks. In this work, we present a biometric system to encrypt the UAV communication by generating a key which is derived from Beta component of the EEG signal of a user. We have developed a safety mechanism that would be activated in case the communication of the UAV from the ground control station gets attacked. This system has been validated on a commercial UAV under malicious attack conditions during which we implement a procedure where the UAV return safely to a "home" position. | An interesting perspective which attempts to solve the cyber-security aspect of UAVs considers the whole scenario of vendor and an attacker as a zero-sum network interdiction game. It is represented as a game where the vendor, also assumed an evader, seeks to choose the optimal path strategy for its UAV, from a source location to a destination location, to evade attacks along the way and minimize its expected delivery time. On the other hand, the attacker or interdictor, aims at choosing the optimal attack locations along the paths traversed by the UAV to interdict the UAV, causing cyber or physical damage, with the goal of maximizing the travel time. Later on it is shown that this network interdiction game is equivalent to a zero-sum matrix game whose Nash equilibrium (NE) can be derived by solving two linear programming (LP) problems. Solving the LP's would give the expected delivery time under different conditions @cite_19 . | {
"cite_N": [
"@cite_19"
],
"mid": [
"2592577925"
],
"abstract": [
"The use of unmanned aerial vehicles (UAVs) as delivery systems of online goods is rapidly becoming a global norm, as corroborated by Amazon's “Prime Air” and Google's “Project Wing” projects. However, the real-world deployment of such drone delivery systems faces many cyber-physical security challenges. In this paper, a novel mathematical framework for analyzing and enhancing the security of drone delivery systems is introduced. In this regard, a zero-sum network interdiction game is formulated between a vendor, operating a drone delivery system, and a malicious attacker. In this game, the vendor seeks to find the optimal path that its UAV should follow, to deliver a purchase from the vendor's warehouse to a customer location, to minimize the delivery time. Meanwhile, an attacker seeks to choose an optimal location to interdict the potential paths of the UAVs, so as to inflict cyber or physical damage to it, thus, maximizing its delivery time. First, the Nash equilibrium point of this game is characterized. Then, to capture the subjective behavior of both the vendor and attacker, new notions from prospect theory are incorporated into the game. These notions allow capturing the vendor's and attacker's i) subjective perception of attack success probabilities, and ii) their disparate subjective valuations of the achieved delivery times relative to a certain target delivery time. Simulation results have shown that the subjective decision making of the vendor and attacker leads to adopting risky path selection strategies which inflict delays to the delivery, thus, yielding unexpected delivery times which surpass the target delivery time set by the vendor."
]
} |
1704.04574 | 2956100892 | Unmanned aerial vehicles (UAVs) have gained much attention in recent years for both commercial and military applications. The progress in this field has gained much popularity and the research has encompassed various fields of scientific domain. Cyber securing a UAV communication has been one of the active research field since the attack on Predator UAV video stream hijacking in 2009. Since UAVs rely heavily on on-board autopilot to function, it is important to develop an autopilot system that is robust to possible cyber attacks. In this work, we present a biometric system to encrypt the UAV communication by generating a key which is derived from Beta component of the EEG signal of a user. We have developed a safety mechanism that would be activated in case the communication of the UAV from the ground control station gets attacked. This system has been validated on a commercial UAV under malicious attack conditions during which we implement a procedure where the UAV return safely to a "home" position. | One potential solution is to use biometric information to secure communication between a UAV and its command and control station. This would allow the UAV to verify that its stated operator is issuing the commands to the UAV. To the best of our knowledge, biometric UAV authentication has been limited to facial recognition alone. Facial authentication is problematic since it can be easily deceived by an attacker if they have a picture of the actual operator @cite_15 . In this way, a more secure biometric feature is needed. We propose to use EEG signal characteristics to secure communication between an operator and a UAV. | {
"cite_N": [
"@cite_15"
],
"mid": [
"1979339361"
],
"abstract": [
"A common technique to by-pass 2-D face recognition systems is to use photographs of spoofed identities. Unfortunately, research in counter-measures to this type of attack have not kept-up - even if such threats have been known for nearly a decade, there seems to exist no consensus on best practices, techniques or protocols for developing and testing spoofing-detectors for face recognition. We attribute the reason for this delay, partly, to the unavailability of public databases and protocols to study solutions and compare results. To this purpose we introduce the publicly available PRINT-ATTACK database and exemplify how to use its companion protocol with a motion-based algorithm that detects correlations between the person's head movements and the scene context. The results are to be used as basis for comparison to other counter-measure techniques. The PRINT-ATTACK database contains 200 videos of real-accesses and 200 videos of spoof attempts using printed photographs of 50 different identities."
]
} |
1704.04601 | 2949460680 | This paper proposes to address the word sense ambiguity issue in an unsupervised manner, where word sense representations are learned along a word sense selection mechanism given contexts. Prior work focused on designing a single model to deliver both mechanisms, and thus suffered from either coarse-grained representation learning or inefficient sense selection. The proposed modular approach, MUSE, implements flexible modules to optimize distinct mechanisms, achieving the first purely sense-level representation learning system with linear-time sense selection. We leverage reinforcement learning to enable joint training on the proposed modules, and introduce various exploration techniques on sense selection for better robustness. The experiments on benchmark data show that the proposed approach achieves the state-of-the-art performance on synonym selection as well as on contextual word similarities in terms of MaxSimC. | first proposed multi-sense word representations on the vector space based on clustering techniques. With the power of deep learning, some work exploited neural networks to learn embeddings with sense selection based on clustering @cite_2 @cite_24 . replaced the clustering procedure with a word sense disambiguation model using WordNet @cite_0 . and further leveraged a weighting mechanism and interactive process in the clustering procedure. Moreover, leveraged bilingual resources for clustering. However, most of the above approaches separated the clustering procedure and the representation learning procedure without a joint objective, which may suffer from the error propagation issue. Instead, the proposed approach, MUSE, enables joint training on sense selection and representation learning. | {
"cite_N": [
"@cite_24",
"@cite_0",
"@cite_2"
],
"mid": [
"2949364118",
"2081580037",
"2164019165"
],
"abstract": [
"There is rising interest in vector-space word embeddings and their use in NLP, especially given recent methods for their fast estimation at very large scale. Nearly all this work, however, assumes a single vector per word type ignoring polysemy and thus jeopardizing their usefulness for downstream tasks. We present an extension to the Skip-gram model that efficiently learns multiple embeddings per word type. It differs from recent related work by jointly performing word sense discrimination and embedding learning, by non-parametrically estimating the number of senses per word type, and by its efficiency and scalability. We present new state-of-the-art results in the word similarity in context task and demonstrate its scalability by training with one machine on a corpus of nearly 1 billion tokens in less than 6 hours.",
"Because meaningful sentences are composed of meaningful words, any system that hopes to process natural languages as people do must have information about words and their meanings. This information is traditionally provided through dictionaries, and machine-readable dictionaries are now widely available. But dictionary entries evolved for the convenience of human readers, not for machines. WordNet 1 provides a more effective combination of traditional lexicographic information and modern computing. WordNet is an online lexical database designed for use under program control. English nouns, verbs, adjectives, and adverbs are organized into sets of synonyms, each representing a lexicalized concept. Semantic relations link the synonym sets [4].",
"Unsupervised word representations are very useful in NLP tasks both as inputs to learning algorithms and as extra word features in NLP systems. However, most of these models are built with only local context and one representation per word. This is problematic because words are often polysemous and global context can also provide useful information for learning word meanings. We present a new neural network architecture which 1) learns word embeddings that better capture the semantics of words by incorporating both local and global document context, and 2) accounts for homonymy and polysemy by learning multiple embeddings per word. We introduce a new dataset with human judgments on pairs of words in sentential context, and evaluate our model on it, showing that our model outperforms competitive baselines and other neural language models."
]
} |
1704.04601 | 2949460680 | This paper proposes to address the word sense ambiguity issue in an unsupervised manner, where word sense representations are learned along a word sense selection mechanism given contexts. Prior work focused on designing a single model to deliver both mechanisms, and thus suffered from either coarse-grained representation learning or inefficient sense selection. The proposed modular approach, MUSE, implements flexible modules to optimize distinct mechanisms, achieving the first purely sense-level representation learning system with linear-time sense selection. We leverage reinforcement learning to enable joint training on the proposed modules, and introduce various exploration techniques on sense selection for better robustness. The experiments on benchmark data show that the proposed approach achieves the state-of-the-art performance on synonym selection as well as on contextual word similarities in terms of MaxSimC. | Instead of clustering, probabilistic modeling methods have been applied for learning multi-sense embeddings in order to make the sense selection more flexible, where and conducted probabilistic modeling with EM training. exploited Chinese Restaurant Process to infer the sense identity. Furthermore, developed a non-parametric Bayesian extension on the skip-gram model @cite_10 . Despite reasonable modeling on sense selection, all above methods mixed word-level and sense-level tokens during representation learning---unable to conduct representation learning in the pure sense level due to the complicated computation in their EM algorithms. Recently, proposed an EM algorithm to learn purely sense-level representations, where the computational cost is high when decoding the sense identity sequence, because it takes exponential time to search all sense combination within a context window. Our modular design addresses such drawback, where the sense selection module decodes a sense sequence with linear-time complexity, while the sense representation module remains representation learning in the pure sense level. | {
"cite_N": [
"@cite_10"
],
"mid": [
"2950133940"
],
"abstract": [
"The recently introduced continuous Skip-gram model is an efficient method for learning high-quality distributed vector representations that capture a large number of precise syntactic and semantic word relationships. In this paper we present several extensions that improve both the quality of the vectors and the training speed. By subsampling of the frequent words we obtain significant speedup and also learn more regular word representations. We also describe a simple alternative to the hierarchical softmax called negative sampling. An inherent limitation of word representations is their indifference to word order and their inability to represent idiomatic phrases. For example, the meanings of \"Canada\" and \"Air\" cannot be easily combined to obtain \"Air Canada\". Motivated by this example, we present a simple method for finding phrases in text, and show that learning good vector representations for millions of phrases is possible."
]
} |
1704.04937 | 2952815199 | Browsers can detect malicious websites that are provisioned with forged or fake TLS SSL certificates. However, they are not so good at detecting malicious websites if they are provisioned with mistakenly issued certificates or certificates that have been issued by a compromised certificate authority. Google proposed certificate transparency which is an open framework to monitor and audit certificates in real time. Thereafter, a few other certificate transparency schemes have been proposed which can even handle revocation. All currently known constructions use Merkle hash trees and have proof size logarithmic in the number of certificates domain owners. We present a new certificate transparency scheme with short (constant size) proofs. Our construction makes use of dynamic bilinear-map accumulators. The scheme has many desirable properties like efficient revocation, low verification cost and update costs comparable to the existing schemes. We provide proofs of security and evaluate the performance of our scheme. | Recently, Dowling @cite_33 proposed a formal framework of secure logging in general. Formal security proofs for secure logging have been presented. The underlying framework consider Google's framework and does not support revocation of domain owners. However, only the proof of presence has been considered in this work. A different line of work is accountable PKI (APKI) @cite_10 @cite_31 , which has been used to address the decrease in trust on CAs. APKI efficiently handles revocation and is still a tree-based protocol. Deployment challenges for log based PKI enhancements have been presented in @cite_8 . Accountability in future internet has been discussed in @cite_28 . Key transparency is another closely related line of work. It differs from the certificate transparency in that, in key transparency, there is a requirement to protect the secrecy of the public keys of the domain owners. Coniks @cite_29 is an example of a key transparency scheme. | {
"cite_N": [
"@cite_33",
"@cite_8",
"@cite_28",
"@cite_29",
"@cite_31",
"@cite_10"
],
"mid": [
"",
"2060371875",
"2083605389",
"2116428199",
"2099175354",
"2294157280"
],
"abstract": [
"",
"Log-based PKI enhancements propose to improve the current TLS PKI by creating public logs to monitor CA operations, thus providing transparency and accountability. In this paper we take the first steps in studying the deployment process of log-based PKI enhancements in two ways. First, we model the influences that parties in the PKI have to incentivize one another to deploy a PKI enhancement, and determine that potential PKI enhancements should focus their initial efforts on convincing browser vendors to deploy. Second, as a promising vendor-based solution we propose deployment status filters, which use a Bloom filter to monitor deployment status and efficiently defend against downgrade attacks from the enhanced protocol to the current TLS PKI. Our results provide promising deployment strategies for log-based PKI enhancements and raise additional questions for further fruitful research.",
"",
"We present CONIKS, an end-user key verification service capable of integration in end-to-end encrypted communication systems. CONIKS builds on transparency log proposals for web server certificates but solves several new challenges specific to key verification for end users. CONIKS obviates the need for global third-party monitors and enables users to efficiently monitor their own key bindings for consistency, downloading less than 20 kB per day to do so even for a provider with billions of users. CONIKS users and providers can collectively audit providers for non-equivocation, and this requires downloading a constant 2.5 kB per provider per day. Additionally, CONIKS preserves the level of privacy offered by today's major communication services, hiding the list of usernames present and even allowing providers to conceal the total number of users in the system.",
"We present ARPKI, a public-key infrastructure that ensures that certificate-related operations, such as certificate issuance, update, revocation, and validation, are transparent and accountable. ARPKI is the first such infrastructure that systematically takes into account requirements identified by previous research. Moreover, ARPKI is co-designed with a formal model, and we verify its core security property using the Tamarin prover. We present a proof-of-concept implementation providing all features required for deployment. ARPKI efficiently handles the certification process with low overhead and without incurring additional latency to TLS. ARPKI offers extremely strong security guarantees, where compromising n-1 trusted signing and verifying entities is insufficient to launch an impersonation attack. Moreover, it deters misbehavior as all its operations are publicly visible.",
"Recent trends in public-key infrastructure research explore the tradeoff between decreased trust in Certificate Authorities (CAs), resilience against attacks, communication overhead (bandwidth and latency) for setting up an SSL TLS connection, and availability with respect to verifiability of public key information. In this paper, we propose AKI as a new public-key validation infrastructure, to reduce the level of trust in CAs. AKI integrates an architecture for key revocation of all entities (e.g., CAs, domains) with an architecture for accountability of all infrastructure parties through checks-and-balances. AKI efficiently handles common certification operations, and gracefully handles catastrophic events such as domain key loss or compromise. We propose AKI to make progress towards a public-key validation infrastructure with key revocation that reduces trust in any single entity."
]
} |
1704.04937 | 2952815199 | Browsers can detect malicious websites that are provisioned with forged or fake TLS SSL certificates. However, they are not so good at detecting malicious websites if they are provisioned with mistakenly issued certificates or certificates that have been issued by a compromised certificate authority. Google proposed certificate transparency which is an open framework to monitor and audit certificates in real time. Thereafter, a few other certificate transparency schemes have been proposed which can even handle revocation. All currently known constructions use Merkle hash trees and have proof size logarithmic in the number of certificates domain owners. We present a new certificate transparency scheme with short (constant size) proofs. Our construction makes use of dynamic bilinear-map accumulators. The scheme has many desirable properties like efficient revocation, low verification cost and update costs comparable to the existing schemes. We provide proofs of security and evaluate the performance of our scheme. | All these schemes maintain central log maintainers. Some Blockchain based solutions are also available. A decentralized PKI setting has been proposed in @cite_3 where domain owners are incentivized in a cryptocurrency called Certcoin to perform audits. The blockchain implementation of Coniks on Ethereum appeared in @cite_0 . Wilson and Ateniese recently exploited Bitcoin and its blockchain to store and retrieve Bitcoin-based PGP certificates @cite_1 . | {
"cite_N": [
"@cite_0",
"@cite_1",
"@cite_3"
],
"mid": [
"2513866749",
"1879433355",
""
],
"abstract": [
"CONIKS is a proposed key transparency system which enables a centralized service provider to maintain an auditable yet privacy-preserving directory of users’ public keys. In the original CONIKS design, users must monitor that their data is correctly included in every published snapshot of the directory, necessitating either slow updates or trust in an unspecified third-party to audit that the data structure has stayed consistent. We demonstrate that the data structures for CONIKS are very similar to those used in Ethereum, a consensus computation platform with a Turing-complete programming environment. We can take advantage of this to embed the core CONIKS data structures into an Ethereum contract with only minor modifications. Users may then trust the Ethereum network to audit the data structure for consistency and non-equivocation. Users who do not trust (or are unaware of) Ethereum can self-audit the CONIKS data structure as before. We have implemented a prototype contract for our hybrid EthIKS scheme, demonstrating that it adds only modest bandwidth overhead to CONIKS proofs and costs hundredths of pennies per key update in fees at today’s rates.",
"PGP is built upon a Distributed Web of Trust in which a user’s trustworthiness is established by others who can vouch through a digital signature for that user’s identity. Preventing its wholesale adoption are a number of inherent weaknesses to include (but not limited to) the following: 1) Trust Relationships are built on a subjective honor system, 2) Only first degree relationships can be fully trusted, 3) Levels of trust are difficult to quantify with actual values, and 4) Issues with the Web of Trust itself (Certification and Endorsement). Although the security that PGP provides is proven to be reliable, it has largely failed to garner large scale adoption. In this paper, we propose several novel contributions to address the aforementioned issues with PGP and associated Web of Trust. To address the subjectivity of the Web of Trust, we provide a new certificate format based on Bitcoin which allows a user to verify a PGP certificate using Bitcoin identity-verification transactions - forming first degree trust relationships that are tied to actual values (i.e., number of Bitcoins transferred during transaction). Secondly, we present the design of a novel Distributed PGP key server that leverages the Bitcoin transaction blockchain to store and retrieve our certificates.",
""
]
} |
1704.04997 | 2612496865 | Professional-grade software applications are powerful but complicated @math expert users can achieve impressive results, but novices often struggle to complete even basic tasks. Photo editing is a prime example: after loading a photo, the user is confronted with an array of cryptic sliders like "clarity", "temp", and "highlights". An automatically generated suggestion could help, but there is no single "correct" edit for a given image @math different experts may make very different aesthetic decisions when faced with the same image, and a single expert may make different choices depending on the intended use of the image (or on a whim). We therefore want a system that can propose multiple diverse, high-quality edits while also learning from and adapting to a user's aesthetic preferences. In this work, we develop a statistical model that meets these objectives. Our model builds on recent advances in neural network generative modeling and scalable inference, and uses hierarchical structure to learn editing patterns across many diverse users. Empirically, we find that our model outperforms other approaches on this challenging multimodal prediction task. | @cite_3 extend the VAE inference scheme to latent graphical models with neural network observation distributions. This SVAE framework combines the interpretability of graphical models with the flexible representations found by deep learning. For example, consider a latent Gaussian mixture model (GMM) with nonlinear observations: [ Dir ( ), ( , ) & NIW ( ), s_n iid , z_n|s_n, , ^ K _ k=1 & N ( s_n , s_n ), y_n | z_n, & N ( (z_n; ), (z_n; )) ,. ] Note that the nonlinear observation model for each data point @math resembles that of the VAE, while the latent variable model for @math is a GMM (see Figure (b)). This latent GMM can represent explicit latent cluster assignments while also capturing complex non-Gaussian cluster shapes in the observations. | {
"cite_N": [
"@cite_3"
],
"mid": [
"2962695963"
],
"abstract": [
"We propose a general modeling and inference framework that combines the complementary strengths of probabilistic graphical models and deep learning methods. Our model family composes latent graphical models with neural network observation likelihoods. For inference, we use recognition networks to produce local evidence potentials, then combine them with the model distribution using efficient message-passing algorithms. All components are trained simultaneously with a single stochastic variational inference objective. We illustrate this framework by automatically segmenting and categorizing mouse behavior from raw depth video, and demonstrate several other example models."
]
} |
1704.04997 | 2612496865 | Professional-grade software applications are powerful but complicated @math expert users can achieve impressive results, but novices often struggle to complete even basic tasks. Photo editing is a prime example: after loading a photo, the user is confronted with an array of cryptic sliders like "clarity", "temp", and "highlights". An automatically generated suggestion could help, but there is no single "correct" edit for a given image @math different experts may make very different aesthetic decisions when faced with the same image, and a single expert may make different choices depending on the intended use of the image (or on a whim). We therefore want a system that can propose multiple diverse, high-quality edits while also learning from and adapting to a user's aesthetic preferences. In this work, we develop a statistical model that meets these objectives. Our model builds on recent advances in neural network generative modeling and scalable inference, and uses hierarchical structure to learn editing patterns across many diverse users. Empirically, we find that our model outperforms other approaches on this challenging multimodal prediction task. | Due to the non-conjugate likelihood function @math , standard variational inference methods cannot be applied to the latent GMM. To solve this problem, the SVAE replaces the non-conjugate likelihood with a recognition model @math that generates conjugate evidence potentials. We can then define a surrogate objective @math with conjugacy structure: [ L ( , z , ) & E _ q( )q(z) & [ p( )p(z| ) r(y; ), t_z(z) q( )q(z) ] ] where the potentials @math have a conjugate form to @math . By choosing @math to optimize this surrogate objective, writing @math , the SVAE objective is then @math which can be shown to lower bound the variational inference objective in eq. . As in the stochastic variational inference (SVI) algorithm @cite_1 , there is a simple expression for the natural gradient of this objective with respect to the variational parameters with conjugate priors; the gradients w.r.t. other variational parameters, such as those parameterizing neural networks, can be computed using the reparameterization trick. | {
"cite_N": [
"@cite_1"
],
"mid": [
"2395531316"
],
"abstract": [
"Color enhancement is a very important aspect of photo editing. Even when photographers have tens of or hundreds of photographs, they must enhance each photo one by one by manually tweaking sliders in software such as brightness and contrast, because automatic color enhancement is not always satisfactory for them. To support this repetitive manual task, we present self-reinforcing color enhancement, where the system implicitly and progressively learns the user's preferences by training on their photo editing history. The more photos the user enhances, the more effectively the system supports the user. We present a working prototype system called SelPh, and then describe the algorithms used to perform the self-reinforcement. We conduct a user study to investigate how photographers would use a self-reinforcing system to enhance a collection of photos. The results indicate that the participants were satisfied with the proposed system and strongly agreed that the self-reinforcing approach is preferable to the traditional workflow."
]
} |
1704.04565 | 2953025129 | We present a solution to the problem of paraphrase identification of questions. We focus on a recent dataset of question pairs annotated with binary paraphrase labels and show that a variant of the decomposable attention model (, 2016) results in accurate performance on this task, while being far simpler than many competing neural architectures. Furthermore, when the model is pretrained on a noisy dataset of automatically collected question paraphrases, it obtains the best reported performance on the dataset. | Character-level modeling of text is a popular approach. While conceptually simple, character @math -gram embeddings are a highly competitive representation @cite_11 @cite_17 @cite_6 . More complex representations built directly from individual characters have also been proposed @cite_12 @cite_16 @cite_9 @cite_7 @cite_18 . These representations are robust to out-of-vocabulary items, often producing improved results. Our pretraining procedure is reminiscent of several recent papers [] wieting-EtAl:2016:EMNLP2016 who aim for general purpose character @math -gram embeddings. In contrast, we pretrain all model parameters on automatic but in-domain paraphrase data. We employ the same neural architecture as our end task, similar to prior work on multi-task learning [] sogaard-goldberg:2016:P16-2 , but use a simpler learning setup. | {
"cite_N": [
"@cite_18",
"@cite_11",
"@cite_7",
"@cite_9",
"@cite_6",
"@cite_16",
"@cite_12",
"@cite_17"
],
"mid": [
"2949563612",
"2136189984",
"2311921240",
"1938755728",
"",
"2339995566",
"1816313093",
"2463895987"
],
"abstract": [
"We introduce a model for constructing vector representations of words by composing characters using bidirectional LSTMs. Relative to traditional word representation models that have independent vectors for each word type, our model requires only a single vector per character type and a fixed set of parameters for the compositional model. Despite the compactness of this model and, more importantly, the arbitrary nature of the form-function relationship in language, our \"composed\" word representations yield state-of-the-art results in language modeling and part-of-speech tagging. Benefits over traditional baselines are particularly pronounced in morphologically rich languages (e.g., Turkish).",
"Latent semantic models, such as LSA, intend to map a query to its relevant documents at the semantic level where keyword-based matching often fails. In this study we strive to develop a series of new latent semantic models with a deep structure that project queries and documents into a common low-dimensional space where the relevance of a document given a query is readily computed as the distance between them. The proposed deep structured semantic models are discriminatively trained by maximizing the conditional likelihood of the clicked documents given a query using the clickthrough data. To make our models applicable to large-scale Web search applications, we also use a technique called word hashing, which is shown to effectively scale up our semantic models to handle large vocabularies which are common in such tasks. The new models are evaluated on a Web document ranking task using a real-world data set. Results show that our best model significantly outperforms other latent semantic models, which were considered state-of-the-art in the performance prior to the work presented in this paper.",
"The existing machine translation systems, whether phrase-based or neural, have relied almost exclusively on word-level modelling with explicit segmentation. In this paper, we ask a fundamental question: can neural machine translation generate a character sequence without any explicit segmentation? To answer this question, we evaluate an attention-based encoder-decoder with a subword-level encoder and a character-level decoder on four language pairs--En-Cs, En-De, En-Ru and En-Fi-- using the parallel corpora from WMT'15. Our experiments show that the models with a character-level decoder outperform the ones with a subword-level decoder on all of the four language pairs. Furthermore, the ensembles of neural models with a character-level decoder outperform the state-of-the-art non-neural machine translation systems on En-Cs, En-De and En-Fi and perform comparably on En-Ru.",
"We describe a simple neural language model that relies only on character-level inputs. Predictions are still made at the word-level. Our model employs a convolutional neural network (CNN) and a highway network over characters, whose output is given to a long short-term memory (LSTM) recurrent neural network language model (RNN-LM). On the English Penn Treebank the model is on par with the existing state-of-the-art despite having 60 fewer parameters. On languages with rich morphology (Arabic, Czech, French, German, Spanish, Russian), the model outperforms word-level morpheme-level LSTM baselines, again with fewer parameters. The results suggest that on many languages, character inputs are sufficient for language modeling. Analysis of word representations obtained from the character composition part of the model reveals that the model is able to encode, from characters only, both semantic and orthographic information.",
"",
"Nearly all previous work on neural machine translation (NMT) has used quite restricted vocabularies, perhaps with a subsequent method to patch in unknown words. This paper presents a novel word-character solution to achieving open vocabulary NMT. We build hybrid systems that translate mostly at the word level and consult the character components for rare words. Our character-level recurrent neural networks compute source word representations and recover unknown target words when needed. The twofold advantage of such a hybrid approach is that it is much faster and easier to train than character-based ones; at the same time, it never produces unknown words as in the case of word-based models. On the WMT'15 English to Czech translation task, this hybrid approach offers an addition boost of +2.1-11.4 BLEU points over models that already handle unknown words. Our best system achieves a new state-of-the-art result with 20.7 BLEU score. We demonstrate that our character models can successfully learn to not only generate well-formed words for Czech, a highly-inflected language with a very complex vocabulary, but also build correct representations for English source words.",
"Neural machine translation (NMT) models typically operate with a fixed vocabulary, but translation is an open-vocabulary problem. Previous work addresses the translation of out-of-vocabulary words by backing off to a dictionary. In this paper, we introduce a simpler and more effective approach, making the NMT model capable of open-vocabulary translation by encoding rare and unknown words as sequences of subword units. This is based on the intuition that various word classes are translatable via smaller units than words, for instance names (via character copying or transliteration), compounds (via compositional translation), and cognates and loanwords (via phonological and morphological transformations). We discuss the suitability of different word segmentation techniques, including simple character n-gram models and a segmentation based on the byte pair encoding compression algorithm, and empirically show that subword models improve over a back-off dictionary baseline for the WMT 15 translation tasks English-German and English-Russian by 1.1 and 1.3 BLEU, respectively.",
""
]
} |
1704.04856 | 2612963851 | We propose a model to automatically describe changes introduced in the source code of a program using natural language. Our method receives as input a set of code commits, which contains both the modifications and message introduced by an user. These two modalities are used to train an encoder-decoder architecture. We evaluated our approach on twelve real world open source projects from four different programming languages. Quantitative and qualitative results showed that the proposed approach can generate feasible and semantically sound descriptions not only in standard in-project settings, but also in a cross-project setting. | The emergence of unifying paradigms that explicitly relate programming and natural languages in distributional terms @cite_12 and the availability of large corpus mainly from open source software opened the door for the use of language modeling for several tasks @cite_8 . Examples of this are approaches for learning program representations @cite_14 , bug localization @cite_10 , API suggestion @cite_13 and code completion @cite_20 . | {
"cite_N": [
"@cite_14",
"@cite_8",
"@cite_10",
"@cite_20",
"@cite_13",
"@cite_12"
],
"mid": [
"2282866165",
"2156981320",
"",
"2143861926",
"",
"2142403498"
],
"abstract": [
"Programming language processing (similar to natural language processing) is a hot research topic in the field of software engineering; it has also aroused growing interest in the artificial intelligence community. However, different from a natural language sentence, a program contains rich, explicit, and complicated structural information. Hence, traditional NLP models may be inappropriate for programs. In this paper, we propose a novel tree-based convolutional neural network (TBCNN) for programming language processing, in which a convolution kernel is designed over programs' abstract syntax trees to capture structural information. TBCNN is a generic architecture for programming language processing; our experiments show its effectiveness in two different program analysis tasks: classifying programs according to functionality, and detecting code snippets of certain patterns. TBCNN outperforms baseline methods, including several neural models for NLP.",
"We present a new approach for predicting program properties from massive codebases (aka \"Big Code\"). Our approach first learns a probabilistic model from existing data and then uses this model to predict properties of new, unseen programs. The key idea of our work is to transform the input program into a representation which allows us to phrase the problem of inferring program properties as structured prediction in machine learning. This formulation enables us to leverage powerful probabilistic graphical models such as conditional random fields (CRFs) in order to perform joint prediction of program properties. As an example of our approach, we built a scalable prediction engine called JSNice for solving two kinds of problems in the context of JavaScript: predicting (syntactic) names of identifiers and predicting (semantic) type annotations of variables. Experimentally, JSNice predicts correct names for 63 of name identifiers and its type annotation predictions are correct in 81 of the cases. In the first week since its release, JSNice was used by more than 30,000 developers and in only few months has become a popular tool in the JavaScript developer community. By formulating the problem of inferring program properties as structured prediction and showing how to perform both learning and inference in this context, our work opens up new possibilities for attacking a wide range of difficult problems in the context of \"Big Code\" including invariant generation, decompilation, synthesis and others.",
"",
"We address the problem of synthesizing code completions for programs using APIs. Given a program with holes, we synthesize completions for holes with the most likely sequences of method calls. Our main idea is to reduce the problem of code completion to a natural-language processing problem of predicting probabilities of sentences. We design a simple and scalable static analysis that extracts sequences of method calls from a large codebase, and index these into a statistical language model. We then employ the language model to find the highest ranked sentences, and use them to synthesize a code completion. Our approach is able to synthesize sequences of calls across multiple objects together with their arguments. Experiments show that our approach is fast and effective. Virtually all computed completions typecheck, and the desired completion appears in the top 3 results in 90 of the cases.",
"",
"Natural languages like English are rich, complex, and powerful. The highly creative and graceful use of languages like English and Tamil, by masters like Shakespeare and Avvaiyar, can certainly delight and inspire. But in practice, given cognitive constraints and the exigencies of daily life, most human utterances are far simpler and much more repetitive and predictable. In fact, these utterances can be very usefully modeled using modern statistical methods. This fact has led to the phenomenal success of statistical approaches to speech recognition, natural language translation, question-answering, and text mining and comprehension. We begin with the conjecture that most software is also natural, in the sense that it is created by humans at work, with all the attendant constraints and limitations — and thus, like natural language, it is also likely to be repetitive and predictable. We then proceed to ask whether a) code can be usefully modeled by statistical language models and b) such models can be leveraged to support software engineers. Using the widely adopted n-gram model, we provide empirical evidence supportive of a positive answer to both these questions. We show that code is also very repetitive, and in fact even more so than natural languages. As an example use of the model, we have developed a simple code completion engine for Java that, despite its simplicity, already improves Eclipse's built-in completion capability. We conclude the paper by laying out a vision for future research in this area."
]
} |
1704.04856 | 2612963851 | We propose a model to automatically describe changes introduced in the source code of a program using natural language. Our method receives as input a set of code commits, which contains both the modifications and message introduced by an user. These two modalities are used to train an encoder-decoder architecture. We evaluated our approach on twelve real world open source projects from four different programming languages. Quantitative and qualitative results showed that the proposed approach can generate feasible and semantically sound descriptions not only in standard in-project settings, but also in a cross-project setting. | Source code summarization has received special attention, ranging from the use of information retrieval techniques to the addition of physiological features such as eye tracking @cite_23 . In recent years several representation learning approaches have been proposed, such as @cite_18 , where the authors employ a convolutional architecture embedded inside an attention mechanism to learn an efficient mapping between source code tokens and natural language keywords. | {
"cite_N": [
"@cite_18",
"@cite_23"
],
"mid": [
"2257123346",
"2019246026"
],
"abstract": [
"Attention mechanisms in neural networks have proved useful for problems in which the input and output do not have fixed dimension. Often there exist features that are locally translation invariant and would be valuable for directing the model's attention, but previous attentional architectures are not constructed to learn such features specifically. We introduce an attentional neural network that employs convolution on the input tokens to detect local time-invariant and long-range topical attention features in a context-dependent way. We apply this architecture to the problem of extreme summarization of source code snippets into short, descriptive function name-like summaries. Using those features, the model sequentially generates a summary by marginalizing over two attention mechanisms: one that predicts the next summary token based on the attention weights of the input tokens and another that is able to copy a code token as-is directly into the summary. We demonstrate our convolutional attention neural network's performance on 10 popular Java projects showing that it achieves better performance compared to previous attentional mechanisms.",
"Although version control systems allow developers to describe and explain the rationale behind code changes in commit messages, the state of practice indicates that most of the time such commit messages are either very short or even empty. In fact, in a recent study of 23K+ Java projects it has been found that only 10 of the messages are descriptive and over 66 of those messages contained fewer words as compared to a typical English sentence (i.e., 15-20 words). However, accurate and complete commit messages summarizing software changes are important to support a number of development and maintenance tasks. In this paper we present an approach, coined as Change Scribe, which is designed to generate commit messages automatically from change sets. Change Scribe generates natural language commit messages by taking into account commit stereotype, the type of changes (e.g., files rename, changes done only to property files), as well as the impact set of the underlying changes. We evaluated Change Scribe in a survey involving 23 developers in which the participants analyzed automatically generated commit messages from real changes and compared them with commit messages written by the original developers of six open source systems. The results demonstrate that automatically generated messages by Change Scribe are preferred in about 62 of the cases for large commits, and about 54 for small commits."
]
} |
1704.04856 | 2612963851 | We propose a model to automatically describe changes introduced in the source code of a program using natural language. Our method receives as input a set of code commits, which contains both the modifications and message introduced by an user. These two modalities are used to train an encoder-decoder architecture. We evaluated our approach on twelve real world open source projects from four different programming languages. Quantitative and qualitative results showed that the proposed approach can generate feasible and semantically sound descriptions not only in standard in-project settings, but also in a cross-project setting. | In terms of specifically working on code change summarization, propose a method based on a set of rules that considers the type and impact of the changes, and @cite_7 combines summarization with symbolic execution. To the best of our knowledge, our approach represents the first attempt to generate natural language descriptions from code changes without the use of hand-crafted features, a desirable setting given the heterogeneity of the data involved. | {
"cite_N": [
"@cite_7"
],
"mid": [
"2057049321"
],
"abstract": [
"Source code modifications are often documented with log messages. Such messages are a key component of software maintenance: they can help developers validate changes, locate and triage defects, and understand modifications. However, this documentation can be burdensome to create and can be incomplete or inaccurate. We present an automatic technique for synthesizing succinct human-readable documentation for arbitrary program differences. Our algorithm is based on a combination of symbolic execution and a novel approach to code summarization. The documentation it produces describes the effect of a change on the runtime behavior of a program, including the conditions under which program behavior changes and what the new behavior is. We compare our documentation to 250 human-written log messages from 5 popular open source projects. Employing a human study, we find that our generated documentation is suitable for supplementing or replacing 89 of existing log messages that directly describe a code change."
]
} |
1704.04852 | 2950302274 | We describe a method for formation-change trajectory planning for large quadrotor teams in obstacle-rich environments. Our method decomposes the planning problem into two stages: a discrete planner operating on a graph representation of the workspace, and a continuous refinement that converts the non-smooth graph plan into a set of C^k-continuous trajectories, locally optimizing an integral-squared-derivative cost. We account for the downwash effect, allowing safe flight in dense formations. We demonstrate the computational efficiency in simulation with up to 200 robots and the physical plausibility with an experiment with 32 nano-quadrotors. Our approach can compute safe and smooth trajectories for hundreds of quadrotors in dense environments with obstacles in a few minutes. | Many works have approached the problem from a graph search perspective @cite_12 @cite_6 . These methods are adept at dealing with maze-like environments and scenarios with high congestion. Some represent the search graph implicitly @cite_14 , so they are not always restricted to a predefined set of points in configuration space. However, directly interpreting a graph plan as a trajectory results in a piecewise linear path, requiring the robot to fully stop at each graph vertex to maintain dynamic feasibility. It is possible to use these planners to resolve ordering conflicts and refine the output for execution on robots @cite_23 . | {
"cite_N": [
"@cite_14",
"@cite_23",
"@cite_6",
"@cite_12"
],
"mid": [
"2136323427",
"2566753361",
"",
"2124015815"
],
"abstract": [
"We present a sampling-based framework for multi-robot motion planning which combines an implicit representation of a roadmap with a novel approach for pathfinding in geometrically embedded graphs tailored for our setting. Our pathfinding algorithm, discrete-RRT (dRRT), is an adaptation of the celebrated RRT algorithm for the discrete case of a graph, and it enables a rapid exploration of the high-dimensional configuration space by carefully walking through an implicit representation of a tensor product of roadmaps for the individual robots. We demonstrate our approach experimentally on scenarios of up to 60 degrees of freedom where our algorithm is faster by a factor of at least ten when compared to existing algorithms that we are aware of.",
"We study formation change for robot groups in known environments. We are given a team of robots partitioned into groups, where robots in the same group are interchangeable with each other. A formation specifies the locations occupied by each group. The objective is to find collision-free paths that move all robots from a given start formation to a given goal formation. Our algorithm TAPF* has the following features: (a) it incorporates kinematic constraints of robots in form of velocity limits; (b) it maintains a user-specified safety distance between robots; (c) it attempts to minimize the makespan; and (d) it runs efficiently for hundreds of robots and dozens of groups even in dense 3D environments with narrow corridors and other occlusions. We demonstrate the efficiency and effectiveness of TAPF* in simulation and on robots.",
"",
"Multirobot path planning is difficult because the full configuration space of the system grows exponentially with the number of robots. Planning in the joint configuration space of a set of robots is only necessary if they are strongly coupled, which is often not true if the robots are well separated in the workspace. Therefore, we initially plan for each robot separately, and only couple sets of robots after they have been found to interact, thus minimizing the dimensionality of the search space. We present a general strategy called subdimensional expansion, which dynamically generates low dimensional search spaces embedded in the full configuration space. We also present an implementation of subdimensional expansion for robot configuration spaces that can be represented as a graph, called M*, and show that M* is complete and finds minimal cost paths."
]
} |
1704.04852 | 2950302274 | We describe a method for formation-change trajectory planning for large quadrotor teams in obstacle-rich environments. Our method decomposes the planning problem into two stages: a discrete planner operating on a graph representation of the workspace, and a continuous refinement that converts the non-smooth graph plan into a set of C^k-continuous trajectories, locally optimizing an integral-squared-derivative cost. We account for the downwash effect, allowing safe flight in dense formations. We demonstrate the computational efficiency in simulation with up to 200 robots and the physical plausibility with an experiment with 32 nano-quadrotors. Our approach can compute safe and smooth trajectories for hundreds of quadrotors in dense environments with obstacles in a few minutes. | Spline-based refinement of waypoint plans was described in @cite_16 and @cite_17 . Our method builds upon these works by adding support for three-dimensional ellipsoidal robots, environmental obstacles, and an anytime refinement stage to further improve the plan after generating an initial set of smooth trajectories. We demonstrate that our iterative refinement produces trajectories with significantly smoother dynamics. | {
"cite_N": [
"@cite_16",
"@cite_17"
],
"mid": [
"2565998605",
"1937654348"
],
"abstract": [
"In this work, we consider the labeled multi-robot planning problem. In this paradigm, a team of robots at fixed start positions must navigate to pre-specified and noninterchangable goal positions. While many algorithms have been proposed for finding optimal solutions to this problem, most methods assume that the robots are kinematic agents, whereas in reality, robots often have high-order dynamics that must be respected by their trajectories. Here, we propose a centralized method for generating trajectories for teams of robots with general nth-order dynamics navigating to labeled goals. Our algorithm is safe and complete and additionally allows for decoupled optimization of each robot's trajectory as a Quadratic Program with linear constraints. We present simulation results for teams of up to 20 robots.",
"The thesis describes a new method for obtaining minimizers for optimal control problems whose minima serve as control policies for guiding nonlinear dynamical systems to achieve prescribed goals under imposed trajectory and actuator constraints. One of the major contributions of the present work resides in the approximation of such minimizers by piecewise polynomial functions expressed in terms of a linear combination of non-uniform rational B-spline (NURBS) basis functions and the judicious exploitation of the properties of the resulting NURBS curves to improve the computational effort often associated with solving optimal control problems for constrained dynamical systems. In particular, by exploiting the two structures combined in a NURBS curve, NURBS basis functions and an associated union of overlapping polytopes constructed from the coefficients of the linear combination, we are able to separate an optimal control problem into two subproblems | guidance and obstacle avoidance, making the original problem tractable. This is accomplished by laying out the union of overlapping polytopes in such a way that they delineate a section of space that avoids all obstacles and then manipulating the NURBS basis functions to obtain trajectories that are guaranteed to remain bounded by this section of space without explicitly including the conjunction of disjunctions naturally induced from obstacles into the guidance problem. In addition, we show how one can construct systematically a feasible region that corresponds to a NURBS parameterization starting from an ordered union of pairwise adjacently overlapping nonempty compact convex sets. Specifically, we show how to setup a nonlinear programming problem to solve for the feasible region in terms of an ordered union of pairwise adjacently overlapping polytopes with nonempty interiors by maximizing the sum of their volumes and starting from a feasible region described by an ordered union of pairwise adjacently overlapping nonempty convex compact simi-algebraic sets. Finally, we show how this strategy can be implemented practically for an autonomous system traversing an urban environment. Finally, this work culminated in the filing of patent US20070179685 on behalf of Northrop Grumman for the Space Technology sector and in the development of the NURBS-based OTG software package. This C++ package contains the theoretical results of this thesis in the form of an object-oriented implementation optimized for real-time trajectory generation."
]
} |
1704.04235 | 2607350342 | Domain adaptation is transfer learning which aims to generalize a learning model across training and testing data with different distributions. Most previous research tackle this problem in seeking a shared feature representation between source and target domains while reducing the mismatch of their data distributions. In this paper, we propose a close yet discriminative domain adaptation method, namely CDDA, which generates a latent feature representation with two interesting properties. First, the discrepancy between the source and target domain, measured in terms of both marginal and conditional probability distribution via Maximum Mean Discrepancy is minimized so as to attract two domains close to each other. More importantly, we also design a repulsive force term, which maximizes the distances between each label dependent sub-domain to all others so as to drag different class dependent sub-domains far away from each other and thereby increase the discriminative power of the adapted domain. Moreover, given the fact that the underlying data manifold could have complex geometric structure, we further propose the constraints of label smoothness and geometric structure consistency for label propagation. Extensive experiments are conducted on 36 cross-domain image classification tasks over four public datasets. The comprehensive results show that the proposed method consistently outperforms the state-of-the-art methods with significant margins. | In machine learning, domain adaptation is transfer learning which aims to learn an effective predictive model for a target domain without labeled data in leveraging abundant existing labeled data of a different but related source domain. Because the collection of large labeled data as needed in traditional machine learning is often prohibitive for many real-life applications, there is an increasing interest on this yet topic @cite_2 @cite_5 . According to the taxonomy made in recent surveys @cite_2 @cite_5 @cite_6 , the proposed method falls down into the feature representation category. | {
"cite_N": [
"@cite_5",
"@cite_6",
"@cite_2"
],
"mid": [
"2395579298",
"2096943734",
"2165698076"
],
"abstract": [
"Machine learning and data mining techniques have been used in numerous real-world applications. An assumption of traditional machine learning methodologies is the training data and testing data are taken from the same domain, such that the input feature space and data distribution characteristics are the same. However, in some real-world machine learning scenarios, this assumption does not hold. There are cases where training data is expensive or difficult to collect. Therefore, there is a need to create high-performance learners trained with more easily obtained data from different domains. This methodology is referred to as transfer learning. This survey paper formally defines transfer learning, presents information on current solutions, and reviews applications applied to transfer learning. Lastly, there is information listed on software downloads for various transfer learning solutions and a discussion of possible future research work. The transfer learning solutions surveyed are independent of data size and can be applied to big data environments.",
"Transfer learning is established as an effective technology in computer vision for leveraging rich labeled data in the source domain to build an accurate classifier for the target domain. However, most prior methods have not simultaneously reduced the difference in both the marginal distribution and conditional distribution between domains. In this paper, we put forward a novel transfer learning approach, referred to as Joint Distribution Adaptation (JDA). Specifically, JDA aims to jointly adapt both the marginal distribution and conditional distribution in a principled dimensionality reduction procedure, and construct new feature representation that is effective and robust for substantial distribution difference. Extensive experiments verify that JDA can significantly outperform several state-of-the-art methods on four types of cross-domain image classification problems.",
"A major assumption in many machine learning and data mining algorithms is that the training and future data must be in the same feature space and have the same distribution. However, in many real-world applications, this assumption may not hold. For example, we sometimes have a classification task in one domain of interest, but we only have sufficient training data in another domain of interest, where the latter data may be in a different feature space or follow a different data distribution. In such cases, knowledge transfer, if done successfully, would greatly improve the performance of learning by avoiding much expensive data-labeling efforts. In recent years, transfer learning has emerged as a new learning framework to address this problem. This survey focuses on categorizing and reviewing the current progress on transfer learning for classification, regression, and clustering problems. In this survey, we discuss the relationship between transfer learning and other related machine learning techniques such as domain adaptation, multitask learning and sample selection bias, as well as covariate shift. We also explore some potential future issues in transfer learning research."
]
} |
1704.04235 | 2607350342 | Domain adaptation is transfer learning which aims to generalize a learning model across training and testing data with different distributions. Most previous research tackle this problem in seeking a shared feature representation between source and target domains while reducing the mismatch of their data distributions. In this paper, we propose a close yet discriminative domain adaptation method, namely CDDA, which generates a latent feature representation with two interesting properties. First, the discrepancy between the source and target domain, measured in terms of both marginal and conditional probability distribution via Maximum Mean Discrepancy is minimized so as to attract two domains close to each other. More importantly, we also design a repulsive force term, which maximizes the distances between each label dependent sub-domain to all others so as to drag different class dependent sub-domains far away from each other and thereby increase the discriminative power of the adapted domain. Moreover, given the fact that the underlying data manifold could have complex geometric structure, we further propose the constraints of label smoothness and geometric structure consistency for label propagation. Extensive experiments are conducted on 36 cross-domain image classification tasks over four public datasets. The comprehensive results show that the proposed method consistently outperforms the state-of-the-art methods with significant margins. | Note that we do not discuss the line of work in the literature on transfer learning which is embedded into deep convolutional neural network as the features used in this work are not deep features; Nevertheless we have noticed their impressive performance, thanks to the combination of the latest advances in transfer learning discussed above with the cutting-edge understanding on the transferability @cite_20 of state-of-the-art deep neural networks, , Deep Adaptation Network(DAN) @cite_3 , Mixing seamlessly our proposed transfer knowledge model with state-of-the-art deep networks will be the subject of our upcoming investigation. | {
"cite_N": [
"@cite_3",
"@cite_20"
],
"mid": [
"2951670162",
"22861983"
],
"abstract": [
"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.",
"The exponential increase in the availability of online reviews and recommendations makes sentiment classification an interesting topic in academic and industrial research. Reviews can span so many different domains that it is difficult to gather annotated training data for all of them. Hence, this paper studies the problem of domain adaptation for sentiment classifiers, hereby a system is trained on labeled reviews from one source domain but is meant to be deployed on another. We propose a deep learning approach which learns to extract a meaningful representation for each review in an unsupervised fashion. Sentiment classifiers trained with this high-level feature representation clearly outperform state-of-the-art methods on a benchmark composed of reviews of 4 types of Amazon products. Furthermore, this method scales well and allowed us to successfully perform domain adaptation on a larger industrial-strength dataset of 22 domains."
]
} |
1704.04460 | 2605927250 | In this work we introduce Qumin, a novel quantum programming language with a focus on providing an easy to use, minimalist, high-level, and easily extensible platform for quantum programming. Qumin's design concentrates on encompassing the various interactions between classical and quantum computation via the use of two sublanguages: an untyped one that handles classical preparation and control, and one linearly typed that explicitly handles quantum routines. This allows the classical part of the language to be freely used for general programming while placing restrictions on the quantum part that enforce rules of quantum computing like the no-cloning of qubits. We describe both the language's theoretical foundations in terms of lambda calculi and linear type systems, and more practical matters such as implementations of algorithms and useful programming tools like matrix and oracle generators that streamline the interaction of the classical and quantum fragments of a program. Finally, we provide an experimental open-source implementation of an interpreter, typechecker and related tools for the language (which can be found in this https URL ). | Works dealing with the design of quantum programming languages like @cite_4 @cite_5 @cite_16 have discussed some basic requirements one would expect a quantum programming language to fulfill. These vary accordingly to the underlying paradigm, with frequent requirements amongst other being: completeness, extensibility, abstracting away and being independent from the underlying machinery and being expressive enough to allow one to defne quantum data structures, oracles, handling of measurement, and handling of quantum memory registers. | {
"cite_N": [
"@cite_5",
"@cite_16",
"@cite_4"
],
"mid": [
"2121369607",
"1999626800",
"1537615043"
],
"abstract": [
"It is becoming increasingly clear that, if a useful device for quantum computation will ever be built, it will be embodied by a classical computing machine with control over a truly quantum subsystem, this apparatus performing a mixture of classical and quantum computation. This paper investigates a possible approach to the problem of programming such machines: a template high level quantum language is presented which complements a generic general purpose classical language with a set of quantum primitives. The underlying scheme involves a run-time environment which calculates the byte-code for the quantum operations and pipes it to a quantum device controller or to a simulator. This language can compactly express existing quantum algorithms and reduce them to sequences of elementary operations; it also easily lends itself to automatic, hardware independent, circuit simplification. A publicly available preliminary implementation of the proposed ideas has been realised using the language.",
"We propose the design of a programming language for quantum computing. Traditionally, quantum algorithms are frequently expressed at the hardware level, for instance in terms of the quantum circuit model or quantum Turing machines. These approaches do not encourage structured programming or abstractions such as data types. In this paper, we describe the syntax and semantics of a simple quantum programming language with high-level features such as loops, recursive procedures, and structured data types. The language is functional in nature, statically typed, free of run-time errors, and has an interesting denotational semantics in terms of complete partial orders of superoperators.",
"A few conventions for thinking about and writing quantum pseudocode are proposed. The conventions can be used for presenting any quantum algorithm down to the lowest level and are consistent with a quantum random access machine (QRAM) model for quantum computing. In principle a formal version of quantum pseudocode could be used in a future extension of a conventional language."
]
} |
1704.04460 | 2605927250 | In this work we introduce Qumin, a novel quantum programming language with a focus on providing an easy to use, minimalist, high-level, and easily extensible platform for quantum programming. Qumin's design concentrates on encompassing the various interactions between classical and quantum computation via the use of two sublanguages: an untyped one that handles classical preparation and control, and one linearly typed that explicitly handles quantum routines. This allows the classical part of the language to be freely used for general programming while placing restrictions on the quantum part that enforce rules of quantum computing like the no-cloning of qubits. We describe both the language's theoretical foundations in terms of lambda calculi and linear type systems, and more practical matters such as implementations of algorithms and useful programming tools like matrix and oracle generators that streamline the interaction of the classical and quantum fragments of a program. Finally, we provide an experimental open-source implementation of an interpreter, typechecker and related tools for the language (which can be found in this https URL ). | QPL, a functional programming language with well-defined operational semantics in terms of complete partial orders and superoperators first appeared in @cite_16 , and was a big influence in the development of other functional quantum programming languages. | {
"cite_N": [
"@cite_16"
],
"mid": [
"1999626800"
],
"abstract": [
"We propose the design of a programming language for quantum computing. Traditionally, quantum algorithms are frequently expressed at the hardware level, for instance in terms of the quantum circuit model or quantum Turing machines. These approaches do not encourage structured programming or abstractions such as data types. In this paper, we describe the syntax and semantics of a simple quantum programming language with high-level features such as loops, recursive procedures, and structured data types. The language is functional in nature, statically typed, free of run-time errors, and has an interesting denotational semantics in terms of complete partial orders of superoperators."
]
} |
1704.04460 | 2605927250 | In this work we introduce Qumin, a novel quantum programming language with a focus on providing an easy to use, minimalist, high-level, and easily extensible platform for quantum programming. Qumin's design concentrates on encompassing the various interactions between classical and quantum computation via the use of two sublanguages: an untyped one that handles classical preparation and control, and one linearly typed that explicitly handles quantum routines. This allows the classical part of the language to be freely used for general programming while placing restrictions on the quantum part that enforce rules of quantum computing like the no-cloning of qubits. We describe both the language's theoretical foundations in terms of lambda calculi and linear type systems, and more practical matters such as implementations of algorithms and useful programming tools like matrix and oracle generators that streamline the interaction of the classical and quantum fragments of a program. Finally, we provide an experimental open-source implementation of an interpreter, typechecker and related tools for the language (which can be found in this https URL ). | QML @cite_0 is a functional quantum programming language, whose semantics are defined in terms of a category of Finite Quantum Computations, and allow QML to be described in terms of superoperators and be translated to quantum circuits. QML's central features are built around the idea of controlling decoherence, a strict linear type system and quantum control in addition to quantum data. Interestingly QML's type system allows for contraction by way of sharing - not cloning qubits. | {
"cite_N": [
"@cite_0"
],
"mid": [
"2157601714"
],
"abstract": [
"We introduce the language QML, a functional language for quantum computations on finite types. Its design is guided by its categorical semantics: QML programs are interpreted by morphisms in the category FQC of finite quantum computations, which provides a constructive semantics of irreversible quantum computations realisable as quantum gates. QML integrates reversible and irreversible quantum computations in one language, using first order strict linear logic to make weakenings explicit. Strict programs are free from decoherence and hence preserve superpositions and entanglement -which is essential for quantum parallelism."
]
} |
1704.04460 | 2605927250 | In this work we introduce Qumin, a novel quantum programming language with a focus on providing an easy to use, minimalist, high-level, and easily extensible platform for quantum programming. Qumin's design concentrates on encompassing the various interactions between classical and quantum computation via the use of two sublanguages: an untyped one that handles classical preparation and control, and one linearly typed that explicitly handles quantum routines. This allows the classical part of the language to be freely used for general programming while placing restrictions on the quantum part that enforce rules of quantum computing like the no-cloning of qubits. We describe both the language's theoretical foundations in terms of lambda calculi and linear type systems, and more practical matters such as implementations of algorithms and useful programming tools like matrix and oracle generators that streamline the interaction of the classical and quantum fragments of a program. Finally, we provide an experimental open-source implementation of an interpreter, typechecker and related tools for the language (which can be found in this https URL ). | There also exist a number of imperative languages. To start with, we have the aforementioned works by Knill which includes an imperative-style pseudocode and with a full scheme and implementation in C++. One of the first attempts at writing a full imperative language, called QCL, is described by " O mer in @cite_17 , with a C-like syntax and high-level quantum programming features such as automatic memory management, user-defined operators, and automatic derivations of them. It also includes a classical sublanguage. Sanders and Zuliani @cite_24 have also proposed an imperative language, qGCL, which is based on a guarded command language and enjoys rigorous semantics and an associated refinement calculus. Another imperative language, LanQ, was introduced by Mlna r ' i k @cite_2 , who also provided operational semantics for it, as well as a type soundness proof for the non-communicating part of the language. | {
"cite_N": [
"@cite_24",
"@cite_2",
"@cite_17"
],
"mid": [
"",
"102856817",
"30747476"
],
"abstract": [
"",
"We present an imperative quantum programming language LanQ which was designed to support combination of quantum and classical programming and basic process operations - process creation and interprocess communication. The language can thus be used for implementing both classical and quantum algorithms and protocols. Its syntax is similar to that of C language what makes it easy to learn for existing programmers. In this paper, we present operational semantics of the language and a proof of type soundness of the noncommunicating part of the language. We provide an example run of a quantum random number generator.",
"Despite many common concepts with classical computer science, quantum computing is still widely considered as a special discipline within the broad field of theoretical physics. One reason for the slow adoption of QC by the computer science community is the confusing variety of formalisms (Dirac notation, matrices, gates, operators, etc.), none of which has any similarity with classical programming languages, as well as the rather “physical” terminology in most of the available literature. QCL (Quantum Computation Language) tries to fill this gap: QCL is a hight level, architecture independent programming language for quantum computers, with a syntax derived from classical procedural languages like C or Pascal. This allows for the complete implementation and simulation of quantum algorithms (including classical components) in one consistent formalism. Chapter 1 is an introduction into the basic concepts of quantum programming, a complete language reference of QCL can be found in chapter 2 and chapter 3 gives some examples including the QCL implementation of Shor’s factorisation algorithm. The sourcecode of the QCL interpreter is available at http: tph.tuwien.ac.at oemer qcl.html."
]
} |
1704.04460 | 2605927250 | In this work we introduce Qumin, a novel quantum programming language with a focus on providing an easy to use, minimalist, high-level, and easily extensible platform for quantum programming. Qumin's design concentrates on encompassing the various interactions between classical and quantum computation via the use of two sublanguages: an untyped one that handles classical preparation and control, and one linearly typed that explicitly handles quantum routines. This allows the classical part of the language to be freely used for general programming while placing restrictions on the quantum part that enforce rules of quantum computing like the no-cloning of qubits. We describe both the language's theoretical foundations in terms of lambda calculi and linear type systems, and more practical matters such as implementations of algorithms and useful programming tools like matrix and oracle generators that streamline the interaction of the classical and quantum fragments of a program. Finally, we provide an experimental open-source implementation of an interpreter, typechecker and related tools for the language (which can be found in this https URL ). | A more recent example of a domain-specific language can be found in @cite_12 . LIQUi @math 's language design focuses on a statically type-checked functional language with an isolated physical model, eschewing the use of linear types to allow instead for the manipulation of qubits in mutable ways not allowed by linear types. | {
"cite_N": [
"@cite_12"
],
"mid": [
"1614518745"
],
"abstract": [
"Languages, compilers, and computer-aided design tools will be essential for scalable quantum computing, which promises an exponential leap in our ability to execute complex tasks. LIQUi|> is a modular software architecture designed to control quantum hardware. It enables easy programming, compilation, and simulation of quantum algorithms and circuits, and is independent of a specific quantum architecture. LIQUi|> contains an embedded, domain-specific language designed for programming quantum algorithms, with F# as the host language. It also allows the extraction of a circuit data structure that can be used for optimization, rendering, or translation. The circuit can also be exported to external hardware and software environments. Two different simulation environments are available to the user which allow a trade-off between number of qubits and class of operations. LIQUi|> has been implemented on a wide range of runtimes as back-ends with a single user front-end. We describe the significant components of the design architecture and how to express any given quantum algorithm."
]
} |
1704.04441 | 2606308499 | This paper investigates the robustness of NLP against perturbed word forms. While neural approaches can achieve (almost) human-like accuracy for certain tasks and conditions, they often are sensitive to small changes in the input such as non-canonical input (e.g., typos). Yet both stability and robustness are desired properties in applications involving user-generated content, and the more as humans easily cope with such noisy or adversary conditions. In this paper, we study the impact of noisy input. We consider different noise distributions (one type of noise, combination of noise types) and mismatched noise distributions for training and testing. Moreover, we empirically evaluate the robustness of different models (convolutional neural networks, recurrent neural networks, non-neural models), different basic units (characters, byte pair encoding units), and different NLP tasks (morphological tagging, machine translation). | A large body of work on regularization techniques to learn robust representations and models exists. Examples include @math -regularization, dropout @cite_4 , Jacobian-based sensitivity penalty @cite_2 @cite_7 , and data noising. Compared to other application domains such as vision @cite_18 @cite_21 and speech @cite_5 @cite_14 @cite_27 @cite_9 , working on noisy data @cite_16 @cite_26 @cite_24 and in particular data noising @cite_22 , do not have a long and extensive history in NLP. | {
"cite_N": [
"@cite_18",
"@cite_14",
"@cite_4",
"@cite_26",
"@cite_7",
"@cite_22",
"@cite_9",
"@cite_21",
"@cite_24",
"@cite_27",
"@cite_2",
"@cite_5",
"@cite_16"
],
"mid": [
"2310919327",
"2187428966",
"1904365287",
"2138738738",
"2963696733",
"2740839465",
"2564008415",
"1945616565",
"",
"1979651826",
"",
"2136439176",
"2157765050"
],
"abstract": [
"",
"This paper presents the progress of acoustic models for lowresourced languages (Assamese, Bengali, Haitian Creole, Lao, Zulu) developed within the second evaluation campaign of the IARPA Babel project. This year, the main focus of the project is put on training high-performing automatic speech recognition (ASR) and keyword search (KWS) systems from language resources limited to about 10 hours of transcribed speech data. Optimizing the structure of Multilayer Perceptron (MLP) based feature extraction and switching from the sigmoid activation function to rectified linear units results in about 5 relative improvement over baseline MLP features. Further improvements are obtained when the MLPs are trained on multiple feature streams and by exploiting label preserving data augmentation techniques like vocal tract length perturbation. Systematic application of these methods allows to improve the unilingual systems by 4-6 absolute in WER and 0.064-0.105 absolute in MTWV. Transfer and adaptation of multilingually trained MLPs lead to additional gains, clearly exceeding the project goal of 0.3 MTWV even when only the limited language pack of the target language is used. Index Terms: ASR, KWS, MTWV, MLP, rectified linear units, multilingual, low-resource",
"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.",
"Part-of-speech information is a pre-requisite in many NLP algorithms. However, Twitter text is difficult to part-of-speech tag: it is noisy, with linguistic errors and idiosyncratic style. We present a detailed error analysis of existing taggers, motivating a series of tagger augmentations which are demonstrated to improve performance. We identify and evaluate techniques for improving English part-of-speech tagging performance in this genre. Further, we present a novel approach to system combination for the case where available taggers use different tagsets, based on voteconstrained bootstrapping with unlabeled data. Coupled with assigning prior probabilities to some tokens and handling of unknown words and slang, we reach 88.7 tagging accuracy (90.5 on development data). This is a new high in PTB-compatible tweet part-of-speech tagging, reducing token error by 26.8 and sentence error by 12.2 . The model, training data and tools are made available.",
"",
"",
"Speech recognition performance using deep neural network based acoustic models is known to degrade when the acoustic environment and the speaker population in the target utterances are significantly different from the conditions represented in the training data. To address these mismatched scenarios, multi-style training (MTR) has been used to perturb utterances in an existing uncorrupted and potentially mismatched training speech corpus to better match target domain utterances. This paper addresses the problem of determining the distribution of perturbation levels for a given set of perturbation types that best matches the target speech utterances. An approach is presented that, given a small set of utterances from a target domain, automatically identifies an empirical distribution of perturbation levels that can be applied to utterances in an existing training set. Distributions are estimated for perturbation types that include acoustic background environments, reverberant room configurations, and speaker related variation like frequency and temporal warping. The end goal is for the resulting perturbed training set to characterize the variability in the target domain and thereby optimize ASR performance. An experimental study is performed to evaluate the impact of this approach on ASR performance when the target utterances are taken from a simulated far-field acoustic environment.",
"Several machine learning models, including neural networks, consistently misclassify adversarial examples---inputs formed by applying small but intentionally worst-case perturbations to examples from the dataset, such that the perturbed input results in the model outputting an incorrect answer with high confidence. Early attempts at explaining this phenomenon focused on nonlinearity and overfitting. We argue instead that the primary cause of neural networks' vulnerability to adversarial perturbation is their linear nature. This explanation is supported by new quantitative results while giving the first explanation of the most intriguing fact about them: their generalization across architectures and training sets. Moreover, this view yields a simple and fast method of generating adversarial examples. Using this approach to provide examples for adversarial training, we reduce the test set error of a maxout network on the MNIST dataset.",
"",
"This paper investigates data augmentation for deep neural network acoustic modeling based on label-preserving transformations to deal with data sparsity. Two data augmentation approaches, vocal tract length perturbation (VTLP) and stochastic feature mapping (SFM), are investigated for both deep neural networks (DNNs) and convolutional neural networks (CNNs). The approaches are focused on increasing speaker and speech variations of the limited training data such that the acoustic models trained with the augmented data are more robust to such variations. In addition, a two-stage data augmentation scheme based on a stacked architecture is proposed to combine VTLP and SFM as complementary approaches. Experiments are conducted on Assamese and Haitian Creole, two development languages of the IARPA Babel program, and improved performance on automatic speech recognition (ASR) and keyword search (KWS) is reported.",
"",
"A new training procedure called multi-style training has been developed to improve performance when a recognizer is used under stress or in high noise but cannot be trained in these conditions. Instead of speaking normally during training, talkers use different, easily produced, talking styles. This technique was tested using a speech data base that included stress speech produced during a workload task and when intense noise was presented through earphones. A continuous-distribution talker-dependent Hidden Markov Model (HMM) recognizer was trained both normally (5 normally spoken tokens) and with multi-style training (one token each from normal, fast, clear, loud, and question-pitch talking styles). The average error rate under stress and normal conditions fell by more than a factor of two with multi-style training and the average error rate under conditions sampled during training fell by a factor of four.",
"We address the problem of part-of-speech tagging for English data from the popular micro-blogging service Twitter. We develop a tagset, annotate data, develop features, and report tagging results nearing 90 accuracy. The data and tools have been made available to the research community with the goal of enabling richer text analysis of Twitter and related social media data sets."
]
} |
1704.04441 | 2606308499 | This paper investigates the robustness of NLP against perturbed word forms. While neural approaches can achieve (almost) human-like accuracy for certain tasks and conditions, they often are sensitive to small changes in the input such as non-canonical input (e.g., typos). Yet both stability and robustness are desired properties in applications involving user-generated content, and the more as humans easily cope with such noisy or adversary conditions. In this paper, we study the impact of noisy input. We consider different noise distributions (one type of noise, combination of noise types) and mismatched noise distributions for training and testing. Moreover, we empirically evaluate the robustness of different models (convolutional neural networks, recurrent neural networks, non-neural models), different basic units (characters, byte pair encoding units), and different NLP tasks (morphological tagging, machine translation). | The arbitrary relationship between the orthography of a word and its meaning in general is a well known assumption in linguistics @cite_6 . However, the word form often carries additional important information. This is, for example, the case in morphologically rich languages or in non-normalized text where small perturbations result in similar word forms. Recently, sub-word units have attracted some attention in NLP to handle rarely and unseen words and to reduce the computational complexity in neural network approaches @cite_28 @cite_8 @cite_13 @cite_15 @cite_3 . Examples for sub-word units include BPE based units @cite_13 , characters @cite_28 @cite_15 @cite_3 or even bytes @cite_8 . A comparison of BPE and characters for machine translation regarding grammaticality can be found in @cite_0 . | {
"cite_N": [
"@cite_8",
"@cite_28",
"@cite_6",
"@cite_3",
"@cite_0",
"@cite_15",
"@cite_13"
],
"mid": [
"2185720331",
"2949563612",
"",
"2740646860",
"2220350356",
"2311921240",
"1816313093"
],
"abstract": [
"We describe an LSTM-based model which we call Byte-to-Span (BTS) that reads text as bytes and outputs span annotations of the form [start, length, label] where start positions, lengths, and labels are separate entries in our vocabulary. Because we operate directly on unicode bytes rather than language-specific words or characters, we can analyze text in many languages with a single model. Due to the small vocabulary size, these multilingual models are very compact, but produce results similar to or better than the state-of- the-art in Part-of-Speech tagging and Named Entity Recognition that use only the provided training datasets (no external data sources). Our models are learning \"from scratch\" in that they do not rely on any elements of the standard pipeline in Natural Language Processing (including tokenization), and thus can run in standalone fashion on raw text.",
"We introduce a model for constructing vector representations of words by composing characters using bidirectional LSTMs. Relative to traditional word representation models that have independent vectors for each word type, our model requires only a single vector per character type and a fixed set of parameters for the compositional model. Despite the compactness of this model and, more importantly, the arbitrary nature of the form-function relationship in language, our \"composed\" word representations yield state-of-the-art results in language modeling and part-of-speech tagging. Benefits over traditional baselines are particularly pronounced in morphologically rich languages (e.g., Turkish).",
"",
"",
"We introduce a neural machine translation model that views the input and output sentences as sequences of characters rather than words. Since word-level information provides a crucial source of bias, our input model composes representations of character sequences into representations of words (as determined by whitespace boundaries), and then these are translated using a joint attention translation model. In the target language, the translation is modeled as a sequence of word vectors, but each word is generated one character at a time, conditional on the previous character generations in each word. As the representation and generation of words is performed at the character level, our model is capable of interpreting and generating unseen word forms. A secondary benefit of this approach is that it alleviates much of the challenges associated with preprocessing tokenization of the source and target languages. We show that our model can achieve translation results that are on par with conventional word-based models.",
"The existing machine translation systems, whether phrase-based or neural, have relied almost exclusively on word-level modelling with explicit segmentation. In this paper, we ask a fundamental question: can neural machine translation generate a character sequence without any explicit segmentation? To answer this question, we evaluate an attention-based encoder-decoder with a subword-level encoder and a character-level decoder on four language pairs--En-Cs, En-De, En-Ru and En-Fi-- using the parallel corpora from WMT'15. Our experiments show that the models with a character-level decoder outperform the ones with a subword-level decoder on all of the four language pairs. Furthermore, the ensembles of neural models with a character-level decoder outperform the state-of-the-art non-neural machine translation systems on En-Cs, En-De and En-Fi and perform comparably on En-Ru.",
"Neural machine translation (NMT) models typically operate with a fixed vocabulary, but translation is an open-vocabulary problem. Previous work addresses the translation of out-of-vocabulary words by backing off to a dictionary. In this paper, we introduce a simpler and more effective approach, making the NMT model capable of open-vocabulary translation by encoding rare and unknown words as sequences of subword units. This is based on the intuition that various word classes are translatable via smaller units than words, for instance names (via character copying or transliteration), compounds (via compositional translation), and cognates and loanwords (via phonological and morphological transformations). We discuss the suitability of different word segmentation techniques, including simple character n-gram models and a segmentation based on the byte pair encoding compression algorithm, and empirically show that subword models improve over a back-off dictionary baseline for the WMT 15 translation tasks English-German and English-Russian by 1.1 and 1.3 BLEU, respectively."
]
} |
1704.04095 | 2607103158 | In this study we determined neural network weights and biases by Imperialist Competitive Algorithm (ICA) in order to train network for predicting earthquake intensity in Richter. For this reason, we used dependent parameters like earthquake occurrence time, epicenter's latitude and longitude in degree, focal depth in kilometer, and the seismological center distance from epicenter and earthquake focal center in kilometer which has been provided by Berkeley data base. The studied neural network has two hidden layer: its first layer has 16 neurons and the second layer has 24 neurons. By using ICA algorithm, average error for testing data is 0.0007 with a variance equal to 0.318. The earthquake prediction error in Richter by MSE criteria for ICA algorithm is 0.101, but by using GA, the MSE value is 0.115. | In @cite_10 , for detecting and forecasting natural disasters like earthquake, some data mining techniques have been used. By analyzing data mining approaches like logical models, neural networks, Bayesian networks, and decision tree, it has been illustrated that all of these methods can be used for prognosticating earthquake, tsunami, landslides, and other microseisms. | {
"cite_N": [
"@cite_10"
],
"mid": [
"2566496798"
],
"abstract": [
"A natural disaster is the effect of a natural hazard (e.g., flood, tornado, hurricane, volcanic eruption, earthquake, heatwave, or landslide). Earthquakes, landslides, tsunamis and volcanos are complex physical phenomenon that leads to financial, environmental or human losses. Prediction of such geological disasters is the need of the day. Also, prediction of these disasters is a complex process that depends on many physical and environmental parameters. Many approaches exist in the literature based on scientific and statistical analysis. Data mining techniques can also be used for prediction of these natural hazards. This paper presents a review of application of data mining in the prediction of natural geological calamities. 16 journal articles on the subject published between 1989 and 2011 was analyzed. The main data mining techniques used for earthquake prediction are logistic models, neural networks, the Bayesian belief network, and decision trees, all of which provide primary solutions to the problems inherent in the prediction of earthquakes, tsunamis, landslides and other micro seismic activities. This paper also aims to encourage additional research on topics, and concludes with several suggestions for further research."
]
} |
1704.04095 | 2607103158 | In this study we determined neural network weights and biases by Imperialist Competitive Algorithm (ICA) in order to train network for predicting earthquake intensity in Richter. For this reason, we used dependent parameters like earthquake occurrence time, epicenter's latitude and longitude in degree, focal depth in kilometer, and the seismological center distance from epicenter and earthquake focal center in kilometer which has been provided by Berkeley data base. The studied neural network has two hidden layer: its first layer has 16 neurons and the second layer has 24 neurons. By using ICA algorithm, average error for testing data is 0.0007 with a variance equal to 0.318. The earthquake prediction error in Richter by MSE criteria for ICA algorithm is 0.101, but by using GA, the MSE value is 0.115. | Negarestani used neural networks for measuring Radon density in soil by local parameters @cite_3 . Their analysis illustrates that by neural network approach unnatural phenomenon on earth like earthquake can be distinguished by soil's Radon density. | {
"cite_N": [
"@cite_3"
],
"mid": [
"2114943843"
],
"abstract": [
"A layered neural network (LNN) has been employed to estimate the radon concentration in soil related to the environmental parameters. This technique can find any functional relationship between the radon concentration and the environmental parameters. Analysis of the data obtained from a site in Thailand indicates that this approach is able to differentiate time variation of radon concentration caused by environmental parameters from those arising by anomaly phenomena in the earth (e.g. earthquake). This method is compared with a linear computational technique based on impulse responses from multivariable time series. It is indicated that the proposed method can give a better estimation of radon variations related to environmental parameters that may have a non-linear effect on the radon concentration in soil, such as rainfall."
]
} |
1704.04095 | 2607103158 | In this study we determined neural network weights and biases by Imperialist Competitive Algorithm (ICA) in order to train network for predicting earthquake intensity in Richter. For this reason, we used dependent parameters like earthquake occurrence time, epicenter's latitude and longitude in degree, focal depth in kilometer, and the seismological center distance from epicenter and earthquake focal center in kilometer which has been provided by Berkeley data base. The studied neural network has two hidden layer: its first layer has 16 neurons and the second layer has 24 neurons. By using ICA algorithm, average error for testing data is 0.0007 with a variance equal to 0.318. The earthquake prediction error in Richter by MSE criteria for ICA algorithm is 0.101, but by using GA, the MSE value is 0.115. | In another work Hanna used a General Regression Neural Network (GRNN) for estimating soil slide potential @cite_4 . Their data for training network belonged to two disastrous earthquakes in Turkey and Taiwan in 1999. In their research, 620 data set including twelve soil and earthquake parameters have been introduced. For enhancing accuracy, an iterative process has been applied. | {
"cite_N": [
"@cite_4"
],
"mid": [
"2114938644"
],
"abstract": [
"Abstract Simplified methods have been practiced by researchers to assess nonlinear liquefaction potential of soil. Derived from several field and laboratory tests, various simplified procedures such as stress-based, strain-based, Chinese criteria, etc. have been developed by utilizing case studies and undisturbed soil specimens. In order to address the collective knowledge built up in conventional liquefaction engineering, an alternative general regression neural network model is proposed in this paper. To meet this objective, a total of 620 sets of data including 12 soil and seismic parameters are introduced into the model. The data includes the results of field tests from the two major earthquakes that took place in Turkey and Taiwan in 1999 and some of the desired input parameters are obtained from correlations existing in the literature. The proposed GRNN model was developed in four phases, mainly: identification phase, collection phase, implementation phase, and verification phase. An iterative procedure was followed to maximize the accuracy of the proposed model. The case records were divided randomly into testing, training, and validation datasets. Generating a model that takes into account of 12 soil and seismic parameters is not feasible by using simplified techniques; however, the proposed GRNN model effectively explored the complex relationship between the introduced soil and seismic input parameters and validated the liquefaction decision obtained by simplified methods. The proposed GRNN model predicted well the occurrence nonoccurrence of soil liquefaction in these sites. The model provides a viable tool to geotechnical engineers in assessing seismic condition in sites susceptible to liquefaction."
]
} |
1704.04095 | 2607103158 | In this study we determined neural network weights and biases by Imperialist Competitive Algorithm (ICA) in order to train network for predicting earthquake intensity in Richter. For this reason, we used dependent parameters like earthquake occurrence time, epicenter's latitude and longitude in degree, focal depth in kilometer, and the seismological center distance from epicenter and earthquake focal center in kilometer which has been provided by Berkeley data base. The studied neural network has two hidden layer: its first layer has 16 neurons and the second layer has 24 neurons. By using ICA algorithm, average error for testing data is 0.0007 with a variance equal to 0.318. The earthquake prediction error in Richter by MSE criteria for ICA algorithm is 0.101, but by using GA, the MSE value is 0.115. | Next, by using a neural network approach and 955 data set related to Alishan's neighbourhood highways in Taiwan, created a model for research on earthquake @cite_8 . | {
"cite_N": [
"@cite_8"
],
"mid": [
"1969129136"
],
"abstract": [
"Abstract This paper is aimed at creating an empirical model for assessing failure potential of highway slopes, with a special attention to the failure characteristics of the highway slopes in the Alishan, Taiwan area prior to, and post, the 1999 Chi-Chi, Taiwan earthquake. The basis of the study is a large database of 955 slope records from four highways in the Alishan area. Artificial neural network (ANN) is utilized to “learn” from this database. The developed ANN model is then used to study the effect of the Chi-Chi earthquake on the slope failure characteristics in the Alishan area. Significant changes in the degrees of influence of several factors (variables) are found and possible reasons for such changes are discussed. The novelty of this paper lies in the fact that the developed ANN models are used as a tool to investigate the slope failure characteristics before and after the Chi-Chi earthquake."
]
} |
1704.04313 | 2950164958 | Extracting per-frame features using convolutional neural networks for real-time processing of video data is currently mainly performed on powerful GPU-accelerated workstations and compute clusters. However, there are many applications such as smart surveillance cameras that require or would benefit from on-site processing. To this end, we propose and evaluate a novel algorithm for change-based evaluation of CNNs for video data recorded with a static camera setting, exploiting the spatio-temporal sparsity of pixel changes. We achieve an average speed-up of 8.6x over a cuDNN baseline on a realistic benchmark with a negligible accuracy loss of less than 0.1 and no retraining of the network. The resulting energy efficiency is 10x higher than that of per-frame evaluation and reaches an equivalent of 328 GOp s W on the Tegra X1 platform. | For our evaluations we are interested in performing object detection or semantic segmentation, which are both often applied to high-resolution images and video streams with frame rates above 10 ,frame s for meaningful applications. With still image object classification being considered solved by having achieved beyond human accuracy @cite_22 @cite_32 , there is now a rapidly increasing interest in extracting information from video data, e.g. video tagging and action recognition on datasets that have recently become available (Sports-1M @cite_41 , Youtube-8M @cite_15 ). | {
"cite_N": [
"@cite_41",
"@cite_15",
"@cite_32",
"@cite_22"
],
"mid": [
"2016053056",
"2524365899",
"2949650786",
"2117539524"
],
"abstract": [
"Convolutional Neural Networks (CNNs) have been established as a powerful class of models for image recognition problems. Encouraged by these results, we provide an extensive empirical evaluation of CNNs on large-scale video classification using a new dataset of 1 million YouTube videos belonging to 487 classes. We study multiple approaches for extending the connectivity of a CNN in time domain to take advantage of local spatio-temporal information and suggest a multiresolution, foveated architecture as a promising way of speeding up the training. Our best spatio-temporal networks display significant performance improvements compared to strong feature-based baselines (55.3 to 63.9 ), but only a surprisingly modest improvement compared to single-frame models (59.3 to 60.9 ). We further study the generalization performance of our best model by retraining the top layers on the UCF-101 Action Recognition dataset and observe significant performance improvements compared to the UCF-101 baseline model (63.3 up from 43.9 ).",
"Many recent advancements in Computer Vision are attributed to large datasets. Open-source software packages for Machine Learning and inexpensive commodity hardware have reduced the barrier of entry for exploring novel approaches at scale. It is possible to train models over millions of examples within a few days. Although large-scale datasets exist for image understanding, such as ImageNet, there are no comparable size video classification datasets. In this paper, we introduce YouTube-8M, the largest multi-label video classification dataset, composed of 8 million videos (500K hours of video), annotated with a vocabulary of 4800 visual entities. To get the videos and their labels, we used a YouTube video annotation system, which labels videos with their main topics. While the labels are machine-generated, they have high-precision and are derived from a variety of human-based signals including metadata and query click signals. We filtered the video labels (Knowledge Graph entities) using both automated and manual curation strategies, including asking human raters if the labels are visually recognizable. Then, we decoded each video at one-frame-per-second, and used a Deep CNN pre-trained on ImageNet to extract the hidden representation immediately prior to the classification layer. Finally, we compressed the frame features and make both the features and video-level labels available for download. We trained various (modest) classification models on the dataset, evaluated them using popular evaluation metrics, and report them as baselines. Despite the size of the dataset, some of our models train to convergence in less than a day on a single machine using TensorFlow. We plan to release code for training a TensorFlow model and for computing metrics.",
"Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers---8x deeper than VGG nets but still having lower complexity. An ensemble of these residual nets achieves 3.57 error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28 relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.",
"The 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."
]
} |
1704.04313 | 2950164958 | Extracting per-frame features using convolutional neural networks for real-time processing of video data is currently mainly performed on powerful GPU-accelerated workstations and compute clusters. However, there are many applications such as smart surveillance cameras that require or would benefit from on-site processing. To this end, we propose and evaluate a novel algorithm for change-based evaluation of CNNs for video data recorded with a static camera setting, exploiting the spatio-temporal sparsity of pixel changes. We achieve an average speed-up of 8.6x over a cuDNN baseline on a realistic benchmark with a negligible accuracy loss of less than 0.1 and no retraining of the network. The resulting energy efficiency is 10x higher than that of per-frame evaluation and reaches an equivalent of 328 GOp s W on the Tegra X1 platform. | We are specifically interested in video sequences obtained from a static camera. While some such dataset exist, most of them are specifically targeted at person tracking and or re-identification and do not provide labeled data for multi-class object detection or segmentation. However, the dataset used in @cite_8 provides ground truth labels for 10-class semantic segmentation from an urban street surveillance perspective, and while they work with individual images, several surrounding unlabeled frames and a trained convolutional network are available. An example image labeled with the provided CNN is shown in fig:labelling , and a sample sequence of 3 images is visualized in fig:gloria_seq . | {
"cite_N": [
"@cite_8"
],
"mid": [
"2535437746"
],
"abstract": [
"Detecting and classifying targets in video streams from surveillance cameras is a cumbersome, error-prone and expensive task. Often, the incurred costs are prohibitive for real-time monitoring. This leads to data being stored locally or transmitted to a central storage site for post-incident examination. The required communication links and archiving of the video data are still expensive and this setup excludes preemptive actions to respond to imminent threats. An effective way to overcome these limitations is to build a smart camera that analyzes the data on-site, close to the sensor, and transmits alerts when relevant video sequences are detected. Deep neural networks (DNNs) have come to outperform humans in visual classifications tasks and are also performing exceptionally well on other computer vision tasks. The concept of DNNs and Convolutional Networks (ConvNets) can easily be extended to make use of higher-dimensional input data such as multispectral data. We explore this opportunity in terms of achievable accuracy and required computational effort. To analyze the precision of DNNs for scene labeling in an urban surveillance scenario we have created a dataset with 8 classes obtained in a field experiment. We combine an RGB camera with a 25-channel VIS-NIR snapshot sensor to assess the potential of multispectral image data for target classification. We evaluate several new DNNs, showing that the spectral information fused together with the RGB frames can be used to improve the accuracy of the system or to achieve similar accuracy with a 3x smaller computation effort. We achieve a very high per-pixel accuracy of 99.1 . Even for scarcely occurring, but particularly interesting classes, such as cars, 75 of the pixels are labeled correctly with errors occurring only around the border of the objects. This high accuracy was obtained with a training set of only 30 labeled images, paving the way for fast adaptation to various application scenarios."
]
} |
1704.04313 | 2950164958 | Extracting per-frame features using convolutional neural networks for real-time processing of video data is currently mainly performed on powerful GPU-accelerated workstations and compute clusters. However, there are many applications such as smart surveillance cameras that require or would benefit from on-site processing. To this end, we propose and evaluate a novel algorithm for change-based evaluation of CNNs for video data recorded with a static camera setting, exploiting the spatio-temporal sparsity of pixel changes. We achieve an average speed-up of 8.6x over a cuDNN baseline on a realistic benchmark with a negligible accuracy loss of less than 0.1 and no retraining of the network. The resulting energy efficiency is 10x higher than that of per-frame evaluation and reaches an equivalent of 328 GOp s W on the Tegra X1 platform. | Further research has focused on optimizing semantic segmentation and object detection algorithms to better reuse already computed features by eliminating any non-convolutional elements from the network @cite_9 @cite_14 @cite_7 . Simplifying the operations in a network, such as low-rank approximations of 2D convolutions or by simply designing smaller networks with state-of-the-art methods have been evaluated in @cite_12 @cite_5 @cite_36 . | {
"cite_N": [
"@cite_14",
"@cite_7",
"@cite_36",
"@cite_9",
"@cite_5",
"@cite_12"
],
"mid": [
"2953106684",
"2952632681",
"",
"1483870316",
"2279098554",
"2950967261"
],
"abstract": [
"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.",
"Convolutional networks are powerful visual models that yield hierarchies of features. We show that convolutional networks by themselves, trained end-to-end, pixels-to-pixels, exceed the state-of-the-art in semantic segmentation. Our key insight is to build \"fully convolutional\" networks that take input of arbitrary size and produce correspondingly-sized output with efficient inference and learning. We define and detail the space of fully convolutional networks, explain their application to spatially dense prediction tasks, and draw connections to prior models. We adapt contemporary classification networks (AlexNet, the VGG net, and GoogLeNet) into fully convolutional networks and transfer their learned representations by fine-tuning to the segmentation task. We then define a novel architecture that combines semantic information from a deep, coarse layer with appearance information from a shallow, fine layer to produce accurate and detailed segmentations. Our fully convolutional network achieves state-of-the-art segmentation of PASCAL VOC (20 relative improvement to 62.2 mean IU on 2012), NYUDv2, and SIFT Flow, while inference takes one third of a second for a typical image.",
"",
"We present YOLO, a new approach to object detection. Prior work on object detection repurposes classifiers to perform detection. Instead, we frame object detection as a regression problem to spatially separated bounding boxes and associated class probabilities. A single neural network predicts bounding boxes and class probabilities directly from full images in one evaluation. Since the whole detection pipeline is a single network, it can be optimized end-to-end directly on detection performance. Our unified architecture is extremely fast. Our base YOLO model processes images in real-time at 45 frames per second. A smaller version of the network, Fast YOLO, processes an astounding 155 frames per second while still achieving double the mAP of other real-time detectors. Compared to state-of-the-art detection systems, YOLO makes more localization errors but is far less likely to predict false detections where nothing exists. Finally, YOLO learns very general representations of objects. It outperforms all other detection methods, including DPM and R-CNN, by a wide margin when generalizing from natural images to artwork on both the Picasso Dataset and the People-Art Dataset.",
"Recent research on deep neural networks has focused primarily on improving accuracy. For a given accuracy level, it is typically possible to identify multiple DNN architectures that achieve that accuracy level. With equivalent accuracy, smaller DNN architectures offer at least three advantages: (1) Smaller DNNs require less communication across servers during distributed training. (2) Smaller DNNs require less bandwidth to export a new model from the cloud to an autonomous car. (3) Smaller DNNs are more feasible to deploy on FPGAs and other hardware with limited memory. To provide all of these advantages, we propose a small DNN architecture called SqueezeNet. SqueezeNet achieves AlexNet-level accuracy on ImageNet with 50x fewer parameters. Additionally, with model compression techniques we are able to compress SqueezeNet to less than 0.5MB (510x smaller than AlexNet).",
"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."
]
} |
1704.04313 | 2950164958 | Extracting per-frame features using convolutional neural networks for real-time processing of video data is currently mainly performed on powerful GPU-accelerated workstations and compute clusters. However, there are many applications such as smart surveillance cameras that require or would benefit from on-site processing. To this end, we propose and evaluate a novel algorithm for change-based evaluation of CNNs for video data recorded with a static camera setting, exploiting the spatio-temporal sparsity of pixel changes. We achieve an average speed-up of 8.6x over a cuDNN baseline on a realistic benchmark with a negligible accuracy loss of less than 0.1 and no retraining of the network. The resulting energy efficiency is 10x higher than that of per-frame evaluation and reaches an equivalent of 328 GOp s W on the Tegra X1 platform. | Obtaining per-frame features naturally seems like an easier task when these frames belong to a video sequence rather than a random collection of images. Limited movement of objects in a frame can be exploited in object tracking by working with a limited search window within the frame @cite_0 , not only reducing the problem size, but also simplifying the regression task---up until the tracked target is occluded by a large object. | {
"cite_N": [
"@cite_0"
],
"mid": [
"2964253307"
],
"abstract": [
"Machine learning techniques are often used in computer vision due to their ability to leverage large amounts of training data to improve performance. Unfortunately, most generic object trackers are still trained from scratch online and do not benefit from the large number of videos that are readily available for offline training. We propose a method for offline training of neural networks that can track novel objects at test-time at 100 fps. Our tracker is significantly faster than previous methods that use neural networks for tracking, which are typically very slow to run and not practical for real-time applications. Our tracker uses a simple feed-forward network with no online training required. The tracker learns a generic relationship between object motion and appearance and can be used to track novel objects that do not appear in the training set. We test our network on a standard tracking benchmark to demonstrate our tracker’s state-of-the-art performance. Further, our performance improves as we add more videos to our offline training set. To the best of our knowledge, our tracker (Our tracker is available at http: davheld.github.io GOTURN GOTURN.html) is the first neural-network tracker that learns to track generic objects at 100 fps."
]
} |
1704.04326 | 2605638678 | The performance of modern face recognition systems is a function of the dataset on which they are trained. Most datasets are largely biased toward "near-frontal" views with benign lighting conditions, negatively effecting recognition performance on images that do not meet these criteria. The proposed approach demonstrates how a baseline training set can be augmented to increase pose and lighting variability using semi-synthetic images with simulated pose and lighting conditions. The semi-synthetic images are generated using a fast and robust 3-d shape estimation and rendering pipeline which includes the full head and background. Various methods of incorporating the semi-synthetic renderings into the training procedure of a state of the art deep neural network-based recognition system without modifying the structure of the network itself are investigated. Quantitative results are presented on the challenging IJB-A identification dataset using a state of the art recognition pipeline as a baseline. | Rather than using normalization to make the recognition task easier at test time, a third approach is to instead render challenging views to make the recognition task more challenging at training time. In theory, this method forces the network to learn features which are invariant to the transformations included in the training set. @cite_9 demonstrate the feasibility of such an approach by augmenting a baseline training set with renderings at fixed yaw angle intervals. Rather than estimating 3-d shape in the original image, renderings are produced using a set of exemplar'' face shapes for each subject. The proposed approach is similar in goal but has several important distinctions. The 3-d shape of the subject in a given image is estimated before rendering in order to simulate viewpoint change as accurately as possible. Rather than rendering the training subjects from a small, fixed number of views, viewpoint in yaw, pitch, and roll is randomly perturbed to achieve maximal pose invariance. We refer to this random perturbation as pose jittering''. Finally, a rendering procedure for altering lighting conditions in addition to pose is presented ( lighting jittering'') in the proposed approach. | {
"cite_N": [
"@cite_9"
],
"mid": [
"2304348237"
],
"abstract": [
"Face recognition capabilities have recently made extraordinary leaps. Though this progress is at least partially due to ballooning training set sizes – huge numbers of face images downloaded and labeled for identity – it is not clear if the formidable task of collecting so many images is truly necessary. We propose a far more accessible means of increasing training data sizes for face recognition systems: Domain specific data augmentation. We describe novel methods of enriching an existing dataset with important facial appearance variations by manipulating the faces it contains. This synthesis is also used when matching query images represented by standard convolutional neural networks. The effect of training and testing with synthesized images is tested on the LFW and IJB-A (verification and identification) benchmarks and Janus CS2. The performances obtained by our approach match state of the art results reported by systems trained on millions of downloaded images."
]
} |
1704.04360 | 2605612822 | We address the problem of epipolar geometry using the motion of silhouettes. Such methods match epipolar lines or frontier points across views, which are then used as the set of putative correspondences. We introduce an approach that improves by two orders of magnitude the performance over state-of-the-art methods, by significantly reducing the number of outliers in the putative matching. We model the frontier points' correspondence problem as constrained flow optimization, requiring small differences between their coordinates over consecutive frames. Our approach is formulated as a Linear Integer Program and we show that due to the nature of our problem, it can be solved efficiently in an iterative manner. Our method was validated on four standard datasets providing accurate calibrations across very different viewpoints. | The most common uses of silhouettes in multi-camera systems are for shape-from-silhouettes @cite_4 @cite_36 @cite_29 and camera calibration @cite_30 @cite_8 @cite_18 @cite_10 @cite_7 . In shape-from-silhouettes, the goal is to recover the visual hull @cite_11 @cite_20 of the object. In calibration, it is assumed that the motion of the silhouette is fully observed across different views. Correspondences are established between frontier points across different views based on the epipolar tangency constraints @cite_38 and are then used to compute the epipolar geometry. Most methods require a specific configuration that cannot be applied in a general setting, which is considered here. These include calibrated cameras, static objects, orthographic projection models, or known (turntable) motion @cite_33 @cite_37 @cite_22 @cite_3 @cite_30 . Calibration can be carried out without explicitly finding the corresponding tangent epipolar lines, such as in @cite_18 @cite_31 . These methods require a good initial guess to converge. | {
"cite_N": [
"@cite_30",
"@cite_38",
"@cite_18",
"@cite_37",
"@cite_4",
"@cite_33",
"@cite_7",
"@cite_8",
"@cite_36",
"@cite_22",
"@cite_29",
"@cite_3",
"@cite_31",
"@cite_10",
"@cite_20",
"@cite_11"
],
"mid": [
"2151226352",
"2133369382",
"1493881924",
"2142390575",
"2163680646",
"2058603116",
"2131994698",
"2085571680",
"1885010708",
"2137610415",
"2109969143",
"",
"",
"",
"2165102296",
"2117007522"
],
"abstract": [
"We present a new approach to camera calibration as a part of a complete and practical system to recover digital copies of sculpture from uncalibrated image sequences taken under turntable motion. In this paper, we introduce the concept of the silhouette coherence of a set of silhouettes generated by a 3D object. We show how the maximization of the silhouette coherence can be exploited to recover the camera poses and focal length. Silhouette coherence can be considered as a generalization of the well-known epipolar tangency constraint for calculating motion from silhouettes or outlines alone. Further, silhouette coherence exploits all the geometric information encoded in the silhouette (not just at epipolar tangency points) and can be used in many practical situations where point correspondences or outer epipolar tangents are unavailable. We present an algorithm for exploiting silhouette coherence to efficiently and reliably estimate camera motion. We use this algorithm to reconstruct very high quality 3D models from uncalibrated circular motion sequences, even when epipolar tangency points are not available or the silhouettes are truncated. The algorithm has been integrated into a practical system and has been tested on more than 50 uncalibrated sequences to produce high quality photo-realistic models. Three illustrative examples are included in this paper. The algorithm is also evaluated quantitatively by comparing it to a state-of-the-art system that exploits only epipolar tangents",
"The frontier of a curved surface is the envelope of contour generators showing the boundary, at least locally, of the visible region swept out under viewer motion. In general, the outlines of curved surfaces (apparent contours) from different viewpoints are generated by different contour generators on the surface and hence do not provide a constraint on viewer motion. We show that frontier points, however, have projections which correspond to a real point on the surface and can be used to constrain viewer motion by the epipolar constraint. We show how to recover viewer motion from frontier points for both continuous and discrete motion, calibrated and uncalibrated cameras. We present preliminary results of an iterative scheme to recover the epipolar line structure from real image sequences using only the outlines of curved surfaces. A statistical evaluation as also performed to estimate the stability of the solution. >",
"This paper addresses the problem of camera calibration using object silhouettes in image sequences. It is known that silhouettes encode information on camera parameters by the fact that their associated viewing cones should present a common intersection in space. In this paper, we investigate how to evaluate calibration parameters given a set of silhouettes, and how to optimize such parameters with silhouette cues only. The objective is to provide on-line tools for silhouette based modeling applications in multiple camera environments. Our contributions with respect to existing works in this field is first to establish the exact constraint that camera parameters should satisfy with respect to silhouettes, and second to derive from this constraint new practical criteria to evaluate and to optimize camera parameters. Results on both synthetic and real data illustrate the interest of the proposed framework.",
"This paper addresses the problem of estimating the motion of a camera as it observes the outline (or apparent contour) of a solid bounded by a smooth surface in successive image frames. In this context, the surface points that project onto the outline of an object depend on the viewpoint and the only true correspondences between two outlines of the same object are the projections of frontier points where the viewing rays intersect in the tangent plane of the surface. In turn, the epipolar geometry is easily estimated once these correspondences have been identified. Given the apparent contours detected in an image sequence, a robust procedure based on RANSAC and a voting strategy is proposed to simultaneously estimate the camera configurations and a consistent set of frontier point projections by enforcing the redundancy of multiview epipolar geometry. The proposed approach is, in principle, applicable to orthographic, weak-perspective, and affine projection models. Experiments with nine real image sequences are presented for the orthographic projection case, including a quantitative comparison with the ground-truth data for the six data sets for which the latter information is available. Sample visual hulls have been computed from all image sequences for qualitative evaluation.",
"Shape-from-silhouette (SFS), also known as visual hull (VH) construction, is a popular 3D reconstruction method, which estimates the shape of an object from multiple silhouette images. The original SFS formulation assumes that the entire silhouette images are captured either at the same time or while the object is static. This assumption is violated when the object moves or changes shape. Hence the use of SFS with moving objects has been restricted to treating each time instant sequentially and independently. Recently we have successfully extended the traditional SFS formulation to refine the shape of a rigidly moving object over time. We further extend SFS to apply to dynamic articulated objects. Given silhouettes of a moving articulated object, the process of recovering the shape and motion requires two steps: (1) correctly segmenting (points on the boundary of) the silhouettes to each articulated part of the object, (2) estimating the motion of each individual part using the segmented silhouette. In this paper, we propose an iterative algorithm to solve this simultaneous assignment and alignment problem. Once we have estimated the shape and motion of each part of the object, the articulation points between each pair of rigid parts are obtained by solving a simple motion constraint between the connected parts. To validate our algorithm, we first apply it to segment the different body parts and estimate the joint positions of a person. The acquired kinematic (shape and joint) information is then used to track the motion of the person in new video sequences.",
"We propose an exact method for efficiently and robustly computing the visual hull of an object from image contours. Unlike most existing approaches, ours computes an exact description of the visual hull polyhedron associated to polygonal image contours. Furthermore, the proposed approach is fast and allows real-time recovery of both manifold and watertight visual hull polyhedra. The process involves three main steps. First, a coarse geometrical approximation of the visual hull is computed by retrieving its viewing edges, an unconnected subset of the wanted mesh. Then, local orientation and connectivity rules are used to walk along the relevant viewing cone intersection boundaries, so as to iteratively generate the missing surface points and connections. A final connection walkthrough allows us to identify the planar contours for each face of the polyhedron. Implementation details and results with synthetic and real data are presented.",
"This paper addresses the problem of recovering both the intrinsic and extrinsic parameters of a camera from the silhouettes of an object in a turntable sequence. Previous silhouette-based approaches have exploited correspondences induced by epipolar tangents to estimate the image invariants under turntable motion and achieved a weak calibration of the cameras. It is known that the fundamental matrix relating any two views in a turntable sequence can be expressed explicitly in terms of the image invariants, the rotation angle, and a fixed scalar. It will be shown that the imaged circular points for the turntable plane can also be formulated in terms of the same image invariants and fixed scalar. This allows the imaged circular points to be recovered directly from the estimated image invariants, and provide constraints for the estimation of the imaged absolute conic. The camera calibration matrix can thus be recovered. A robust method for estimating the fixed scalar from image triplets is introduced, and a method for recovering the rotation angles using the estimated imaged circular points and epipoles is presented. Using the estimated camera intrinsics and extrinsics, a Euclidean reconstruction can be obtained. Experimental results on real data sequences are presented, which demonstrate the high precision achieved by the proposed method.",
"In this paper we present an automatic method for calibrating a network of cameras that works by analyzing only the motion of silhouettes in the multiple video streams. This is particularly useful for automatic reconstruction of a dynamic event using a camera network in a situation where pre-calibration of the cameras is impractical or even impossible. The key contribution of this work is a RANSAC-based algorithm that simultaneously computes the epipolar geometry and synchronization of a pair of cameras only from the motion of silhouettes in video. Our approach involves first independently computing the fundamental matrix and synchronization for multiple pairs of cameras in the network. In the next stage the calibration and synchronization for the complete network is recovered from the pairwise information. Finally, a visual-hull algorithm is used to reconstruct the shape of the dynamic object from its silhouettes in video. For unsynchronized video streams with sub-frame temporal offsets, we interpolate silhouettes between successive frames to get more accurate visual hulls. We show the effectiveness of our method by remotely calibrating several different indoor camera networks from archived video streams.",
"Two planar mirrors are positioned to show five views of an object, and snapshots are captured from different viewpoints. We present closed form solutions for calculating the focal length, principal point, mirror and camera poses directly from the silhouette outlines of the object and its reflections. In the noisy case, these equations are used to form initial parameter estimates that are refined using iterative minimisation. The self-calibration allows the visual cones from each silhouette to be specified in a common reference frame so that the visual hull can be constructed. The proposed setup provides a simple method for creating 3D multimedia content that does not rely on specialised equipment. Experimental results demonstrate the reconstruction of a toy horse and a locust from real images. Synthetic images are used to quantify the sensitivity of the self-calibration to quantisation noise. In terms of the silhouette calibration ratio, degradation in silhouette quality has a greater effect on silhouette set consistency than computed calibration parameters.",
"This paper addresses the problem of recovering structure and motion from silhouettes. Silhouettes are projections of contour generators which are viewpoint dependent, and hence do not readily provide point correspondences for exploitation in motion estimation. Previous works have exploited correspondences induced by epipolar tangencies, and a successful solution has been developed in the special case of circular motion (turnable sequences). However, the main drawbacks are (1) new views cannot be added easily at a later time, and (2) part of the structure will always remain invisible under circular motion. In this paper we overcome the above problems by incorporating arbitrary general views and estimating the camera poses using silhouettes alone. We present a complete and practical system which produces high quality 3D models from 2D uncalibrated silhouettes. The 3D models thus obtained can be refined incrementally by adding new arbitrary views and estimating their poses. Experimental results on various objects are presented, demonstrating the quality of the reconstructions.",
"We propose a novel method for computing a four-dimensional (4D) representation of the spatio-temporal visual hull of a dynamic scene, based on an extension of a recent provably correct Delaunay meshing algorithm. By considering time as an additional dimension, our approach exploits seamlessly the time coherence between different frames to produce a compact and high-quality 4D mesh representation of the visual hull. The 3D visual hull at a given time instant is easily obtained by intersecting this 4D mesh with a temporal plane, thus enabling interpolation of objects' shape between consecutive frames. In addition, our approach offers easy and extensive control over the size and quality of the output mesh as well as over its associated re- projection error. Our numerical experiments demonstrate the effectiveness and flexibility of our approach for generating compact, high-quality, time-coherent visual hull representations from real silhouette image data.",
"",
"",
"",
"The visual hull is widely used to produce three dimensional models from multiple views, due to the reliability of the resulting surface. This paper presents a novel method for efficiently evaluating the exact view-dependent visual hull without using approximations. Methods for selecting intersections and ordering them via the cross ratio are presented. Results show the high quality of the surfaces produced using this method.",
"Many algorithms for both identifying and reconstructing a 3-D object are based on the 2-D silhouettes of the object. In general, identifying a nonconvex object using a silhouette-based approach implies neglecting some features of its surface as identification clues. The same features cannot be reconstructed by volume intersection techniques using multiple silhouettes of the object. This paper addresses the problem of finding which parts of a nonconvex object are relevant for silhouette-based image understanding. For this purpose, the geometric concept of visual hull of a 3-D object is introduced. This is the closest approximation of object S that can be obtained with the volume intersection approach; it is the maximal object silhouette-equivalent to S, i.e., which can be substituted for S without affecting any silhouette. Only the parts of the surface of S that also lie on the surface of the visual hull can be reconstructed or identified using silhouette-based algorithms. The visual hull depends not only on the object but also on the region allowed to the viewpoint. Two main viewing regions result in the external and internal visual hull. In the former case the viewing region is related to the convex hull of S, in the latter it is bounded by S. The internal visual hull also admits an interpretation not related to silhouettes. Algorithms for computing visual hulls are presented and their complexity analyzed. In general, the visual hull of a 3-D planar face object turns out to be bounded by planar and curved patches. >"
]
} |
1704.04360 | 2605612822 | We address the problem of epipolar geometry using the motion of silhouettes. Such methods match epipolar lines or frontier points across views, which are then used as the set of putative correspondences. We introduce an approach that improves by two orders of magnitude the performance over state-of-the-art methods, by significantly reducing the number of outliers in the putative matching. We model the frontier points' correspondence problem as constrained flow optimization, requiring small differences between their coordinates over consecutive frames. Our approach is formulated as a Linear Integer Program and we show that due to the nature of our problem, it can be solved efficiently in an iterative manner. Our method was validated on four standard datasets providing accurate calibrations across very different viewpoints. | In this paper we consider calibration in the most general setting where only the motion of silhouettes is available without further assumptions. Sinha and Pollefeys @cite_8 considered such a setting for calibration. They searched for the epipoles by randomly sampling lines from the tangent envelope of the silhouette. They used RANSAC for extracting the most plausible solution. Ben- @cite_6 proposed a temporal binary descriptor, denoted as a motion-barcode, for suggesting the corresponding epipolar lines across views. They sampled the lines from the tangent envelope according to the similarity induced by their descriptor. Importantly, they showed that accuracy and runtime are markedly improved. In both methods there is a significant number of outliers and the frontier points in the current frame are matched without taking into account the previous or the next corresponding points. @cite_39 used the same descriptor for calibrating crowded scenes. However, they considered noisy, low-resolution images and provided limited accuracy. | {
"cite_N": [
"@cite_39",
"@cite_6",
"@cite_8"
],
"mid": [
"2483609476",
"2247382355",
"2085571680"
],
"abstract": [
"Computing the epipolar geometry between cameras with very different viewpoints is often very difficult. The appearance of objects can vary greatly, and it is difficult to find corresponding feature points. Prior methods searched for corresponding epipolar lines using points on the convex hull of the silhouette of a single moving object. These methods fail when the scene includes multiple moving objects. This paper extends previous work to scenes having multiple moving objects by using the “Motion Barcodes”, a temporal signature of lines. Corresponding epipolar lines have similar motion barcodes, and candidate pairs of corresponding epipoar lines are found by the similarity of their motion barcodes. As in previous methods we assume that cameras are relatively stationary and that moving objects have already been extracted using background subtraction.",
"Computing the epipolar geometry between cameras with very different viewpoints is often problematic as matching points are hard to find. In these cases, it has been proposed to use information from dynamic objects in the scene for suggesting point and line correspondences. We propose a speed up of about two orders of magnitude, as well as an increase in robustness and accuracy, to methods computing epipolar geometry from dynamic silhouettes. This improvement is based on a new temporal signature: motion barcode for lines. Motion barcode is a binary temporal sequence for lines, indicating for each frame the existence of at least one foreground pixel on that line. The motion barcodes of two corresponding epipolar lines are very similar, so the search for corresponding epipolar lines can be limited only to lines having similar barcodes. The use of motion barcodes leads to increased speed, accuracy, and robustness in computing the epipolar geometry.",
"In this paper we present an automatic method for calibrating a network of cameras that works by analyzing only the motion of silhouettes in the multiple video streams. This is particularly useful for automatic reconstruction of a dynamic event using a camera network in a situation where pre-calibration of the cameras is impractical or even impossible. The key contribution of this work is a RANSAC-based algorithm that simultaneously computes the epipolar geometry and synchronization of a pair of cameras only from the motion of silhouettes in video. Our approach involves first independently computing the fundamental matrix and synchronization for multiple pairs of cameras in the network. In the next stage the calibration and synchronization for the complete network is recovered from the pairwise information. Finally, a visual-hull algorithm is used to reconstruct the shape of the dynamic object from its silhouettes in video. For unsynchronized video streams with sub-frame temporal offsets, we interpolate silhouettes between successive frames to get more accurate visual hulls. We show the effectiveness of our method by remotely calibrating several different indoor camera networks from archived video streams."
]
} |
1704.04360 | 2605612822 | We address the problem of epipolar geometry using the motion of silhouettes. Such methods match epipolar lines or frontier points across views, which are then used as the set of putative correspondences. We introduce an approach that improves by two orders of magnitude the performance over state-of-the-art methods, by significantly reducing the number of outliers in the putative matching. We model the frontier points' correspondence problem as constrained flow optimization, requiring small differences between their coordinates over consecutive frames. Our approach is formulated as a Linear Integer Program and we show that due to the nature of our problem, it can be solved efficiently in an iterative manner. Our method was validated on four standard datasets providing accurate calibrations across very different viewpoints. | We used the similarity proposed by the binary motion-barcode descriptor @cite_6 as one of the cues for our algorithm. Similar motion-based binary descriptors have been proposed and used by @cite_5 , @cite_0 . Both methods assume a planar structure of the scene. Ben- @cite_17 used a similar descriptor for matching events across different views, but their method does not provide accurate localization. Pundik and Moses @cite_9 introduced a similar motion-based descriptor, line signal, and used it for video synchronization. It depends on the color and was used under the assumption of known calibration. | {
"cite_N": [
"@cite_9",
"@cite_6",
"@cite_0",
"@cite_5",
"@cite_17"
],
"mid": [
"1586713560",
"2247382355",
"2128923002",
"2125949781",
"2963810042"
],
"abstract": [
"Time synchronization of video sequences in a multicamera system is necessary for successfully analyzing the acquired visual information. Even if synchronization is established, its quality may deteriorate over time due to a variety of reasons, most notably frame dropping. Consequently, synchronization must be actively maintained. This paper presents a method for online synchronization that relies only on the video sequences. We introduce a novel definition of low level temporal signals computed from epipolar lines. The spatial matching of two such temporal signals is given by the fundamental matrix. Thus, no pixel correspondence is required, bypassing the problem of correspondence changes in the presence of motion. The synchronization is determined from registration of the temporal signals. We consider general video data with substantial movement in the scene, for which high level information may be hard to extract from each individual camera (e.g., computing trajectories in crowded scenes). Furthermore, a trivial correspondence between the sequences is not assumed to exist. The method is online and can be used to resynchronize video sequences every few seconds, with only a small delay. Experiments on indoor and outdoor sequences demonstrate the effectiveness of the method.",
"Computing the epipolar geometry between cameras with very different viewpoints is often problematic as matching points are hard to find. In these cases, it has been proposed to use information from dynamic objects in the scene for suggesting point and line correspondences. We propose a speed up of about two orders of magnitude, as well as an increase in robustness and accuracy, to methods computing epipolar geometry from dynamic silhouettes. This improvement is based on a new temporal signature: motion barcode for lines. Motion barcode is a binary temporal sequence for lines, indicating for each frame the existence of at least one foreground pixel on that line. The motion barcodes of two corresponding epipolar lines are very similar, so the search for corresponding epipolar lines can be limited only to lines having similar barcodes. The use of motion barcodes leads to increased speed, accuracy, and robustness in computing the epipolar geometry.",
"This paper presents a novel approach for matching 2D points between a video projector and a digital camera. Our method is motivated by camera-projector applications for which the projected image needs to be warped to prevent geometric distortion. Since the warping process often needs geometric information on the 3D scene that can only be obtained from triangulation, we propose a technique for matching points in the projector to points in the camera based on arbitrary video sequences. The novelty of our method lies in the fact that it does not require the use of pre-designed structured light patterns as is usually the case. The back bone of our application lies in a function that matches activity patterns instead of colors. This makes our method robust to pose, to severe photometric and geometric distortions. It also does not require calibration of the color response curve of the camera-projector system. We present quantitative and qualitative results with synthetic and real life examples, and compare the proposed method with the scale invariant feature transform (SIFT) method and with a state-of-the-art structured light technique. We show that our method performs almost as well as structured light methods and significantly outperforms SIFT when the contrast of the video captured by the camera has been degraded.",
"In this paper, we consider the problem of finding correspondences between distributed cameras that have partially overlapping field of views. When multiple cameras with adaptable orientations and zooms are deployed, as in many wide area surveillance applications, identifying correspondence between different activities becomes a fundamental issue. We propose a correspondence method based upon activity features that, unlike photometric features, have certain geometry independence properties. The proposed method is robust to pose, illumination and geometric effects, unsupervised (does not require any calibration objects). In addition, these features are amenable to low communication bandwidth and distributed network applications. We present quantitative and qualitative results with synthetic and real life examples, and compare the proposed method with scale invariant feature transform (SIFT) based method. We show that our method significantly outperforms the SIFT method when cameras have significantly different orientations. We then describe extensions of our method in a number of directions including topology reconstruction, camera calibration, and distributed anomaly detection.",
""
]
} |
1704.04516 | 2953048485 | The discriminative power of modern deep learning models for 3D human action recognition is growing ever so potent. In conjunction with the recent resurgence of 3D human action representation with 3D skeletons, the quality and the pace of recent progress have been significant. However, the inner workings of state-of-the-art learning based methods in 3D human action recognition still remain mostly black-box. In this work, we propose to use a new class of models known as Temporal Convolutional Neural Networks (TCN) for 3D human action recognition. Compared to popular LSTM-based Recurrent Neural Network models, given interpretable input such as 3D skeletons, TCN provides us a way to explicitly learn readily interpretable spatio-temporal representations for 3D human action recognition. We provide our strategy in re-designing the TCN with interpretability in mind and how such characteristics of the model is leveraged to construct a powerful 3D activity recognition method. Through this work, we wish to take a step towards a spatio-temporal model that is easier to understand, explain and interpret. The resulting model, Res-TCN, achieves state-of-the-art results on the largest 3D human action recognition dataset, NTU-RGBD. | In the above methods, the key intuition is that a certain subset of joints are more important for recognizing human activities. However, it is difficult to interpret what the model parameters of each LSTM layer represent. In our proposed version of TCN, we show that our model also learns both spatial and temporal attention without the need for initial pre-training stage as in @cite_30 . Moreover, by model design based on temporal convolutions @cite_5 and residual connections @cite_25 , we can begin to directly interpret what our model parameters and features represent. | {
"cite_N": [
"@cite_30",
"@cite_5",
"@cite_25"
],
"mid": [
"2950568498",
"2550143307",
"2302255633"
],
"abstract": [
"Human action recognition is an important task in computer vision. Extracting discriminative spatial and temporal features to model the spatial and temporal evolutions of different actions plays a key role in accomplishing this task. In this work, we propose an end-to-end spatial and temporal attention model for human action recognition from skeleton data. We build our model on top of the Recurrent Neural Networks (RNNs) with Long Short-Term Memory (LSTM), which learns to selectively focus on discriminative joints of skeleton within each frame of the inputs and pays different levels of attention to the outputs of different frames. Furthermore, to ensure effective training of the network, we propose a regularized cross-entropy loss to drive the model learning process and develop a joint training strategy accordingly. Experimental results demonstrate the effectiveness of the proposed model,both on the small human action recognition data set of SBU and the currently largest NTU dataset.",
"The ability to identify and temporally segment fine-grained human actions throughout a video is crucial for robotics, surveillance, education, and beyond. Typical approaches decouple this problem by first extracting local spatiotemporal features from video frames and then feeding them into a temporal classifier that captures high-level temporal patterns. We describe a class of temporal models, which we call Temporal Convolutional Networks (TCNs), that use a hierarchy of temporal convolutions to perform fine-grained action segmentation or detection. Our Encoder-Decoder TCN uses pooling and upsampling to efficiently capture long-range temporal patterns whereas our Dilated TCN uses dilated convolutions. We show that TCNs are capable of capturing action compositions, segment durations, and long-range dependencies, and are over a magnitude faster to train than competing LSTM-based Recurrent Neural Networks. We apply these models to three challenging fine-grained datasets and show large improvements over the state of the art.",
"Deep residual networks have emerged as a family of extremely deep architectures showing compelling accuracy and nice convergence behaviors. In this paper, we analyze the propagation formulations behind the residual building blocks, which suggest that the forward and backward signals can be directly propagated from one block to any other block, when using identity mappings as the skip connections and after-addition activation. A series of ablation experiments support the importance of these identity mappings. This motivates us to propose a new residual unit, which makes training easier and improves generalization. We report improved results using a 1001-layer ResNet on CIFAR-10 (4.62 error) and CIFAR-100, and a 200-layer ResNet on ImageNet. Code is available at: https: github.com KaimingHe resnet-1k-layers."
]
} |
1704.04517 | 2606220156 | We introduce a novel framework for evaluating multimodal deep learning models with respect to their language understanding and generalization abilities. In this approach, artificial data is automatically generated according to the experimenter's specifications. The content of the data, both during training and evaluation, can be controlled in detail, which enables tasks to be created that require true generalization abilities, in particular the combination of previously introduced concepts in novel ways. We demonstrate the potential of our methodology by evaluating various visual question answering models on four different tasks, and show how our framework gives us detailed insights into their capabilities and limitations. By open-sourcing our framework, we hope to stimulate progress in the field of multimodal language understanding. | When focusing on language tasks, the simulation paradigm faces the problem that interesting language generation is a difficult task in its own right, and that the difficulty increases with the complexity of the underlying world. The bAbI tasks @cite_3 are generated by internally simulating a short scene and extracting a few simple sentences from it. A similar approach is taken by , but here the simulation is more complex, comprising a text-based role-playing game. The MazeBase game environment @cite_33 uses language as a mean to represent the game world. However, the descriptions are in an abstract, formulaic format, and the focus of the simulation is much more on the planning than the language component. The long-term research proposal of also simulates a world where an agent learns to solve tasks by communication with a teacher module. At least for a start, this module is supposed to be scripted to automatically generate appropriate responses, given its internal knowledge of the world state. | {
"cite_N": [
"@cite_33",
"@cite_3"
],
"mid": [
"2174196774",
"2962790689"
],
"abstract": [
"This paper introduces MazeBase: an environment for simple 2D games, designed as a sandbox for machine learning approaches to reasoning and planning. Within it, we create 10 simple games embodying a range of algorithmic tasks (e.g. if-then statements or set negation). A variety of neural models (fully connected, convolutional network, memory network) are deployed via reinforcement learning on these games, with and without a procedurally generated curriculum. Despite the tasks' simplicity, the performance of the models is far from optimal, suggesting directions for future development. We also demonstrate the versatility of MazeBase by using it to emulate small combat scenarios from StarCraft. Models trained on the MazeBase version can be directly applied to StarCraft, where they consistently beat the in-game AI.",
"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."
]
} |
1704.04517 | 2606220156 | We introduce a novel framework for evaluating multimodal deep learning models with respect to their language understanding and generalization abilities. In this approach, artificial data is automatically generated according to the experimenter's specifications. The content of the data, both during training and evaluation, can be controlled in detail, which enables tasks to be created that require true generalization abilities, in particular the combination of previously introduced concepts in novel ways. We demonstrate the potential of our methodology by evaluating various visual question answering models on four different tasks, and show how our framework gives us detailed insights into their capabilities and limitations. By open-sourcing our framework, we hope to stimulate progress in the field of multimodal language understanding. | Automatically generated data is common for tasks specifically focusing on the ability to efficiently process data of a certain formal structure. Here, data is deliberately stripped of any real-world connection to create an abstract capability check. Recent work in the context of deep learning has investigated sequence patterns @cite_1 , combinatorial problems @cite_11 , or executing programming language code @cite_17 , amongst others. This kind of task is particularly common for neural network models (see, for instance, more than twenty years ago). The reason for interest in abstract capability checks is that the learning process and decisions of deep networks are more difficult to interpret than shallower machine learning methods. and are more similar to our work in focusing on specific linguistic aspects. Both generate artificial data automatically based on abstract models for tasks targeting logical semantics and quantifiers, respectively. | {
"cite_N": [
"@cite_1",
"@cite_17",
"@cite_11"
],
"mid": [
"2949626814",
"1581407678",
""
],
"abstract": [
"Despite the recent achievements in machine learning, we are still very far from achieving real artificial intelligence. In this paper, we discuss the limitations of standard deep learning approaches and show that some of these limitations can be overcome by learning how to grow the complexity of a model in a structured way. Specifically, we study the simplest sequence prediction problems that are beyond the scope of what is learnable with standard recurrent networks, algorithmically generated sequences which can only be learned by models which have the capacity to count and to memorize sequences. We show that some basic algorithms can be learned from sequential data using a recurrent network associated with a trainable memory.",
"Recurrent Neural Networks (RNNs) with Long Short-Term Memory units (LSTM) are widely used because they are expressive and are easy to train. Our interest lies in empirically evaluating the expressiveness and the learnability of LSTMs in the sequence-to-sequence regime by training them to evaluate short computer programs, a domain that has traditionally been seen as too complex for neural networks. We consider a simple class of programs that can be evaluated with a single left-to-right pass using constant memory. Our main result is that LSTMs can learn to map the character-level representations of such programs to their correct outputs. Notably, it was necessary to use curriculum learning, and while conventional curriculum learning proved ineffective, we developed a new variant of curriculum learning that improved our networks' performance in all experimental conditions. The improved curriculum had a dramatic impact on an addition problem, making it possible to train an LSTM to add two 9-digit numbers with 99 accuracy.",
""
]
} |
1704.04517 | 2606220156 | We introduce a novel framework for evaluating multimodal deep learning models with respect to their language understanding and generalization abilities. In this approach, artificial data is automatically generated according to the experimenter's specifications. The content of the data, both during training and evaluation, can be controlled in detail, which enables tasks to be created that require true generalization abilities, in particular the combination of previously introduced concepts in novel ways. We demonstrate the potential of our methodology by evaluating various visual question answering models on four different tasks, and show how our framework gives us detailed insights into their capabilities and limitations. By open-sourcing our framework, we hope to stimulate progress in the field of multimodal language understanding. | The multimodal tasks of image captioning and VQA are closely related to our evaluation goal, but usually consist of real-world photos and human-written descriptions. Although we contrast such real-world'' data with artificial simulations, it should be clear that this is very unlike the visual experience of an entity situated in the real world. However, there have been experiments in which parts of the data are artificial and or generated automatically, for instance, automatic question generation from annotation @cite_9 or systematic modification of captions @cite_24 . Abstract Clipart scenes have been used for image captioning @cite_10 @cite_20 and to balance existing VQA datasets @cite_36 . Most similar to the framework is the CLEVR dataset @cite_30 . It contains images of rendered abstract 3-dimensional scenes and complex questions generated from a variety of templates. As with our work, they propose that their artificial dataset evaluation on real-world VQA datasets. | {
"cite_N": [
"@cite_30",
"@cite_36",
"@cite_9",
"@cite_24",
"@cite_10",
"@cite_20"
],
"mid": [
"2561715562",
"2273038706",
"2396147015",
"",
"1981636936",
"1996418862"
],
"abstract": [
"When building artificial intelligence systems that can reason and answer questions about visual data, we need diagnostic tests to analyze our progress and discover short-comings. Existing benchmarks for visual question answering can help, but have strong biases that models can exploit to correctly answer questions without reasoning. They also conflate multiple sources of error, making it hard to pinpoint model weaknesses. We present a diagnostic dataset that tests a range of visual reasoning abilities. It contains minimal biases and has detailed annotations describing the kind of reasoning each question requires. We use this dataset to analyze a variety of modern visual reasoning systems, providing novel insights into their abilities and limitations.",
"The complex compositional structure of language makes problems at the intersection of vision and language challenging. But language also provides a strong prior that can result in good superficial performance, without the underlying models truly understanding the visual content. This can hinder progress in pushing state of art in the computer vision aspects of multi-modal AI. In this paper, we address binary Visual Question Answering (VQA) on abstract scenes. We formulate this problem as visual verification of concepts inquired in the questions. Specifically, we convert the question to a tuple that concisely summarizes the visual concept to be detected in the image. If the concept can be found in the image, the answer to the question is \"yes\", and otherwise \"no\". Abstract scenes play two roles (1) They allow us to focus on the high-level semantics of the VQA task as opposed to the low-level recognition problems, and perhaps more importantly, (2) They provide us the modality to balance the dataset such that language priors are controlled, and the role of vision is essential. In particular, we collect fine-grained pairs of scenes for every question, such that the answer to the question is \"yes\" for one scene, and \"no\" for the other for the exact same question. Indeed, language priors alone do not perform better than chance on our balanced dataset. Moreover, our proposed approach matches the performance of a state-of-the-art VQA approach on the unbalanced dataset, and outperforms it on the balanced dataset.",
"",
"",
"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 over real images. 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 real images. Importantly, abstract images also allow the ability to generate sets of semantically similar scenes. Finding analogous sets of real images that are semantically similar would be nearly impossible. We create 1,002 sets of 10 semantically similar abstract images 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. Finally, we study the relation between the saliency and memorability of objects and their semantic importance.",
"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."
]
} |
1704.04455 | 2952535097 | Information extraction (IE) from text has largely focused on relations between individual entities, such as who has won which award. However, some facts are never fully mentioned, and no IE method has perfect recall. Thus, it is beneficial to also tap contents about the cardinalities of these relations, for example, how many awards someone has won. We introduce this novel problem of extracting cardinalities and discusses the specific challenges that set it apart from standard IE. We present a distant supervision method using conditional random fields. A preliminary evaluation results in precision between 3 and 55 , depending on the difficulty of relations. | Automated KB construction is a major effort for quite a while. Some approaches, such as YAGO @cite_8 or DBpedia @cite_6 , focus on structured parts of Wikipedia, while other approaches such as OLLIE @cite_11 , ClauseIE @cite_7 or NELL @cite_13 , focus on unstructured contents across the whole Web. In the latter, usually the schema is also not predefined, thus such approaches are called Open IE. Most state-of-the-art systems now rely on distant supervision @cite_9 @cite_4 . | {
"cite_N": [
"@cite_4",
"@cite_7",
"@cite_8",
"@cite_9",
"@cite_6",
"@cite_13",
"@cite_11"
],
"mid": [
"2107598941",
"1529731474",
"2022166150",
"1954715867",
"102708294",
"",
"2129842875"
],
"abstract": [
"Modern models of relation extraction for tasks like ACE are based on supervised learning of relations from small hand-labeled corpora. We investigate an alternative paradigm that does not require labeled corpora, avoiding the domain dependence of ACE-style algorithms, and allowing the use of corpora of any size. Our experiments use Freebase, a large semantic database of several thousand relations, to provide distant supervision. For each pair of entities that appears in some Freebase relation, we find all sentences containing those entities in a large unlabeled corpus and extract textual features to train a relation classifier. Our algorithm combines the advantages of supervised IE (combining 400,000 noisy pattern features in a probabilistic classifier) and unsupervised IE (extracting large numbers of relations from large corpora of any domain). Our model is able to extract 10,000 instances of 102 relations at a precision of 67.6 . We also analyze feature performance, showing that syntactic parse features are particularly helpful for relations that are ambiguous or lexically distant in their expression.",
"We propose ClausIE, a novel, clause-based approach to open information extraction, which extracts relations and their arguments from natural language text. ClausIE fundamentally differs from previous approaches in that it separates the detection of useful'' pieces of information expressed in a sentence from their representation in terms of extractions. In more detail, ClausIE exploits linguistic knowledge about the grammar of the English language to first detect clauses in an input sentence and to subsequently identify the type of each clause according to the grammatical function of its constituents. Based on this information, ClausIE is able to generate high-precision extractions; the representation of these extractions can be flexibly customized to the underlying application. ClausIE is based on dependency parsing and a small set of domain-independent lexica, operates sentence by sentence without any post-processing, and requires no training data (whether labeled or unlabeled). Our experimental study on various real-world datasets suggests that ClausIE obtains higher recall and higher precision than existing approaches, both on high-quality text as well as on noisy text as found in the web.",
"We present YAGO, a light-weight and extensible ontology with high coverage and quality. YAGO builds on entities and relations and currently contains more than 1 million entities and 5 million facts. This includes the Is-A hierarchy as well as non-taxonomic relations between entities (such as HASONEPRIZE). The facts have been automatically extracted from Wikipedia and unified with WordNet, using a carefully designed combination of rule-based and heuristic methods described in this paper. The resulting knowledge base is a major step beyond WordNet: in quality by adding knowledge about individuals like persons, organizations, products, etc. with their semantic relationships - and in quantity by increasing the number of facts by more than an order of magnitude. Our empirical evaluation of fact correctness shows an accuracy of about 95 . YAGO is based on a logically clean model, which is decidable, extensible, and compatible with RDFS. Finally, we show how YAGO can be further extended by state-of-the-art information extraction techniques.",
"Recently, there has been much effort in making databases for Inolecular biology more accessible osld interoperable. However, information in text. form, such as MEDLINE records, remains a greatly underutilized source of biological information. We have begun a research effort aimed at automatically mapping information from text. sources into structured representations, such as knowledge bases. Our approach to this task is to use machine-learning methods to induce routines for extracting facts from text. We describe two learning methods that we have applied to this task -a statistical text classification method, and a relational learning method -and our initial experiments in learning such information-extraction routines. We also present an approach to decreasing the cost of learning information-extraction routines by learning from \"weakly\" labeled training data.",
"DBpedia is a community effort to extract structured information from Wikipedia and to make this information available on the Web. DBpedia allows you to ask sophisticated queries against datasets derived from Wikipedia and to link other datasets on the Web to Wikipedia data. We describe the extraction of the DBpedia datasets, and how the resulting information is published on the Web for human-andmachine-consumption. We describe some emerging applications from the DBpedia community and show how website authors can facilitate DBpedia content within their sites. Finally, we present the current status of interlinking DBpedia with other open datasets on the Web and outline how DBpedia could serve as a nucleus for an emerging Web of open data.",
"",
"Open Information Extraction (IE) systems extract relational tuples from text, without requiring a pre-specified vocabulary, by identifying relation phrases and associated arguments in arbitrary sentences. However, state-of-the-art Open IE systems such as ReVerb and woe share two important weaknesses -- (1) they extract only relations that are mediated by verbs, and (2) they ignore context, thus extracting tuples that are not asserted as factual. This paper presents ollie, a substantially improved Open IE system that addresses both these limitations. First, ollie achieves high yield by extracting relations mediated by nouns, adjectives, and more. Second, a context-analysis step increases precision by including contextual information from the sentence in the extractions. ollie obtains 2.7 times the area under precision-yield curve (AUC) compared to ReVerb and 1.9 times the AUC of woeparse."
]
} |
1704.04455 | 2952535097 | Information extraction (IE) from text has largely focused on relations between individual entities, such as who has won which award. However, some facts are never fully mentioned, and no IE method has perfect recall. Thus, it is beneficial to also tap contents about the cardinalities of these relations, for example, how many awards someone has won. We introduce this novel problem of extracting cardinalities and discusses the specific challenges that set it apart from standard IE. We present a distant supervision method using conditional random fields. A preliminary evaluation results in precision between 3 and 55 , depending on the difficulty of relations. | Despite all efforts, KBs are immensely incomplete. For instance, the average number of children per person in Wikidata @cite_1 is just 0.02 @cite_15 . | {
"cite_N": [
"@cite_15",
"@cite_1"
],
"mid": [
"2513612540",
"2080133951"
],
"abstract": [
"Knowledge bases such as Wikidata, DBpedia, YAGO, or the Google Knowledge Vault collect a vast number of facts about the world. But while quite some facts are known about the world, little is known about how much is unknown. For example, while the knowledge base may tell us that Barack Obama is the father of Malia Obama and Sasha Obama, it does not tell us whether these are all of his children. This is not just an epistemic challenge, but also a practical problem for data producers and consumers. We envision that KBs become annotated with information about their recall on specific topics. We show what such annotations could look like, how they could be obtained, and survey related work.",
"This collaboratively edited knowledgebase provides a common source of data for Wikipedia, and everyone else."
]
} |
1704.04455 | 2952535097 | Information extraction (IE) from text has largely focused on relations between individual entities, such as who has won which award. However, some facts are never fully mentioned, and no IE method has perfect recall. Thus, it is beneficial to also tap contents about the cardinalities of these relations, for example, how many awards someone has won. We introduce this novel problem of extracting cardinalities and discusses the specific challenges that set it apart from standard IE. We present a distant supervision method using conditional random fields. A preliminary evaluation results in precision between 3 and 55 , depending on the difficulty of relations. | Numbers in text are an important source of information. Much work has been done on understanding numbers that express temporal information @cite_2 @cite_3 , and more recently, on numbers that express physical quantities or measures, either mentioned in text @cite_0 or in the context of web tables @cite_14 @cite_5 . | {
"cite_N": [
"@cite_14",
"@cite_3",
"@cite_0",
"@cite_2",
"@cite_5"
],
"mid": [
"2533904613",
"",
"2515878185",
"2123167824",
"2521441617"
],
"abstract": [
"HTML tables and spreadsheets on the Internet or in enterprise intranets often contain valuable information, but are created ad-hoc. As a result, they usually lack systematic names for column headers and clear vocabulary for cell values. This limits the re-use of such tables and creates a huge heterogeneity problem when comparing or aggregating multiple tables. This paper aims to overcome this problem by automatically canonicalizing header names and cell values onto concepts, classes, entities and uniquely represented quantities registered in a knowledge base. To this end, we devise a probabilistic graphical model that captures coherence dependencies between cells in tables and candidate items in the space of concepts, entities and quantities. We give specific consideration to quantities which are mapped into a \"measure, value, unit\" triple over a taxonomy of physical (e.g. power consumption), monetary (e.g. revenue), temporal (e.g. date) and dimensionless (e.g. counts) measures. Our experiments with Web tables from diverse domains demonstrate the viability of our method and its benefits over baselines.",
"",
"How much is 131 million US dollars? To help readers put such numbers in context, we propose a new task of automatically generating short descriptions known as perspectives, e.g. \"$131 million is about the cost to employ everyone in Texas over a lunch period\". First, we collect a dataset of numeric mentions in news articles, where each mention is labeled with a set of rated perspectives. We then propose a system to generate these descriptions consisting of two steps: formula construction and description generation. In construction, we compose formulae from numeric facts in a knowledge base and rank the resulting formulas based on familiarity, numeric proximity and semantic compatibility. In generation, we convert a formula into natural language using a sequence-to-sequence recurrent neural network. Our system obtains a 15.2 F1 improvement over a non-compositional baseline at formula construction and a 12.5 BLEU point improvement over a baseline description generation.",
"Research on information extraction (IE) seeks to distill relational tuples from natural language text, such as the contents of the WWW. Most IE work has focussed on identifying static facts, encoding them as binary relations. This is unfortunate, because the vast majority of facts are fluents, only holding true during an interval of time. It is less helpful to extract PresidentOf(Bill-Clinton, USA) without the temporal scope 1 20 93 - 1 20 01. This paper presents TIE, a novel, information-extraction system, which distills facts from text while inducing as much temporal information as possible. In addition to recognizing temporal relations between times and events, TIE performs global inference, enforcing transitivity to bound the start and ending times for each event. We introduce the notion of temporal entropy as a way to evaluate the performance of temporal IE systems and present experiments showing that TIE outperforms three alternative approaches.",
"With the success of Open Data a huge amount of tabular data sources became available that could potentially be mapped and linked into the Web of (Linked) Data. Most existing approaches to “semantically label” such tabular data rely on mappings of textual information to classes, properties, or instances in RDF knowledge bases in order to link – and eventually transform – tabular data into RDF. However, as we will illustrate, Open Data tables typically contain a large portion of numerical columns and or non-textual headers; therefore solutions that solely focus on textual “cues” are only partially applicable for mapping such data sources. We propose an approach to find and rank candidates of semantic labels and context descriptions for a given bag of numerical values. To this end, we apply a hierarchical clustering over information taken from DBpedia to build a background knowledge graph of possible “semantic contexts” for bags of numerical values, over which we perform a nearest neighbour search to rank the most likely candidates. Our evaluation shows that our approach can assign fine-grained semantic labels, when there is enough supporting evidence in the background knowledge graph. In other cases, our approach can nevertheless assign high level contexts to the data, which could potentially be used in combination with other approaches to narrow down the search space of possible labels."
]
} |
1704.04452 | 2605592210 | Concept maps can be used to concisely represent important information and bring structure into large document collections. Therefore, we study a variant of multi-document summarization that produces summaries in the form of concept maps. However, suitable evaluation datasets for this task are currently missing. To close this gap, we present a newly created corpus of concept maps that summarize heterogeneous collections of web documents on educational topics. It was created using a novel crowdsourcing approach that allows us to efficiently determine important elements in large document collections. We release the corpus along with a baseline system and proposed evaluation protocol to enable further research on this variant of summarization. | Other types of information representation that also model concepts and their relationships are knowledge bases, such as Freebase @cite_13 , and ontologies. However, they both differ in important aspects: Whereas concept maps follow an open label paradigm and are meant to be interpretable by humans, knowledge bases and ontologies are usually more strictly typed and made to be machine-readable. Moreover, approaches to automatically construct them from text typically try to extract as much information as possible, while we want to summarize a document. | {
"cite_N": [
"@cite_13"
],
"mid": [
"2164777277"
],
"abstract": [
"Abstract This paper presents a general statistical methodology for the analysis of multivariate categorical data arising from observer reliability studies. The procedure essentially involves the construction of functions of the observed proportions which are directed at the extent to which the observers agree among themselves and the construction of test statistics for hypotheses involving these functions. Tests for interobserver bias are presented in terms of first-order marginal homogeneity and measures of interobserver agreement are developed as generalized kappa-type statistics. These procedures are illustrated with a clinical diagnosis example from the epidemiological literature."
]
} |
1704.04451 | 2606877214 | Coreference evaluation metrics are hard to optimize directly as they are non-differentiable functions, not easily decomposable into elementary decisions. Consequently, most approaches optimize objectives only indirectly related to the end goal, resulting in suboptimal performance. Instead, we propose a differentiable relaxation that lends itself to gradient-based optimisation, thus bypassing the need for reinforcement learning or heuristic modification of cross-entropy. We show that by modifying the training objective of a competitive neural coreference system, we obtain a substantial gain in performance. This suggests that our approach can be regarded as a viable alternative to using reinforcement learning or more computationally expensive imitation learning. | Using differentiable relaxations of evaluation metrics as in our work is related to a line of research in reinforcement learning where a non-differentiable action-value function is replaced by a differentiable critic @cite_16 @cite_14 . The critic is trained so that it is as close to the true action-value function as possible. This technique is applied to machine translation @cite_21 where evaluation metrics (e.g., BLUE) are non-differentiable. A disadvantage of using critics is that there is no guarantee that the critic converges to the true evaluation metric given finite training data. In contrast, our differentiable relaxations do not need to train, and the convergence is guaranteed as @math . | {
"cite_N": [
"@cite_14",
"@cite_16",
"@cite_21"
],
"mid": [
"",
"2155027007",
"2587713090"
],
"abstract": [
"",
"Function approximation is essential to reinforcement learning, but the standard approach of approximating a value function and determining a policy from it has so far proven theoretically intractable. In this paper we explore an alternative approach in which the policy is explicitly represented by its own function approximator, independent of the value function, and is updated according to the gradient of expected reward with respect to the policy parameters. Williams's REINFORCE method and actor-critic methods are examples of this approach. Our main new result is to show that the gradient can be written in a form suitable for estimation from experience aided by an approximate action-value or advantage function. Using this result, we prove for the first time that a version of policy iteration with arbitrary differentiable function approximation is convergent to a locally optimal policy.",
"Recent research in neural machine translation has largely focused on two aspects; neural network architectures and end-to-end learning algorithms. The problem of decoding, however, has received relatively little attention from the research community. In this paper, we solely focus on the problem of decoding given a trained neural machine translation model. Instead of trying to build a new decoding algorithm for any specific decoding objective, we propose the idea of trainable decoding algorithm in which we train a decoding algorithm to find a translation that maximizes an arbitrary decoding objective. More specifically, we design an actor that observes and manipulates the hidden state of the neural machine translation decoder and propose to train it using a variant of deterministic policy gradient. We extensively evaluate the proposed algorithm using four language pairs and two decoding objectives and show that we can indeed train a trainable greedy decoder that generates a better translation (in terms of a target decoding objective) with minimal computational overhead."
]
} |
1704.03822 | 2950090724 | For machines to interact with the physical world, they must understand the physical properties of objects and materials they encounter. We use fabrics as an example of a deformable material with a rich set of mechanical properties. A thin flexible fabric, when draped, tends to look different from a heavy stiff fabric. It also feels different when touched. Using a collection of 118 fabric sample, we captured color and depth images of draped fabrics along with tactile data from a high resolution touch sensor. We then sought to associate the information from vision and touch by jointly training CNNs across the three modalities. Through the CNN, each input, regardless of the modality, generates an embedding vector that records the fabric's physical property. By comparing the embeddings, our system is able to look at a fabric image and predict how it will feel, and vice versa. We also show that a system jointly trained on vision and touch data can outperform a similar system trained only on visual data when tested purely with visual inputs. | Researchers in computer vision and graphics have been trying to track or represent fabrics or clothes, but their visual representation is difficult to obtain compared with rigid objects', and the uncertainty and complexity of the shapes and motion make the fabrics or clothes more difficult to predict. To track the exact shape of the clothes, White al @cite_11 made dense patterns on fabrics or clothes, and used multiple cameras to track their motion thus to reconstruct the 3D shapes of the clothes. Han al @cite_15 represented the cloth shape with a 2-layer model: one represents the general shape, the other one represents fold shapes, which are measured by shape-from-shading methods. Some other researches tried to represent the fabrics by physical parameters, and estimated the parameters from the visual appearance. Baht al @cite_21 used a model made of physical properties, including density, bend stiffness, stretch stiffness, damping resistance and friction, to describe and simulate clothes. They estimate the properties by comparing the real clothes' motion video with the simulated videos. Bouman al @cite_17 measured fabric properties (stiffness and density) directly from the video of fabric motion using hand-crafted features, when the fabrics were hanged and exposed to different winds. | {
"cite_N": [
"@cite_15",
"@cite_21",
"@cite_17",
"@cite_11"
],
"mid": [
"2136393262",
"2101219494",
"2172212081",
"2124527723"
],
"abstract": [
"In this paper, we present a two-level generative model for representing the images and surface depth maps of drapery and clothes. The upper level consists of a number of folds which will generate the high contrast (ridge) areas with a dictionary of shading primitives (for 2D images) and fold primitives (for 3D depth maps). These primitives are represented in parametric forms and are learned in a supervised learning phase using 3D surfaces of clothes acquired through photometric stereo. The lower level consists of the remaining flat areas which fill between the folds with a smoothness prior (Markov random field). We show that the classical ill-posed problem-shape from shading (SFS) can be much improved by this two-level model for its reduced dimensionality and incorporation of middle-level visual knowledge, i.e., the dictionary of primitives. Given an input image, we first infer the folds and compute a sketch graph using a sketch pursuit algorithm as in the primal sketch (, 2003). The 3D folds are estimated by parameter fitting using the fold dictionary and they form the \"skeleton\" of the drapery cloth surfaces. Then, the lower level is computed by conventional SFS method using the fold areas as boundary conditions. The two levels interact at the final stage by optimizing a joint Bayesian posterior probability on the depth map. We show a number of experiments which demonstrate more robust results in comparison with state-of-the-art work. In a broader scope, our representation can be viewed as a two-level inhomogeneous MRF model which is applicable to general shape-from-X problems. Our study is an attempt to revisit Marr's idea (Marr and Freeman, 1982) of computing the 2frac12D sketch from primal sketch. In a companion paper (Barbu and Zhu, 2005), we study shape from stereo based on a similar two-level generative sketch representation.",
"Cloth simulations are notoriously difficult to tune due to the many parameters that must be adjusted to achieve the look of a particular fabric. In this paper, we present an algorithm for estimating the parameters of a cloth simulation from video data of real fabric. A perceptually motivated metric based on matching between folds is used to compare video of real cloth with simulation. This metric compares two video sequences of cloth and returns a number that measures the differences in their folds. Simulated annealing is used to minimize the frame by frame error between the metric for a given simulation and the real-world footage. To estimate all the cloth parameters, we identify simple static and dynamic calibration experiments that use small swatches of the fabric. To demonstrate the power of this approach, we use our algorithm to find the parameters for four different fabrics. We show the match between the video footage and simulated motion on the calibration experiments, on new video sequences for the swatches, and on a simulation of a full skirt.",
"Passively estimating the intrinsic material properties of deformable objects moving in a natural environment is essential for scene understanding. We present a framework to automatically analyze videos of fabrics moving under various unknown wind forces, and recover two key material properties of the fabric: stiffness and area weight. We extend features previously developed to compactly represent static image textures to describe video textures, such as fabric motion. A discriminatively trained regression model is then used to predict the physical properties of fabric from these features. The success of our model is demonstrated on a new, publicly available database of fabric videos with corresponding measured ground truth material properties. We show that our predictions are well correlated with ground truth measurements of stiffness and density for the fabrics. Our contributions include: (a) a database that can be used for training and testing algorithms for passively predicting fabric properties from video, (b) an algorithm for predicting the material properties of fabric from a video, and (c) a perceptual study of humans' ability to estimate the material properties of fabric from videos and images.",
"We capture the shape of moving cloth using a custom set of color markers printed on the surface of the cloth. The output is a sequence of triangle meshes with static connectivity and with detail at the scale of individual markers in both smooth and folded regions. We compute markers' coordinates in space using correspondence across multiple synchronized video cameras. Correspondence is determined from color information in small neighborhoods and refined using a novel strain pruning process. Final correspondence does not require neighborhood information. We use a novel data driven hole-filling technique to fill occluded regions. Our results include several challenging examples: a wrinkled shirt sleeve, a dancing pair of pants, and a rag tossed onto a cup. Finally, we demonstrate that cloth capture is reusable by animating a pair of pants using human motion capture data."
]
} |
1704.03822 | 2950090724 | For machines to interact with the physical world, they must understand the physical properties of objects and materials they encounter. We use fabrics as an example of a deformable material with a rich set of mechanical properties. A thin flexible fabric, when draped, tends to look different from a heavy stiff fabric. It also feels different when touched. Using a collection of 118 fabric sample, we captured color and depth images of draped fabrics along with tactile data from a high resolution touch sensor. We then sought to associate the information from vision and touch by jointly training CNNs across the three modalities. Through the CNN, each input, regardless of the modality, generates an embedding vector that records the fabric's physical property. By comparing the embeddings, our system is able to look at a fabric image and predict how it will feel, and vice versa. We also show that a system jointly trained on vision and touch data can outperform a similar system trained only on visual data when tested purely with visual inputs. | As the best high-resolution tactile sensor, GelSight is good at measuring the fine structure of object surfaces. The tactile signal from the sensor comes in the form of image, thus some typical computer vision algorithms could be applied on processing the signal. Li and Adelson @cite_10 showed GelSight worked well on recognizing materials classes according to their texture. Moreover, GelSight also showed potentials to obtain physical properties of objects through physical interaction with the objects. Yuan al @cite_1 @cite_5 pressed GelSight on soft objects, and estimate the objects' hardness from the image sequences from GelSight. | {
"cite_N": [
"@cite_1",
"@cite_5",
"@cite_10"
],
"mid": [
"2562116405",
"",
"1965953026"
],
"abstract": [
"Hardness sensing is a valuable capability for a robot touch sensor. We describe a novel method of hardness sensing that does not require accurate control of contact conditions. A GelSight sensor is a tactile sensor that provides high resolution tactile images, which enables a robot to infer object properties such as geometry and fine texture, as well as contact force and slip conditions. The sensor is pressed on silicone samples by a human or a robot and we measure the sample hardness only with data from the sensor, without a separate force sensor and without precise knowledge of the contact trajectory. We describe the features that show object hardness. For hemispherical objects, we develop a model to measure the sample hardness, and the estimation error is about 4 in the range of 8 Shore 00 to 45 Shore A. With this technology, a robot is able to more easily infer the hardness of the touched objects, thereby improving its object recognition as well as manipulation strategy.",
"",
"Sensing surface textures by touch is a valuable capability for robots. Until recently it was difficult to build a compliant sensor with high sensitivity and high resolution. The GelSight sensor is compliant and offers sensitivity and resolution exceeding that of the human fingertips. This opens the possibility of measuring and recognizing highly detailed surface textures. The GelSight sensor, when pressed against a surface, delivers a height map. This can be treated as an image, and processed using the tools of visual texture analysis. We have devised a simple yet effective texture recognition system based on local binary patterns, and enhanced it by the use of a multi-scale pyramid and a Hellinger distance metric. We built a database with 40 classes of tactile textures using materials such as fabric, wood, and sandpaper. Our system can correctly categorize materials from this database with high accuracy. This suggests that the GelSight sensor can be useful for material recognition by robots."
]
} |
1704.03822 | 2950090724 | For machines to interact with the physical world, they must understand the physical properties of objects and materials they encounter. We use fabrics as an example of a deformable material with a rich set of mechanical properties. A thin flexible fabric, when draped, tends to look different from a heavy stiff fabric. It also feels different when touched. Using a collection of 118 fabric sample, we captured color and depth images of draped fabrics along with tactile data from a high resolution touch sensor. We then sought to associate the information from vision and touch by jointly training CNNs across the three modalities. Through the CNN, each input, regardless of the modality, generates an embedding vector that records the fabric's physical property. By comparing the embeddings, our system is able to look at a fabric image and predict how it will feel, and vice versa. We also show that a system jointly trained on vision and touch data can outperform a similar system trained only on visual data when tested purely with visual inputs. | Joint neural network is the network architecture that joins two or more separate networks for different inputs. Chopra al @cite_8 first proposed a Siamese Neural Network (SNN), that learned low dimensional embedding vectors from a single-modal input. The SNN has two identical neural networks with shared weights, and outputs the distance of embedding vectors from the two inputs. In the training, the network uses energy-based contrastive loss @cite_20 to minimize the distance of the embeddings from similar input pairs while making the distance of dissimilar input pairs' embeddings larger than margin. SNN has been applied in face verification @cite_8 and sentence embedding @cite_24 . | {
"cite_N": [
"@cite_24",
"@cite_20",
"@cite_8"
],
"mid": [
"2508865106",
"2138621090",
""
],
"abstract": [
"We present a siamese adaptation of the Long Short-Term Memory (LSTM) network for labeled data comprised of pairs of variable-length sequences. Our model is applied to assess semantic similarity between sentences, where we exceed state of the art, outperforming carefully handcrafted features and recently proposed neural network systems of greater complexity. For these applications, we provide word-embedding vectors supplemented with synonymic information to the LSTMs, which use a fixed size vector to encode the underlying meaning expressed in a sentence (irrespective of the particular wording syntax). By restricting subsequent operations to rely on a simple Manhattan metric, we compel the sentence representations learned by our model to form a highly structured space whose geometry reflects complex semantic relationships. Our results are the latest in a line of findings that showcase LSTMs as powerful language models capable of tasks requiring intricate understanding.",
"Dimensionality reduction involves mapping a set of high dimensional input points onto a low dimensional manifold so that 'similar\" points in input space are mapped to nearby points on the manifold. We present a method - called Dimensionality Reduction by Learning an Invariant Mapping (DrLIM) - for learning a globally coherent nonlinear function that maps the data evenly to the output manifold. The learning relies solely on neighborhood relationships and does not require any distancemeasure in the input space. The method can learn mappings that are invariant to certain transformations of the inputs, as is demonstrated with a number of experiments. Comparisons are made to other techniques, in particular LLE.",
""
]
} |
1704.03822 | 2950090724 | For machines to interact with the physical world, they must understand the physical properties of objects and materials they encounter. We use fabrics as an example of a deformable material with a rich set of mechanical properties. A thin flexible fabric, when draped, tends to look different from a heavy stiff fabric. It also feels different when touched. Using a collection of 118 fabric sample, we captured color and depth images of draped fabrics along with tactile data from a high resolution touch sensor. We then sought to associate the information from vision and touch by jointly training CNNs across the three modalities. Through the CNN, each input, regardless of the modality, generates an embedding vector that records the fabric's physical property. By comparing the embeddings, our system is able to look at a fabric image and predict how it will feel, and vice versa. We also show that a system jointly trained on vision and touch data can outperform a similar system trained only on visual data when tested purely with visual inputs. | In recent years, people have been using joint neural network for cross-modality learning -- mostly 2 modalities. A traditional method is to extract features from one modality and project the other modality to this feature space. Frome al @cite_2 proposed hinge rank loss to transform visual data to text. Li al @cite_23 learned the joint embedding by associating generated images to the trained embeddings from shape images of objects. Owens al @cite_4 combined CNN and LSTM to predict objects’ hitting sound from videos. They extracted the sound features first, and then regress the features from images by neural networks. Their other work @cite_22 presented a CNN that learn visual representation self-supervised by features extracted from ambient sound. | {
"cite_N": [
"@cite_4",
"@cite_22",
"@cite_23",
"@cite_2"
],
"mid": [
"2196697617",
"2951353470",
"",
"2123024445"
],
"abstract": [
"Objects make distinctive sounds when they are hit or scratched. These sounds reveal aspects of an object's material properties, as well as the actions that produced them. In this paper, we propose the task of predicting what sound an object makes when struck as a way of studying physical interactions within a visual scene. We present an algorithm that synthesizes sound from silent videos of people hitting and scratching objects with a drumstick. This algorithm uses a recurrent neural network to predict sound features from videos and then produces a waveform from these features with an example-based synthesis procedure. We show that the sounds predicted by our model are realistic enough to fool participants in a \"real or fake\" psychophysical experiment, and that they convey significant information about material properties and physical interactions.",
"The sound of crashing waves, the roar of fast-moving cars -- sound conveys important information about the objects in our surroundings. In this work, we show that ambient sounds can be used as a supervisory signal for learning visual models. To demonstrate this, we train a convolutional neural network to predict a statistical summary of the sound associated with a video frame. We show that, through this process, the network learns a representation that conveys information about objects and scenes. We evaluate this representation on several recognition tasks, finding that its performance is comparable to that of other state-of-the-art unsupervised learning methods. Finally, we show through visualizations that the network learns units that are selective to objects that are often associated with characteristic sounds.",
"",
"Modern visual recognition systems are often limited in their ability to scale to large numbers of object categories. This limitation is in part due to the increasing difficulty of acquiring sufficient training data in the form of labeled images as the number of object categories grows. One remedy is to leverage data from other sources - such as text data - both to train visual models and to constrain their predictions. In this paper we present a new deep visual-semantic embedding model trained to identify visual objects using both labeled image data as well as semantic information gleaned from unannotated text. We demonstrate that this model matches state-of-the-art performance on the 1000-class ImageNet object recognition challenge while making more semantically reasonable errors, and also show that the semantic information can be exploited to make predictions about tens of thousands of image labels not observed during training. Semantic knowledge improves such zero-shot predictions achieving hit rates of up to 18 across thousands of novel labels never seen by the visual model."
]
} |
1704.03895 | 2606059535 | Collecting fully annotated image datasets is challenging and expensive. Many types of weak supervision have been explored: weak manual annotations, web search results, temporal continuity, ambient sound and others. We focus on one particular unexplored mode: visual questions that are asked about images. The key observation that inspires our work is that the question itself provides useful information about the image (even without the answer being available). For instance, the question "what is the breed of the dog?" informs the AI that the animal in the scene is a dog and that there is only one dog present. We make three contributions: (1) providing an extensive qualitative and quantitative analysis of the information contained in human visual questions, (2) proposing two simple but surprisingly effective modifications to the standard visual question answering models that allow them to make use of weak supervision in the form of unanswered questions associated with images and (3) demonstrating that a simple data augmentation strategy inspired by our insights results in a 7.1 improvement on the standard VQA benchmark. | Vision and language Computer vision models are generally trained to recognize a fixed vocabulary of visual concepts @cite_47 @cite_61 @cite_64 . But recently, there has been a trend towards more descriptive open-world image understanding. Efforts have included works on image @cite_24 @cite_18 @cite_3 @cite_41 @cite_58 @cite_59 and video @cite_43 captioning, image segmentation from natural language expressions @cite_33 , aligning videos and books @cite_52 , zero-shot recognition from natural text @cite_56 @cite_44 , learning object models from noisy open-world human labels @cite_22 and other methods. While visual questions are certainly not meant to provide a complete description of the image, they still contain some open-world information about the scene encoded in natural text. In this work we take the initial steps towards extracting and harnessing this information. | {
"cite_N": [
"@cite_61",
"@cite_18",
"@cite_64",
"@cite_33",
"@cite_22",
"@cite_41",
"@cite_52",
"@cite_3",
"@cite_56",
"@cite_24",
"@cite_43",
"@cite_44",
"@cite_59",
"@cite_47",
"@cite_58"
],
"mid": [
"",
"2949769367",
"2031489346",
"",
"",
"",
"",
"2122180654",
"870084106",
"",
"2951183276",
"",
"",
"",
""
],
"abstract": [
"",
"This paper presents a novel approach for automatically generating image descriptions: visual detectors, language models, and multimodal similarity models learnt directly from a dataset of image captions. We use multiple instance learning to train visual detectors for words that commonly occur in captions, including many different parts of speech such as nouns, verbs, and adjectives. The word detector outputs serve as conditional inputs to a maximum-entropy language model. The language model learns from a set of over 400,000 image descriptions to capture the statistics of word usage. We capture global semantics by re-ranking caption candidates using sentence-level features and a deep multimodal similarity model. Our system is state-of-the-art on the official Microsoft COCO benchmark, producing a BLEU-4 score of 29.1 . When human judges compare the system captions to ones written by other people on our held-out test set, the system captions have equal or better quality 34 of the time.",
"The Pascal Visual Object Classes (VOC) challenge is a benchmark in visual object category recognition and detection, providing the vision and machine learning communities with a standard dataset of images and annotation, and standard evaluation procedures. Organised annually from 2005 to present, the challenge and its associated dataset has become accepted as the benchmark for object detection. This paper describes the dataset and evaluation procedure. We review the state-of-the-art in evaluated methods for both classification and detection, analyse whether the methods are statistically different, what they are learning from the images (e.g. the object or its context), and what the methods find easy or confuse. The paper concludes with lessons learnt in the three year history of the challenge, and proposes directions for future improvement and extension.",
"",
"",
"",
"",
"In this paper we explore the bi-directional mapping between images and their sentence-based descriptions. We propose learning this mapping using a recurrent neural network. Unlike previous approaches that map both sentences and images to a common embedding, we enable the generation of novel sentences given an image. Using the same model, we can also reconstruct the visual features associated with an image given its visual description. We use a novel recurrent visual memory that automatically learns to remember long-term visual concepts to aid in both sentence generation and visual feature reconstruction. 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 preferred by humans over @math 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.",
"One of the main challenges in Zero-Shot Learning of visual categories is gathering semantic attributes to accompany images. Recent work has shown that learning from textual descriptions, such as Wikipedia articles, avoids the problem of having to explicitly define these attributes. We present a new model that can classify unseen categories from their textual description. Specifically, we use text features to predict the output weights of both the convolutional and the fully connected layers in a deep convolutional neural network (CNN). We take advantage of the architecture of CNNs and learn features at different layers, rather than just learning an embedding space for both modalities, as is common with existing approaches. The proposed model also allows us to automatically generate a list of pseudo- attributes for each visual category consisting of words from Wikipedia articles. We train our models end-to-end us- ing the Caltech-UCSD bird and flower datasets and evaluate both ROC and Precision-Recall curves. Our empirical results show that the proposed model significantly outperforms previous methods.",
"",
"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.",
"",
"",
"",
""
]
} |
1704.04054 | 2605580863 | We propose a new saliency-guided method for generating supervoxels in 3D space. Rather than using an evenly distributed spatial seeding procedure, our method uses visual saliency to guide the process of supervoxel generation. This results in densely distributed, small, and precise supervoxels in salient regions which often contain objects, and larger supervoxels in less salient regions that often correspond to background. Our approach largely improves the quality of the resulting supervoxel segmentation in terms of boundary recall and under-segmentation error on publicly available benchmarks. | Following the classification of @cite_8 , superpixel methods can be classified into watershed-based @cite_15 , density-based @cite_19 , graph-based @cite_11 , contour evolution @cite_25 , path-based @cite_18 , clustering-based @cite_6 @cite_12 and energy optimization methods @cite_1 @cite_13 . Among the most popular methods is the Simple Linear Iterative Clustering (SLIC) superpixel method @cite_6 that applies local @math -means clusterings of the image pixels on a regular grid pattern over the entire image to generate perceptually uniform superpixels. SLIC is appealing since it is simple, fast, and has only two parameters that must be controlled: the number of superpixels that shall be generated and the compactness of the superpixels. Many other superpixel and supervoxel methods follow the idea of SLIC, e.g., @cite_10 @cite_12 , and @cite_17 . | {
"cite_N": [
"@cite_18",
"@cite_8",
"@cite_10",
"@cite_1",
"@cite_17",
"@cite_6",
"@cite_19",
"@cite_15",
"@cite_13",
"@cite_25",
"@cite_12",
"@cite_11"
],
"mid": [
"2149233894",
"2585293115",
"2135249503",
"1994938019",
"2155033583",
"2118246710",
"1508404128",
"",
"1950112384",
"2119531662",
"",
"1999478155"
],
"abstract": [
"Most existing superpixel algorithms ignore the spatial structure and regularity properties, which result in undesirable sizes and location relationships for the subsequent processing. In this paper, we introduce a new method to generate the regularity preserved superpixels. Starting from the lattice seeds, our method relocates them to the pixel with locally maximal edge magnitudes and treats them as the superpixel junctions. Then, the shortest path algorithm is employed to find the local optimal boundary connecting each adjacent junction pair. Thanks to the local constraints, our method obtains homogeneous superpixels with adjacency in lowly textured and uniform regions and simultaneously preserves the boundary adherence in the high contrast contents. Our method preserves the regularity property without significantly sacrificing the segmentation accuracy. Moreover, we extend this regular constraint for generating the supervoxels. Our method obtains the regular supervoxels, which preserves the structural relation on both spatial and temporal spaces of the video. Quantitative and qualitative experimental results on benchmark datasets demonstrate that our simple but effective method outperforms the existing regular superpixel methods.",
"Abstract Superpixels group perceptually similar pixels to create visually meaningful entities while heavily reducing the number of primitives for subsequent processing steps. As of these properties, superpixel algorithms have received much attention since their naming in 2003 (Ren and Malik, 2003). By today, publicly available superpixel algorithms have turned into standard tools in low-level vision. As such, and due to their quick adoption in a wide range of applications, appropriate benchmarks are crucial for algorithm selection and comparison. Until now, the rapidly growing number of algorithms as well as varying experimental setups hindered the development of a unifying benchmark. We present a comprehensive evaluation of 28 state-of-the-art superpixel algorithms utilizing a benchmark focussing on fair comparison and designed to provide new insights relevant for applications. To this end, we explicitly discuss parameter optimization and the importance of strictly enforcing connectivity. Furthermore, by extending well-known metrics, we are able to summarize algorithm performance independent of the number of generated superpixels, thereby overcoming a major limitation of available benchmarks. Furthermore, we discuss runtime, robustness against noise, blur and affine transformations, implementation details as well as aspects of visual quality. Finally, we present an overall ranking of superpixel algorithms which redefines the state-of-the-art and enables researchers to easily select appropriate algorithms and the corresponding implementations which themselves are made publicly available as part of our benchmark at http: www.davidstutz.de projects superpixel-benchmark .",
"Unsupervised over-segmentation of an image into regions of perceptually similar pixels, known as super pixels, is a widely used preprocessing step in segmentation algorithms. Super pixel methods reduce the number of regions that must be considered later by more computationally expensive algorithms, with a minimal loss of information. Nevertheless, as some information is inevitably lost, it is vital that super pixels not cross object boundaries, as such errors will propagate through later steps. Existing methods make use of projected color or depth information, but do not consider three dimensional geometric relationships between observed data points which can be used to prevent super pixels from crossing regions of empty space. We propose a novel over-segmentation algorithm which uses voxel relationships to produce over-segmentations which are fully consistent with the spatial geometry of the scene in three dimensional, rather than projective, space. Enforcing the constraint that segmented regions must have spatial connectivity prevents label flow across semantic object boundaries which might otherwise be violated. Additionally, as the algorithm works directly in 3D space, observations from several calibrated RGB+D cameras can be segmented jointly. Experiments on a large data set of human annotated RGB+D images demonstrate a significant reduction in occurrence of clusters crossing object boundaries, while maintaining speeds comparable to state-of-the-art 2D methods.",
"Superpixel algorithms aim to over-segment the image by grouping pixels that belong to the same object. Many state-of-the-art superpixel algorithms rely on minimizing objective functions to enforce color homogeneity. The optimization is accomplished by sophisticated methods that progressively build the superpixels, typically by adding cuts or growing superpixels. As a result, they are computationally too expensive for real-time applications. We introduce a new approach based on a simple hill-climbing optimization. Starting from an initial superpixel partitioning, it continuously refines the superpixels by modifying the boundaries. We define a robust and fast to evaluate energy function, based on enforcing color similarity between the boundaries and the superpixel color histogram. In a series of experiments, we show that we achieve an excellent compromise between accuracy and efficiency. We are able to achieve a performance comparable to the state-of-the-art, but in real-time on a single Intel i7 CPU at 2.8GHz.",
"With the advances of depth sensing technologies, color image plus depth information (referred to as RGB-D data hereafter) is more and more popular for comprehensive description of 3-D scenes. This paper proposes a two-stage segmentation method for RGB-D data: 1) oversegmentation by 3-D geometry enhanced superpixels and 2) graph-based merging with label cost from superpixels. In the oversegmentation stage, 3-D geometrical information is reconstructed from the depth map. Then, a K-means-like clustering method is applied to the RGB-D data for oversegmentation using an 8-D distance metric constructed from both color and 3-D geometrical information. In the merging stage, treating each superpixel as a node, a graph-based model is set up to relabel the superpixels into semantically-coherent segments. In the graph-based model, RGB-D proximity, texture similarity, and boundary continuity are incorporated into the smoothness term to exploit the correlations of neighboring superpixels. To obtain a compact labeling, the label term is designed to penalize labels linking to similar superpixels that likely belong to the same object. Both the proposed 3-D geometry enhanced superpixel clustering method and the graph-based merging method from superpixels are evaluated by qualitative and quantitative results. By the fusion of color and depth information, the proposed method achieves superior segmentation performance over several state-of-the-art algorithms.",
"Computer vision applications have come to rely increasingly on superpixels in recent years, but it is not always clear what constitutes a good superpixel algorithm. In an effort to understand the benefits and drawbacks of existing methods, we empirically compare five state-of-the-art superpixel algorithms for their ability to adhere to image boundaries, speed, memory efficiency, and their impact on segmentation performance. We then introduce a new superpixel algorithm, simple linear iterative clustering (SLIC), which adapts a k-means clustering approach to efficiently generate superpixels. Despite its simplicity, SLIC adheres to boundaries as well as or better than previous methods. At the same time, it is faster and more memory efficient, improves segmentation performance, and is straightforward to extend to supervoxel generation.",
"We show that the complexity of the recently introduced medoid-shift algorithm in clustering N points is O(N 2), with a small constant, if the underlying distance is Euclidean. This makes medoid shift considerably faster than mean shift, contrarily to what previously believed. We then exploit kernel methods to extend both mean shift and the improved medoid shift to a large family of distances, with complexity bounded by the effective rank of the resulting kernel matrix, and with explicit regularization constraints. Finally, we show that, under certain conditions, medoid shift fails to cluster data points belonging to the same mode, resulting in over-fragmentation. We propose remedies for this problem, by introducing a novel, simple and extremely efficient clustering algorithm, called quick shift, that explicitly trades off under- and over-fragmentation. Like medoid shift, quick shift operates in non-Euclidean spaces in a straightforward manner. We also show that the accelerated medoid shift can be used to initialize mean shift for increased efficiency. We illustrate our algorithms to clustering data on manifolds, image segmentation, and the automatic discovery of visual categories.",
"",
"In this paper, we tackle the problem of unsupervised segmentation in the form of superpixels. Our main emphasis is on speed and accuracy. We build on [31] to define the problem as a boundary and topology preserving Markov random field. We propose a coarse to fine optimization technique that speeds up inference in terms of the number of updates by an order of magnitude. Our approach is shown to outperform [31] while employing a single iteration. We evaluate and compare our approach to state-of-the-art superpixel algorithms on the BSD and KITTI benchmarks. Our approach significantly outperforms the baselines in the segmentation metrics and achieves the lowest error on the stereo task.",
"We describe a geometric-flow-based algorithm for computing a dense oversegmentation of an image, often referred to as superpixels. It produces segments that, on one hand, respect local image boundaries, while, on the other hand, limiting undersegmentation through a compactness constraint. It is very fast, with complexity that is approximately linear in image size, and can be applied to megapixel sized images with high superpixel densities in a matter of minutes. We show qualitative demonstrations of high-quality results on several complex images. The Berkeley database is used to quantitatively compare its performance to a number of oversegmentation algorithms, showing that it yields less undersegmentation than algorithms that lack a compactness constraint while offering a significant speedup over N-cuts, which does enforce compactness.",
"",
"This paper addresses the problem of segmenting an image into regions. We define a predicate for measuring the evidence for a boundary between two regions using a graph-based representation of the image. We then develop an efficient segmentation algorithm based on this predicate, and show that although this algorithm makes greedy decisions it produces segmentations that satisfy global properties. We apply the algorithm to image segmentation using two different kinds of local neighborhoods in constructing the graph, and illustrate the results with both real and synthetic images. The algorithm runs in time nearly linear in the number of graph edges and is also fast in practice. An important characteristic of the method is its ability to preserve detail in low-variability image regions while ignoring detail in high-variability regions."
]
} |
1704.04054 | 2605580863 | We propose a new saliency-guided method for generating supervoxels in 3D space. Rather than using an evenly distributed spatial seeding procedure, our method uses visual saliency to guide the process of supervoxel generation. This results in densely distributed, small, and precise supervoxels in salient regions which often contain objects, and larger supervoxels in less salient regions that often correspond to background. Our approach largely improves the quality of the resulting supervoxel segmentation in terms of boundary recall and under-segmentation error on publicly available benchmarks. | Instead, Voxel Cloud Connectivity Segmentation (VCCS) @cite_10 generates supervoxels from RGB-D data, while guaranteeing that all voxels within each supervoxel are spatially connected. Similarly to SLIC and DASP, VCCS is also a variant of iterative local clustering applied on a regular lattice. However, VCCS also exploits the 3D geometry of the scene and generates a full 3D supervoxel graph. | {
"cite_N": [
"@cite_10"
],
"mid": [
"2135249503"
],
"abstract": [
"Unsupervised over-segmentation of an image into regions of perceptually similar pixels, known as super pixels, is a widely used preprocessing step in segmentation algorithms. Super pixel methods reduce the number of regions that must be considered later by more computationally expensive algorithms, with a minimal loss of information. Nevertheless, as some information is inevitably lost, it is vital that super pixels not cross object boundaries, as such errors will propagate through later steps. Existing methods make use of projected color or depth information, but do not consider three dimensional geometric relationships between observed data points which can be used to prevent super pixels from crossing regions of empty space. We propose a novel over-segmentation algorithm which uses voxel relationships to produce over-segmentations which are fully consistent with the spatial geometry of the scene in three dimensional, rather than projective, space. Enforcing the constraint that segmented regions must have spatial connectivity prevents label flow across semantic object boundaries which might otherwise be violated. Additionally, as the algorithm works directly in 3D space, observations from several calibrated RGB+D cameras can be segmented jointly. Experiments on a large data set of human annotated RGB+D images demonstrate a significant reduction in occurrence of clusters crossing object boundaries, while maintaining speeds comparable to state-of-the-art 2D methods."
]
} |
1704.04054 | 2605580863 | We propose a new saliency-guided method for generating supervoxels in 3D space. Rather than using an evenly distributed spatial seeding procedure, our method uses visual saliency to guide the process of supervoxel generation. This results in densely distributed, small, and precise supervoxels in salient regions which often contain objects, and larger supervoxels in less salient regions that often correspond to background. Our approach largely improves the quality of the resulting supervoxel segmentation in terms of boundary recall and under-segmentation error on publicly available benchmarks. | It should be mentioned that in the literature other methods are also referred to as supervoxels, which use time instead of space as the third dimension and are thus a type of video segmentation @cite_18 @cite_4 . These methods are not applicable to RGB-D point clouds since they usually require a solid 3D volume of time and space as input. Thus, such methods are not comparable with the supervoxel methods operating on 3D data from a sensor such as the Microsoft Kinect. | {
"cite_N": [
"@cite_18",
"@cite_4"
],
"mid": [
"2149233894",
"2081432165"
],
"abstract": [
"Most existing superpixel algorithms ignore the spatial structure and regularity properties, which result in undesirable sizes and location relationships for the subsequent processing. In this paper, we introduce a new method to generate the regularity preserved superpixels. Starting from the lattice seeds, our method relocates them to the pixel with locally maximal edge magnitudes and treats them as the superpixel junctions. Then, the shortest path algorithm is employed to find the local optimal boundary connecting each adjacent junction pair. Thanks to the local constraints, our method obtains homogeneous superpixels with adjacency in lowly textured and uniform regions and simultaneously preserves the boundary adherence in the high contrast contents. Our method preserves the regularity property without significantly sacrificing the segmentation accuracy. Moreover, we extend this regular constraint for generating the supervoxels. Our method obtains the regular supervoxels, which preserves the structural relation on both spatial and temporal spaces of the video. Quantitative and qualitative experimental results on benchmark datasets demonstrate that our simple but effective method outperforms the existing regular superpixel methods.",
"Supervoxel segmentation has strong potential to be incorporated into early video analysis as superpixel segmentation has in image analysis. However, there are many plausible supervoxel methods and little understanding as to when and where each is most appropriate. Indeed, we are not aware of a single comparative study on supervoxel segmentation. To that end, we study five supervoxel algorithms in the context of what we consider to be a good supervoxel: namely, spatiotemporal uniformity, object region boundary detection, region compression and parsimony. For the evaluation we propose a comprehensive suite of 3D volumetric quality metrics to measure these desirable supervoxel characteristics. We use three benchmark video data sets with a variety of content-types and varying amounts of human annotations. Our findings have led us to conclusive evidence that the hierarchical graph-based and segmentation by weighted aggregation methods perform best and almost equally-well on nearly all the metrics and are the methods of choice given our proposed assumptions."
]
} |
1704.03809 | 2606052883 | We present a new model for singing synthesis based on a modified version of the WaveNet architecture. Instead of modeling raw waveform, we model features produced by a parametric vocoder that separates the influence of pitch and timbre. This allows conveniently modifying pitch to match any target melody, facilitates training on more modest dataset sizes, and significantly reduces training and generation times. Our model makes frame-wise predictions using mixture density outputs rather than categorical outputs in order to reduce the required parameter count. As we found overfitting to be an issue with the relatively small datasets used in our experiments, we propose a method to regularize the model and make the autoregressive generation process more robust to prediction errors. Using a simple multi-stream architecture, harmonic, aperiodic and voiced unvoiced components can all be predicted in a coherent manner. We compare our method to existing parametric statistical and state-of-the-art concatenative methods using quantitative metrics and a listening test. While naive implementations of the autoregressive generation algorithm tend to be inefficient, using a smart algorithm we can greatly speed up the process and obtain a system that's competitive in both speed and quality. | More traditional neural parametric speech synthesizers tend to be based on feed-forward architectures such as DNNs and Mixture Density Networks (MDNs) @cite_7 , or on recurrent architectures such as Long Short-Term Memory RNNs (LSTM-RNNs) @cite_10 . Feed-forward networks learn a frame-wise mapping between linguistic and acoustic features, thus potentially producing discontinuous output. This is often partly mitigated by predicting static, delta and delta-delta feature distributions combined with a parameter generation algorithm that maximizes output probability @cite_25 . Recurrent architectures avoid this issue by propagating hidden states (and sometimes the output state) over time. In contrast, autoregressive architectures like the one we propose make predictions based on predicted past acoustic features, allowing, among other things, to better model rapid modulations such as plosive and trill consonants. | {
"cite_N": [
"@cite_10",
"@cite_25",
"@cite_7"
],
"mid": [
"1576227399",
"",
"1990505856"
],
"abstract": [
"Long short-term memory recurrent neural networks (LSTM-RNNs) have been applied to various speech applications including acoustic modeling for statistical parametric speech synthesis. One of the concerns for applying them to text-to-speech applications is its effect on latency. To address this concern, this paper proposes a low-latency, streaming speech synthesis architecture using unidirectional LSTM-RNNs with a recurrent output layer. The use of unidirectional RNN architecture allows frame-synchronous streaming inference of output acoustic features given input linguistic features. The recurrent output layer further encourages smooth transition between acoustic features at consecutive frames. Experimental results in subjective listening tests show that the proposed architecture can synthesize natural sounding speech without requiring utterance-level batch processing.",
"",
"Statistical parametric speech synthesis (SPSS) using deep neural networks (DNNs) has shown its potential to produce naturally-sounding synthesized speech. However, there are limitations in the current implementation of DNN-based acoustic modeling for speech synthesis, such as the unimodal nature of its objective function and its lack of ability to predict variances. To address these limitations, this paper investigates the use of a mixture density output layer. It can estimate full probability density functions over real-valued output features conditioned on the corresponding input features. Experimental results in objective and subjective evaluations show that the use of the mixture density output layer improves the prediction accuracy of acoustic features and the naturalness of the synthesized speech."
]
} |
1704.03949 | 2606996976 | We explore how the availability of a sensemaking tool influences users' knowledge externalization strategies. On a large display, users were asked to solve an intelligence analysis task with or without a bidirectionally linked concept-graph (BLC) to organize insights into concepts (nodes) and relations (edges). In BLC, both nodes and edges maintain "deep links" to the exact source phrases and sections in associated documents. In our control condition, we were able to reproduce previously described spatial organization behaviors using document windows on the large display. When using BLC, however, we found that analysts apply spatial organization to BLC nodes instead, use significantly less display space and have significantly fewer open windows. | Text analytics tools, such as @cite_9 or 's @cite_15 , categorize text entities and present them as graphs, scatter plots, or fairly free-form shoeboxes'', in which the connections between entities are made explicit by color coding and visual links. Mind mapping tools ( @cite_14 ) enable users not only to add text nodes, but also images and document links to the mind map for externalizing their mental concepts. | {
"cite_N": [
"@cite_9",
"@cite_14",
"@cite_15"
],
"mid": [
"2165662062",
"2138538892",
""
],
"abstract": [
"Investigative analysts who work with collections of text documents connect embedded threads of evidence in order to formulate hypotheses about plans and activities of potential interest. As the number of documents and the corresponding number of concepts and entities within the documents grow larger, sense-making processes become more and more difficult for the analysts. We have developed a visual analytic system called Jigsaw that represents documents and their entities visually in order to help analysts examine reports more efficiently and develop theories about potential actions more quickly. Jigsaw provides multiple coordinated views of document entities with a special emphasis on visually illustrating connections between entities across the different documents.",
"The Visual Understanding Environment (VUE) project at Tufts' Academic Technology department provides faculty and students with tools to successfully integrate digital resources into their teaching and learning. VUE provides a visual environment for structuring, presenting, and sharing digital information and an OKI-compliant software bridge for connecting to FEDORA-based digital repositories. Using VUE's concept mapping interface, faculty and students design customized semantic networks of digital resources drawing from digital libraries, local files and the Web. The resulting content maps can then be viewed and exchanged online.",
""
]
} |
1704.03949 | 2606996976 | We explore how the availability of a sensemaking tool influences users' knowledge externalization strategies. On a large display, users were asked to solve an intelligence analysis task with or without a bidirectionally linked concept-graph (BLC) to organize insights into concepts (nodes) and relations (edges). In BLC, both nodes and edges maintain "deep links" to the exact source phrases and sections in associated documents. In our control condition, we were able to reproduce previously described spatial organization behaviors using document windows on the large display. When using BLC, however, we found that analysts apply spatial organization to BLC nodes instead, use significantly less display space and have significantly fewer open windows. | @cite_10 investigated collaborative sensemaking on a large display using either Jigsaw @cite_9 or a simple document viewer with highlighting and annotation. They observed that users had fewer documents open with Jigsaw compared to the document viewer, but speculated that this difference was caused by the different window management behaviors of the two sensemaking tools. However, an alternative explanation could be that the users employed different externalization strategies in Jigsaw, so that the actual for multiple document windows was reduced. | {
"cite_N": [
"@cite_9",
"@cite_10"
],
"mid": [
"2165662062",
"2158062194"
],
"abstract": [
"Investigative analysts who work with collections of text documents connect embedded threads of evidence in order to formulate hypotheses about plans and activities of potential interest. As the number of documents and the corresponding number of concepts and entities within the documents grow larger, sense-making processes become more and more difficult for the analysts. We have developed a visual analytic system called Jigsaw that represents documents and their entities visually in order to help analysts examine reports more efficiently and develop theories about potential actions more quickly. Jigsaw provides multiple coordinated views of document entities with a special emphasis on visually illustrating connections between entities across the different documents.",
"From an exploratory user study using a fictional textual intelligence analysis task on a large, high-resolution vertical display, we investigated how pairs of users interact with the display to construct spatial schemas and externalize information, as well as how they establish shared and private territories. We investigated users' space management strategies depending on the design philosophy of the user interface (visualization- or document-centric). We classified the types of territorial behavior exhibited in terms of how the users interacted with information on the display (integrated or independent workspaces). Next, we examined how territorial behavior impacted the common ground between the pairs of users. Finally, we offer design suggestions for building future co-located collaborative visual analytics tools for use on large, high-resolution vertical displays."
]
} |
1704.03949 | 2606996976 | We explore how the availability of a sensemaking tool influences users' knowledge externalization strategies. On a large display, users were asked to solve an intelligence analysis task with or without a bidirectionally linked concept-graph (BLC) to organize insights into concepts (nodes) and relations (edges). In BLC, both nodes and edges maintain "deep links" to the exact source phrases and sections in associated documents. In our control condition, we were able to reproduce previously described spatial organization behaviors using document windows on the large display. When using BLC, however, we found that analysts apply spatial organization to BLC nodes instead, use significantly less display space and have significantly fewer open windows. | In this work, we investigate the influence of sensemaking tools on knowledge externalization strategies using a lightweight graph-based tool, the ( ). supports sensemaking with arbitrary online information sources and allows users to externalize their knowledge through a graph. It combines features of @cite_7 (attaching lists of document references to nodes and edges) and @cite_0 (referencing specific websites or passages within), and adds window layout and visual cueing capabilities that make attractive for use on large displays. In particular, we are interested in understanding users' spatial organization strategies and | {
"cite_N": [
"@cite_0",
"@cite_7"
],
"mid": [
"2168992605",
"2054054049"
],
"abstract": [
"The World Wide Web is a powerful platform for a wide range of information tasks. Dramatic advances in technology, such as improved search capabilities and the AJAX application model, have enabled entirely new web-based applications and usage patterns, making many tasks easier to perform than ever before. However, few tools have been developed to assist with sensemaking tasks: complex research behaviors in which users gather and comprehend information from many sources to answer potentially vague, non-procedural questions. Sensemaking tasks are common and include, for example, researching vacation destinations or deciding how to invest. This paper presents the ScratchPad, an extension to the standard browser interface that is designed to capture, organize, and exploit the information discovered while performing a sensemaking task.",
"When people work together to analyze a data set, they need to organize their findings, hypotheses, and evidence, share that information with their collaborators, and coordinate activities amongst team members. Sharing externalizations (recorded information such as notes) could increase awareness and assist with team communication and coordination. However, we currently know little about how to provide tool support for this sort of sharing. We explore how linked common work (LCW) can be employed within a collaborative thinking space', to facilitate synchronous collaborative sensemaking activities in Visual Analytics (VA). Collaborative thinking spaces provide an environment for analysts to record, organize, share and connect externalizations. Our tool, CLIP, extends earlier thinking spaces by integrating LCW features that reveal relationships between collaborators' findings. We conducted a user study comparing CLIP to a baseline version without LCW. Results demonstrated that LCW significantly improved analytic outcomes at a collaborative intelligence task. Groups using CLIP were also able to more effectively coordinate their work, and held more discussion of their findings and hypotheses. LCW enabled them to maintain awareness of each other's activities and findings and link those findings to their own work, preventing disruptive oral awareness notifications."
]
} |
1704.03952 | 2606508169 | Reinforcement learning is considered as a promising direction for driving policy learning. However, training autonomous driving vehicle with reinforcement learning in real environment involves non-affordable trial-and-error. It is more desirable to first train in a virtual environment and then transfer to the real environment. In this paper, we propose a novel realistic translation network to make model trained in virtual environment be workable in real world. The proposed network can convert non-realistic virtual image input into a realistic one with similar scene structure. Given realistic frames as input, driving policy trained by reinforcement learning can nicely adapt to real world driving. Experiments show that our proposed virtual to real (VR) reinforcement learning (RL) works pretty well. To our knowledge, this is the first successful case of driving policy trained by reinforcement learning that can adapt to real world driving data. | . Supervised learning methods are obviously straightforward ways to train autonomous vehicles. ALVINN @cite_15 provides an early example of using neural network for autonomous driving. Their model is simple and direct, which maps image inputs to action predictions with a shallow network. Powered by deep learning especially a convolutional neural network, NVIDIA @cite_17 recently provides an attempt to leverage driving video data for simple lane following task. Another work by @cite_8 learns a mapping between input images to a number of key perception indicators, which are closely related to the affordance of a driving state. However, the learned affordance must be associated with actions through hand-engineered rules. These supervised methods work relatively well in simple tasks such as lane-following and driving on a highway. On the other hand, imitation learning can also be regarded as supervised learning approach @cite_34 , where the agent observes the demonstrations performed by some experts and learns to imitate the actions of the experts. However, an intrinsic shortcoming of imitation learning is that it has the covariate shift problem @cite_6 and can not generalize very well to scenes never experienced before. | {
"cite_N": [
"@cite_8",
"@cite_6",
"@cite_15",
"@cite_34",
"@cite_17"
],
"mid": [
"2953248129",
"",
"2167224731",
"2396217537",
"2342840547"
],
"abstract": [
"Today, there are two major paradigms for vision-based autonomous driving systems: mediated perception approaches that parse an entire scene to make a driving decision, and behavior reflex approaches that directly map an input image to a driving action by a regressor. In this paper, we propose a third paradigm: a direct perception approach to estimate the affordance for driving. We propose to map an input image to a small number of key perception indicators that directly relate to the affordance of a road traffic state for driving. Our representation provides a set of compact yet complete descriptions of the scene to enable a simple controller to drive autonomously. Falling in between the two extremes of mediated perception and behavior reflex, we argue that our direct perception representation provides the right level of abstraction. To demonstrate this, we train a deep Convolutional Neural Network using recording from 12 hours of human driving in a video game and show that our model can work well to drive a car in a very diverse set of virtual environments. We also train a model for car distance estimation on the KITTI dataset. Results show that our direct perception approach can generalize well to real driving images. Source code and data are available on our project website.",
"",
"ALVINN (Autonomous Land Vehicle In a Neural Network) is a 3-layer back-propagation network designed for the task of road following. Currently ALVINN takes images from a camera and a laser range finder as input and produces as output the direction the vehicle should travel in order to follow the road. Training has been conducted using simulated road images. Successful tests on the Carnegie Mellon autonomous navigation test vehicle indicate that the network can effectively follow real roads under certain field conditions. The representation developed to perform the task differs dramatically when the network is trained under various conditions, suggesting the possibility of a novel adaptive autonomous navigation system capable of tailoring its processing to the conditions at hand.",
"One way to approach end-to-end autonomous driving is to learn a policy function that maps from a sensory input, such as an image frame from a front-facing camera, to a driving action, by imitating an expert driver, or a reference policy. This can be done by supervised learning, where a policy function is tuned to minimize the difference between the predicted and ground-truth actions. A policy function trained in this way however is known to suffer from unexpected behaviours due to the mismatch between the states reachable by the reference policy and trained policy functions. More advanced algorithms for imitation learning, such as DAgger, addresses this issue by iteratively collecting training examples from both reference and trained policies. These algorithms often requires a large number of queries to a reference policy, which is undesirable as the reference policy is often expensive. In this paper, we propose an extension of the DAgger, called SafeDAgger, that is query-efficient and more suitable for end-to-end autonomous driving. We evaluate the proposed SafeDAgger in a car racing simulator and show that it indeed requires less queries to a reference policy. We observe a significant speed up in convergence, which we conjecture to be due to the effect of automated curriculum learning.",
"We trained a convolutional neural network (CNN) to map raw pixels from a single front-facing camera directly to steering commands. This end-to-end approach proved surprisingly powerful. With minimum training data from humans the system learns to drive in traffic on local roads with or without lane markings and on highways. It also operates in areas with unclear visual guidance such as in parking lots and on unpaved roads. The system automatically learns internal representations of the necessary processing steps such as detecting useful road features with only the human steering angle as the training signal. We never explicitly trained it to detect, for example, the outline of roads. Compared to explicit decomposition of the problem, such as lane marking detection, path planning, and control, our end-to-end system optimizes all processing steps simultaneously. We argue that this will eventually lead to better performance and smaller systems. Better performance will result because the internal components self-optimize to maximize overall system performance, instead of optimizing human-selected intermediate criteria, e.g., lane detection. Such criteria understandably are selected for ease of human interpretation which doesn't automatically guarantee maximum system performance. Smaller networks are possible because the system learns to solve the problem with the minimal number of processing steps. We used an NVIDIA DevBox and Torch 7 for training and an NVIDIA DRIVE(TM) PX self-driving car computer also running Torch 7 for determining where to drive. The system operates at 30 frames per second (FPS)."
]
} |
1704.03952 | 2606508169 | Reinforcement learning is considered as a promising direction for driving policy learning. However, training autonomous driving vehicle with reinforcement learning in real environment involves non-affordable trial-and-error. It is more desirable to first train in a virtual environment and then transfer to the real environment. In this paper, we propose a novel realistic translation network to make model trained in virtual environment be workable in real world. The proposed network can convert non-realistic virtual image input into a realistic one with similar scene structure. Given realistic frames as input, driving policy trained by reinforcement learning can nicely adapt to real world driving. Experiments show that our proposed virtual to real (VR) reinforcement learning (RL) works pretty well. To our knowledge, this is the first successful case of driving policy trained by reinforcement learning that can adapt to real world driving data. | . Reinforcement learning has been applied to a wide variety of robotics related tasks, such as computer games @cite_3 , robot locomotion @cite_9 @cite_31 , and autonomous driving @cite_18 @cite_22 . One of the challenges in practical real-world applications of reinforcement learning is the high-dimensionality of state space as well as the non-trivial large action range. Developing an optimal policy over such high-complexity space is time consuming. Recent work in deep reinforcement learning has made great progress in learning in a high dimensional space with the power of deep neural networks @cite_27 @cite_19 @cite_10 @cite_12 @cite_29 . However, both deep Q-learning method @cite_3 and policy gradient method @cite_12 require the agent to interact with the environment to get reward and feedback. However, it is unrealistic to train autonomous vehicle with reinforcement learning in a real world environment since the car may hurt its surroundings once it takes a wrong action. | {
"cite_N": [
"@cite_18",
"@cite_22",
"@cite_9",
"@cite_29",
"@cite_3",
"@cite_19",
"@cite_27",
"@cite_31",
"@cite_10",
"@cite_12"
],
"mid": [
"2286929393",
"",
"2139053308",
"2260756217",
"2145339207",
"",
"2038794597",
"2051620263",
"",
"2173248099"
],
"abstract": [
"State-of-the-art models for semantic segmentation are based on adaptations of convolutional networks that had originally been designed for image classification. However, dense prediction and image classification are structurally different. In this work, we develop a new convolutional network module that is specifically designed for dense prediction. The presented module uses dilated convolutions to systematically aggregate multi-scale contextual information without losing resolution. The architecture is based on the fact that dilated convolutions support exponential expansion of the receptive field without loss of resolution or coverage. We show that the presented context module increases the accuracy of state-of-the-art semantic segmentation systems. In addition, we examine the adaptation of image classification networks to dense prediction and show that simplifying the adapted network can increase accuracy.",
"",
"This paper presents a machine learning approach to optimizing a quadrupedal trot gait for forward speed. Given a parameterized walk designed for a specific robot, we propose using a form of policy gradient reinforcement learning to automatically search the set of possible parameters with the goal of finding the fastest possible walk. We implement and test our approach on a commercially available quadrupedal robot platform, namely the Sony Aibo robot. After about three hours of learning, all on the physical robots and with no human intervention other than to change the batteries, the robots achieved a gait faster than any previously known gait known for the Aibo, significantly outperforming a variety of existing hand-coded and learned solutions.",
"We propose a conceptually simple and lightweight framework for deep reinforcement learning that uses asynchronous gradient descent for optimization of deep neural network controllers. We present asynchronous variants of four standard reinforcement learning algorithms and show that parallel actor-learners have a stabilizing effect on training allowing all four methods to successfully train neural network controllers. The best performing method, an asynchronous variant of actor-critic, surpasses the current state-of-the-art on the Atari domain while training for half the time on a single multi-core CPU instead of a GPU. Furthermore, we show that asynchronous actor-critic succeeds on a wide variety of continuous motor control problems as well as on a new task of navigating random 3D mazes using a visual input.",
"An artificial agent is developed that learns to play a diverse range of classic Atari 2600 computer games directly from sensory experience, achieving a performance comparable to that of an expert human player; this work paves the way to building general-purpose learning algorithms that bridge the divide between perception and action.",
"",
"The idea of using evolutionary computation to train artificial neural networks, or neuroevolution (NE), for reinforcement learning (RL) tasks has now been around for over 20 years. However, as RL tasks become more challenging, the networks required become larger, as do their genomes. But, scaling NE to large nets (i.e. tens of thousands of weights) is infeasible using direct encodings that map genes one-to-one to network components. In this paper, we scale-up our compressed network encoding where network weight matrices are represented indirectly as a set of Fourier-type coefficients, to tasks that require very-large networks due to the high-dimensionality of their input space. The approach is demonstrated successfully on two reinforcement learning tasks in which the control networks receive visual input: (1) a vision-based version of the octopus control task requiring networks with over 3 thousand weights, and (2) a version of the TORCS driving game where networks with over 1 million weights are evolved to drive a car around a track using video images from the driver's perspective.",
"In this paper we describe a learning framework for a central pattern generator (CPG)-based biped locomotion controller using a policy gradient method. Our goals in this study are to achieve CPG-based biped walking with a 3D hardware humanoid and to develop an efficient learning algorithm with CPG by reducing the dimensionality of the state space used for learning. We demonstrate that an appropriate feedback controller can be acquired within a few thousand trials by numerical simulations and the controller obtained in numerical simulation achieves stable walking with a physical robot in the real world. Numerical simulations and hardware experiments evaluate the walking velocity and stability. The results suggest that the learning algorithm is capable of adapting to environmental changes. Furthermore, we present an online learning scheme with an initial policy for a hardware robot to improve the controller within 200 iterations.",
"",
"We adapt the ideas underlying the success of Deep Q-Learning to the continuous action domain. We present an actor-critic, model-free algorithm based on the deterministic policy gradient that can operate over continuous action spaces. Using the same learning algorithm, network architecture and hyper-parameters, our algorithm robustly solves more than 20 simulated physics tasks, including classic problems such as cartpole swing-up, dexterous manipulation, legged locomotion and car driving. Our algorithm is able to find policies whose performance is competitive with those found by a planning algorithm with full access to the dynamics of the domain and its derivatives. We further demonstrate that for many of the tasks the algorithm can learn policies end-to-end: directly from raw pixel inputs."
]
} |
1704.03952 | 2606508169 | Reinforcement learning is considered as a promising direction for driving policy learning. However, training autonomous driving vehicle with reinforcement learning in real environment involves non-affordable trial-and-error. It is more desirable to first train in a virtual environment and then transfer to the real environment. In this paper, we propose a novel realistic translation network to make model trained in virtual environment be workable in real world. The proposed network can convert non-realistic virtual image input into a realistic one with similar scene structure. Given realistic frames as input, driving policy trained by reinforcement learning can nicely adapt to real world driving. Experiments show that our proposed virtual to real (VR) reinforcement learning (RL) works pretty well. To our knowledge, this is the first successful case of driving policy trained by reinforcement learning that can adapt to real world driving data. | . Image translation aims to predict image in some specific modality, given an image from another modality. This can be treated as a generic method as it predicts pixels from pixels. Recently, the community has made significant progress in generative approaches, mostly based on generative adversarial networks @cite_24 . To name a few, the work of @cite_14 explored the use of VAE-GAN @cite_28 in generating 3D voxel models, and the work of @cite_26 proposed a cascade GAN to generate natural images by structure and style. More recently, the work of @cite_4 developed a general and simple framework for image-to-image translation which can handle various pixel level generative tasks like semantic segmentation, colorization, rendering edge maps, etc. | {
"cite_N": [
"@cite_14",
"@cite_4",
"@cite_26",
"@cite_28",
"@cite_24"
],
"mid": [
"2949551726",
"2552465644",
"2298992465",
"2202109488",
"2099471712"
],
"abstract": [
"We study the problem of 3D object generation. We propose a novel framework, namely 3D Generative Adversarial Network (3D-GAN), which generates 3D objects from a probabilistic space by leveraging recent advances in volumetric convolutional networks and generative adversarial nets. The benefits of our model are three-fold: first, the use of an adversarial criterion, instead of traditional heuristic criteria, enables the generator to capture object structure implicitly and to synthesize high-quality 3D objects; second, the generator establishes a mapping from a low-dimensional probabilistic space to the space of 3D objects, so that we can sample objects without a reference image or CAD models, and explore the 3D object manifold; third, the adversarial discriminator provides a powerful 3D shape descriptor which, learned without supervision, has wide applications in 3D object recognition. Experiments demonstrate that our method generates high-quality 3D objects, and our unsupervisedly learned features achieve impressive performance on 3D object recognition, comparable with those of supervised learning methods.",
"We investigate conditional adversarial networks as a general-purpose solution to image-to-image translation problems. These networks not only learn the mapping from input image to output image, but also learn a loss function to train this mapping. This makes it possible to apply the same generic approach to problems that traditionally would require very different loss formulations. We demonstrate that this approach is effective at synthesizing photos from label maps, reconstructing objects from edge maps, and colorizing images, among other tasks. Indeed, since the release of the pix2pix software associated with this paper, a large number of internet users (many of them artists) have posted their own experiments with our system, further demonstrating its wide applicability and ease of adoption without the need for parameter tweaking. As a community, we no longer hand-engineer our mapping functions, and this work suggests we can achieve reasonable results without hand-engineering our loss functions either.",
"Current generative frameworks use end-to-end learning and generate images by sampling from uniform noise distribution. However, these approaches ignore the most basic principle of image formation: images are product of: (a) Structure: the underlying 3D model; (b) Style: the texture mapped onto structure. In this paper, we factorize the image generation process and propose Style and Structure Generative Adversarial Network ( ( S ^2 )-GAN). Our ( S ^2 )-GAN has two components: the Structure-GAN generates a surface normal map; the Style-GAN takes the surface normal map as input and generates the 2D image. Apart from a real vs. generated loss function, we use an additional loss with computed surface normals from generated images. The two GANs are first trained independently, and then merged together via joint learning. We show our ( S ^2 )-GAN model is interpretable, generates more realistic images and can be used to learn unsupervised RGBD representations.",
"We present an autoencoder that leverages learned representations to better measure similarities in data space. By combining a variational autoencoder with a generative adversarial network we can use learned feature representations in the GAN discriminator as basis for the VAE reconstruction objective. Thereby, we replace element-wise errors with feature-wise errors to better capture the data distribution while offering invariance towards e.g. translation. We apply our method to images of faces and show that it outperforms VAEs with element-wise similarity measures in terms of visual fidelity. Moreover, we show that the method learns an embedding in which high-level abstract visual features (e.g. wearing glasses) can be modified using simple arithmetic.",
"We propose a new framework for estimating generative models via an adversarial process, in which we simultaneously train two models: a generative model G that captures the data distribution, and a discriminative model D that estimates the probability that a sample came from the training data rather than G. The training procedure for G is to maximize the probability of D making a mistake. This framework corresponds to a minimax two-player game. In the space of arbitrary functions G and D, a unique solution exists, with G recovering the training data distribution and D equal to ½ everywhere. In the case where G and D are defined by multilayer perceptrons, the entire system can be trained with backpropagation. There is no need for any Markov chains or unrolled approximate inference networks during either training or generation of samples. Experiments demonstrate the potential of the framework through qualitative and quantitative evaluation of the generated samples."
]
} |
1704.03952 | 2606508169 | Reinforcement learning is considered as a promising direction for driving policy learning. However, training autonomous driving vehicle with reinforcement learning in real environment involves non-affordable trial-and-error. It is more desirable to first train in a virtual environment and then transfer to the real environment. In this paper, we propose a novel realistic translation network to make model trained in virtual environment be workable in real world. The proposed network can convert non-realistic virtual image input into a realistic one with similar scene structure. Given realistic frames as input, driving policy trained by reinforcement learning can nicely adapt to real world driving. Experiments show that our proposed virtual to real (VR) reinforcement learning (RL) works pretty well. To our knowledge, this is the first successful case of driving policy trained by reinforcement learning that can adapt to real world driving data. | . One part of our network is the semantic image segmentation network. There are already many great works in the field of semantic image segmentation. Many of them are based on deep convolutional neural network or fully convolutional neural network @cite_33 . In this paper, we use the SegNet for image segmentation, the structure of the network is revealed in @cite_5 , which is composed of two main parts. The first part is an encoder, which consists of Convolutional, Batch Normalization, ReLU and max pooling layers. The second part is a decoder, which replaces the pooling layers with upsampling layers. | {
"cite_N": [
"@cite_5",
"@cite_33"
],
"mid": [
"2963881378",
"2952632681"
],
"abstract": [
"We present a novel and practical deep fully convolutional neural network architecture for semantic pixel-wise segmentation termed SegNet. This core trainable segmentation engine consists of an encoder network, a corresponding decoder network followed by a pixel-wise classification layer. The architecture of the encoder network is topologically identical to the 13 convolutional layers in the VGG16 network [1] . The role of the decoder network is to map the low resolution encoder feature maps to full input resolution feature maps for pixel-wise classification. The novelty of SegNet lies is in the manner in which the decoder upsamples its lower resolution input feature map(s). Specifically, the decoder uses pooling indices computed in the max-pooling step of the corresponding encoder to perform non-linear upsampling. This eliminates the need for learning to upsample. The upsampled maps are sparse and are then convolved with trainable filters to produce dense feature maps. We compare our proposed architecture with the widely adopted FCN [2] and also with the well known DeepLab-LargeFOV [3] , DeconvNet [4] architectures. This comparison reveals the memory versus accuracy trade-off involved in achieving good segmentation performance. SegNet was primarily motivated by scene understanding applications. Hence, it is designed to be efficient both in terms of memory and computational time during inference. It is also significantly smaller in the number of trainable parameters than other competing architectures and can be trained end-to-end using stochastic gradient descent. We also performed a controlled benchmark of SegNet and other architectures on both road scenes and SUN RGB-D indoor scene segmentation tasks. These quantitative assessments show that SegNet provides good performance with competitive inference time and most efficient inference memory-wise as compared to other architectures. We also provide a Caffe implementation of SegNet and a web demo at http: mi.eng.cam.ac.uk projects segnet .",
"Convolutional networks are powerful visual models that yield hierarchies of features. We show that convolutional networks by themselves, trained end-to-end, pixels-to-pixels, exceed the state-of-the-art in semantic segmentation. Our key insight is to build \"fully convolutional\" networks that take input of arbitrary size and produce correspondingly-sized output with efficient inference and learning. We define and detail the space of fully convolutional networks, explain their application to spatially dense prediction tasks, and draw connections to prior models. We adapt contemporary classification networks (AlexNet, the VGG net, and GoogLeNet) into fully convolutional networks and transfer their learned representations by fine-tuning to the segmentation task. We then define a novel architecture that combines semantic information from a deep, coarse layer with appearance information from a shallow, fine layer to produce accurate and detailed segmentations. Our fully convolutional network achieves state-of-the-art segmentation of PASCAL VOC (20 relative improvement to 62.2 mean IU on 2012), NYUDv2, and SIFT Flow, while inference takes one third of a second for a typical image."
]
} |
1704.03706 | 2951240189 | In application domains such as robotics, it is useful to represent the uncertainty related to the robot's belief about the state of its environment. Algorithms that only yield a single "best guess" as a result are not sufficient. In this paper, we propose object proposal generation based on non-parametric Bayesian inference that allows quantification of the likelihood of the proposals. We apply Markov chain Monte Carlo to draw samples of image segmentations via the distance dependent Chinese restaurant process. Our method achieves state-of-the-art performance on an indoor object discovery data set, while additionally providing a likelihood term for each proposal. We show that the likelihood term can effectively be used to rank proposals according to their quality. | Object discovery methods include window-scoring methods (e.g. @cite_10 ) that slide a window over the image which is evaluated for its objectness, and segment-grouping methods (e.g. @cite_0 ), that start with an oversegmentation of the image and group these segments to obtain object proposals. Segment-grouping methods have the advantage of delivering object contours instead of only bounding boxes, which is especially important in applications such as robotics where the object of interest might have to be manipulated. We concentrate here on the segment-grouping approach. | {
"cite_N": [
"@cite_0",
"@cite_10"
],
"mid": [
"2121660792",
"2066624635"
],
"abstract": [
"Generic object detection is the challenging task of proposing windows that localize all the objects in an image, regardless of their classes. Such detectors have recently been shown to benefit many applications such as speeding-up class-specific object detection, weakly supervised learning of object detectors and object discovery. In this paper, we introduce a novel and very efficient method for generic object detection based on a randomized version of Prim's algorithm. Using the connectivity graph of an image's super pixels, with weights modelling the probability that neighbouring super pixels belong to the same object, the algorithm generates random partial spanning trees with large expected sum of edge weights. Object localizations are proposed as bounding-boxes of those partial trees. Our method has several benefits compared to the state-of-the-art. Thanks to the efficiency of Prim's algorithm, it samples proposals very quickly: 1000 proposals are obtained in about 0.7s. With proposals bound to super pixel boundaries yet diversified by randomization, it yields very high detection rates and windows that tightly fit objects. In extensive experiments on the challenging PASCAL VOC 2007 and 2012 and SUN2012 benchmark datasets, we show that our method improves over state-of-the-art competitors for a wide range of evaluation scenarios.",
"We present a generic objectness measure, quantifying how likely it is for an image window to contain an object of any class. We explicitly train it to distinguish objects with a well-defined boundary in space, such as cows and telephones, from amorphous background elements, such as grass and road. The measure combines in a Bayesian framework several image cues measuring characteristics of objects, such as appearing different from their surroundings and having a closed boundary. These include an innovative cue to measure the closed boundary characteristic. In experiments on the challenging PASCAL VOC 07 dataset, we show this new cue to outperform a state-of-the-art saliency measure, and the combined objectness measure to perform better than any cue alone. We also compare to interest point operators, a HOG detector, and three recent works aiming at automatic object segmentation. Finally, we present two applications of objectness. In the first, we sample a small numberof windows according to their objectness probability and give an algorithm to employ them as location priors for modern class-specific object detectors. As we show experimentally, this greatly reduces the number of windows evaluated by the expensive class-specific model. In the second application, we use objectness as a complementary score in addition to the class-specific model, which leads to fewer false positives. As shown in several recent papers, objectness can act as a valuable focus of attention mechanism in many other applications operating on image windows, including weakly supervised learning of object categories, unsupervised pixelwise segmentation, and object tracking in video. Computing objectness is very efficient and takes only about 4 sec. per image."
]
} |
1704.03706 | 2951240189 | In application domains such as robotics, it is useful to represent the uncertainty related to the robot's belief about the state of its environment. Algorithms that only yield a single "best guess" as a result are not sufficient. In this paper, we propose object proposal generation based on non-parametric Bayesian inference that allows quantification of the likelihood of the proposals. We apply Markov chain Monte Carlo to draw samples of image segmentations via the distance dependent Chinese restaurant process. Our method achieves state-of-the-art performance on an indoor object discovery data set, while additionally providing a likelihood term for each proposal. We show that the likelihood term can effectively be used to rank proposals according to their quality. | Although some segment-grouping approaches such as e.g. @cite_0 apply random sampling to generate object proposals, it is often not possible to estimate a likelihood value for a particular combination of superpixels, nor is it intuitively clear what the overall probability distribution over image segments is that is applied in the sampling. However, both these properties are useful in application domains such as robotics, where decisions are made based on the observed image data, see, e.g., @cite_7 @cite_1 . To address these limitations, we consider non-parametric Bayesian methods for superpixel clustering. Such methods have been previously applied to image segmentation with the aim of replicating human segmentation of images. For example, @cite_5 applies the distance dependent Chinese restaurant process (ddCRP) and @cite_11 proposes a hierarchical Dirichlet process Markov random field for the segmentation task. @cite_8 , multiple segmentation hypotheses are produced applying the spatially dependent Pitman-Yor process. Recent work applies a Poisson process with segment shape priors for segmentation @cite_4 . | {
"cite_N": [
"@cite_4",
"@cite_7",
"@cite_8",
"@cite_1",
"@cite_0",
"@cite_5",
"@cite_11"
],
"mid": [
"2552780611",
"1980076217",
"2025498056",
"1486763584",
"2121660792",
"2122740842",
"1683428144"
],
"abstract": [
"",
"",
"We explore recently proposed Bayesian nonparametric models of image partitions, based on spatially dependent Pitman-Yor processes. These models are attractive because they adapt to images of varying complexity, successfully modeling uncertainty in the structure and scale of human segmentations of natural scenes. By developing substantially improved inference and learning algorithms, we achieve performance comparable to state-of-the-art methods. For learning, we show how the Gaussian process (GP) covariance functions underlying these models can be calibrated to accurately match the statistics of example human segmentations. For inference, we develop a stochastic search-based algorithm which is substantially less susceptible to local optima than conventional variational methods. Our approach utilizes the expectation propagation algorithm to approximately marginalize latent GPs, and a low rank covariance representation to improve computational efficiency. Experiments with two benchmark datasets show that our learning and inference innovations substantially improve segmentation accuracy. By hypothesizing multiple partitions for each image, we also take steps towards capturing the variability of human scene interpretations.",
"Making decisions based on visual input is challenging because determining how the scene should be split into individual objects is often very difficult. While previous work mainly considers decision making and visual processing as two separate tasks, we argue that the inherent uncertainty in object segmentation requires an integrated approach that chooses the best decision over all possible segmentations. Our approach over-segments the visual input and combines the segments into possible objects to get a probability distribution over object compositions, represented as particles. We introduce a Markov chain Monte Carlo procedure that aims to produce exact, independent samples. In experiments, where a 6-DOF robot arm moves object hypotheses captured by an RGB-D visual sensor, our approach of probability distribution based decision making outperforms an approach which utilises the traditional most likely object composition.",
"Generic object detection is the challenging task of proposing windows that localize all the objects in an image, regardless of their classes. Such detectors have recently been shown to benefit many applications such as speeding-up class-specific object detection, weakly supervised learning of object detectors and object discovery. In this paper, we introduce a novel and very efficient method for generic object detection based on a randomized version of Prim's algorithm. Using the connectivity graph of an image's super pixels, with weights modelling the probability that neighbouring super pixels belong to the same object, the algorithm generates random partial spanning trees with large expected sum of edge weights. Object localizations are proposed as bounding-boxes of those partial trees. Our method has several benefits compared to the state-of-the-art. Thanks to the efficiency of Prim's algorithm, it samples proposals very quickly: 1000 proposals are obtained in about 0.7s. With proposals bound to super pixel boundaries yet diversified by randomization, it yields very high detection rates and windows that tightly fit objects. In extensive experiments on the challenging PASCAL VOC 2007 and 2012 and SUN2012 benchmark datasets, we show that our method improves over state-of-the-art competitors for a wide range of evaluation scenarios.",
"The distance dependent Chinese restaurant process (ddCRP) was recently introduced to accommodate random partitions of non-exchangeable data [1]. The dd-CRP clusters data in a biased way: each data point is more likely to be clustered with other data that are near it in an external sense. This paper examines the dd-CRP in a spatial setting with the goal of natural image segmentation. We explore the biases of the spatial ddCRP model and propose a novel hierarchical extension better suited for producing \"human-like\" segmentations. We then study the sensitivity of the models to various distance and appearance hyperparameters, and provide the first rigorous comparison of nonparametric Bayesian models in the image segmentation domain. On unsupervised image segmentation, we demonstrate that similar performance to existing nonparametric Bayesian models is possible with substantially simpler models and algorithms.",
"Infinite Hidden Markov Random Fields have been proposed for image segmentation as a solution to the problem of automatically determining the number of regions in an image; however, the model does not maintain identity of segmented regions among multiple images. In order to identify segmented regions in images, we developed Hierarchical Dirichlet Process Markov Random Fields. Our model maintains global identification of segmented regions in multiple images by incorporating the idea of hierarchical modeling and automatically determines the number of segmented regions in each image. We show an experimental comparison between the previous model and our proposed model by changing the observation features from RGB value to color histogram features."
]
} |
1704.03706 | 2951240189 | In application domains such as robotics, it is useful to represent the uncertainty related to the robot's belief about the state of its environment. Algorithms that only yield a single "best guess" as a result are not sufficient. In this paper, we propose object proposal generation based on non-parametric Bayesian inference that allows quantification of the likelihood of the proposals. We apply Markov chain Monte Carlo to draw samples of image segmentations via the distance dependent Chinese restaurant process. Our method achieves state-of-the-art performance on an indoor object discovery data set, while additionally providing a likelihood term for each proposal. We show that the likelihood term can effectively be used to rank proposals according to their quality. | In our work, similarly to @cite_5 , we apply Markov chain Monte Carlo (MCMC) sampling from a ddCRP posterior to generate clusters of superpixels. However, in contrast to earlier work our main aim is object discovery. We tune our method especially towards this aim by setting the model hyperparameters to produce clusters of superpixels that have a strong link to human object perception as described by the Gestalt principles of human object perception @cite_13 . | {
"cite_N": [
"@cite_5",
"@cite_13"
],
"mid": [
"2122740842",
"2151823074"
],
"abstract": [
"The distance dependent Chinese restaurant process (ddCRP) was recently introduced to accommodate random partitions of non-exchangeable data [1]. The dd-CRP clusters data in a biased way: each data point is more likely to be clustered with other data that are near it in an external sense. This paper examines the dd-CRP in a spatial setting with the goal of natural image segmentation. We explore the biases of the spatial ddCRP model and propose a novel hierarchical extension better suited for producing \"human-like\" segmentations. We then study the sensitivity of the models to various distance and appearance hyperparameters, and provide the first rigorous comparison of nonparametric Bayesian models in the image segmentation domain. On unsupervised image segmentation, we demonstrate that similar performance to existing nonparametric Bayesian models is possible with substantially simpler models and algorithms.",
"In 1912, Max Wertheimer published his paper on phi motion, widely recognized as the start of Gestalt psychology. Because of its continued relevance in modern psychology, this centennial anniversary is an excellent opportunity to take stock of what Gestalt psychology has offered and how it has changed since its inception. We first introduce the key findings and ideas in the Berlin school of Gestalt psychology, and then briefly sketch its development, rise, and fall. Next, we discuss its empirical and conceptual problems, and indicate how they are addressed in contemporary research on perceptual grouping and figure–ground organization. In particular, we review the principles of grouping, both classical (e.g., proximity, similarity, common fate, good continuation, closure, symmetry, parallelism) and new (e.g., synchrony, common region, element and uniform connectedness), and their role in contour integration and completion. We then review classic and new image-based principles of figure–ground organization, how it is influenced by past experience and attention, and how it relates to shape and depth perception. After an integrated review of the neural mechanisms involved in contour grouping, border ownership, and figure–ground perception, we conclude by evaluating what modern vision science has offered compared to traditional Gestalt psychology, whether we can speak of a Gestalt revival, and where the remaining limitations and challenges lie. A better integration of this research tradition with the rest of vision science requires further progress regarding the conceptual and theoretical foundations of the Gestalt approach, which is the focus of a second review article."
]
} |
1704.03672 | 2607372571 | We observed that safety arguments are prone to stay too abstract, e.g. solutions refer to large packages, argument strategies to complex reasoning steps, contexts and assumptions lack traceability. These issues can reduce the confidence we require of such arguments. In this paper, we investigate the construction of confident arguments from (i) hazard analysis (HA) results and (ii) the design of safety measures, i.e., both used for confidence evaluation. We present an argument pattern integrating three HA techniques, i.e., FTA, FMEA, and STPA, as well as the reactions on the results of these analyses, i.e., safety requirements and design increments. We provide an example of how our pattern can help in argument construction and discuss steps towards using our pattern in formal analysis and computer-assisted construction of safety cases. | @cite_7 present patterns arguing for safe control software by using adaptation mechanisms for improving or maintaining safe states. Our three modules (M, CR, and HC) are similar to their core patterns but use fewer argumentation steps. They focus on FMEA, arguing over adaptation as a design measure ( @math ) which goes beyond in our countermeasure module ( @math ). We integrate FTA, FMEA, and STPA and capture how safety requirements were motivated. We aim at a compact pattern and use HA results more directly. Their argument depicts that an identified risk structure ( failure modes as hazards and their causal factors) is acceptable and, by adaptations as countermeasures, how this structure got acceptable. We do not presume adaptation as a countermeasure. Moreover, we allow the goal structure to be built during FTA, FMEA, or STPA whereas they concentrate on an a-posteriori construction. | {
"cite_N": [
"@cite_7"
],
"mid": [
"141099830"
],
"abstract": [
"Abstract : This report results from a contract tasking University of York. The project will undertake one activity: Produce a unified (generic) approach to developing safety cases for adaptive avionics and software and identifying a \"way ahead\" to develop and validate the approach, based on the outline produced in the preceding NASA project. As with the previous report, the authors' focus in addressing safety cases for \"advanced\" control systems is to concentrate on the class of adaptive systems. A system can be considered adaptive if its behavior cannot be predicted solely from knowledge of its initial software design and state. The behavior of an adaptive system is the product of its initial state and the adaptations (state changes) that have taken place according to the stimuli it has encountered. Adaptive systems can be introduced to improve safety (e.g., to continue to control an aircraft safely in the event of losing a control surface), or to improve other system characteristics (e.g., to improve the fuel consumption of an aero-engine). The motivation for introducing an adaptive capability has a significant impact on the nature of the required safety argument. Where improved safety is the goal of the adaptation, the safety argument must justify that the adaptive system is capable of reducing some of the risks associated with hazards already present with the equipment under control. At the same, it is necessary to ensure that the introduction of the adaptive capability does not introduce new, or increase existing, risks. Where adaptation is being introduced for reasons other than safety, safety can be viewed as a constraint. The principal concern is that the adaptive capability doesn't introduce new, or increase existing, risks."
]
} |
1704.03672 | 2607372571 | We observed that safety arguments are prone to stay too abstract, e.g. solutions refer to large packages, argument strategies to complex reasoning steps, contexts and assumptions lack traceability. These issues can reduce the confidence we require of such arguments. In this paper, we investigate the construction of confident arguments from (i) hazard analysis (HA) results and (ii) the design of safety measures, i.e., both used for confidence evaluation. We present an argument pattern integrating three HA techniques, i.e., FTA, FMEA, and STPA, as well as the reactions on the results of these analyses, i.e., safety requirements and design increments. We provide an example of how our pattern can help in argument construction and discuss steps towards using our pattern in formal analysis and computer-assisted construction of safety cases. | Palin and Habli @cite_16 propose safety case pattern catalogs for the construction of a vehicle safety case in accordance with ISO 26262. One of the proposed catalogs, namely the pattern catalog, confirms the necessity of an pattern. Their paper does not elaborate on this pattern. | {
"cite_N": [
"@cite_16"
],
"mid": [
"1532757718"
],
"abstract": [
"A safety case should provide a clear, comprehensible and defensible argument, supported by evidence, that a system is acceptably safe to operate in a particular environment. This approach is not new. For example, in the nuclear industry, safety cases are approaching their 50th birthday. In stark contrast, the automotive industry has never been required to produce a safety case. Instead, it has relied on compliance with extensive regional and national regulation. With the imminent introduction of the automotive safety standard ISO 26262, the production of a safety case is now explicitly required by the standard for electrical and electronic systems. This presents both opportunities and challenges to safety practitioners and researchers within that industry. This paper sets out to look at the issues of what a safety case might look like for a complete vehicle and how the ISO 26262 fits into the existing framework of automotive safety. Using the ideas of modular safety case construction, this approach is developed into a number of reusable safety arguments to form an automotive safety case pattern catalogue. Evaluation of the approach is carried out through an industrial case study."
]
} |
1704.03672 | 2607372571 | We observed that safety arguments are prone to stay too abstract, e.g. solutions refer to large packages, argument strategies to complex reasoning steps, contexts and assumptions lack traceability. These issues can reduce the confidence we require of such arguments. In this paper, we investigate the construction of confident arguments from (i) hazard analysis (HA) results and (ii) the design of safety measures, i.e., both used for confidence evaluation. We present an argument pattern integrating three HA techniques, i.e., FTA, FMEA, and STPA, as well as the reactions on the results of these analyses, i.e., safety requirements and design increments. We provide an example of how our pattern can help in argument construction and discuss steps towards using our pattern in formal analysis and computer-assisted construction of safety cases. | For a truck information and control system, Dardar @cite_29 constructs an ISO 26262-compliant safety case using GSN and SysML. He argues from a trustworthy process based on coarse evidence from several HA and quality assurance techniques. However, for FTA he shows a goal structure that can be seen as an instance of our CR and HC modules for FTA. | {
"cite_N": [
"@cite_29"
],
"mid": [
"611011698"
],
"abstract": [
"Nowadays, road vehicles, including trucks, are characterized by an increasedcomplexity due to a greater variety of software, and a greater number of sensorsand actuators. As a consequence, there is an increased risk in termsof software or hardware failures that could lead to unacceptable hazards.Thus safety, more precisely functional safety, is a crucial property that mustbe ensured to avoid or mitigate these potential unacceptable hazards. Inthe automotive domain, recently (November 2011), the ISO-26262 safetystandard has been introduced to provide appropriate requirements and processes.More specically, the standard denes the system development processthat must be carried out to achieve a system that can be consideredacceptably safe. To be released on the market, systems must be certied,proofs that the systems are acceptably safe must be provided in terms of astructured argument, known as safety case, which inter-relates evidence andclaims. Certication authorities are in charge of evaluating the validity ofsuch safety cases. In the automotive domain, certication and compliancewith the standard ISO-26262 is becoming mandatory. By now, trucks donot have to be compliant with the standard. However, it is likely that by2016 they will have to. Scania is one of the leading companies in trucksdevelopment. To be ready by 2016, Scania is interested in investigatingISO-26262 as well as safety case provision. Thus this thesis focuses on theprovision of a safety case in the context of ISO-26262 for Fuel Level Estimationand Display System (FLEDS), which is one of the safety-criticalsystems in Scania.1"
]
} |
1704.03641 | 2952691609 | Traditionally, Internet Access Providers (APs) only charge end-users for Internet access services; however, to recoup infrastructure costs and increase revenues, some APs have recently adopted two-sided pricing schemes under which both end-users and content providers are charged. Meanwhile, with the rapid growth of traffic, network congestion could seriously degrade user experiences and influence providers' utility. To optimize profit and social welfare, APs and regulators need to design appropriate pricing strategies and regulatory policies that take the effects of network congestion into consideration. In this paper, we model two-sided networks under which users' traffic demands are influenced by exogenous pricing and endogenous congestion parameters and derive the system congestion under an equilibrium. We characterize the structures and sensitivities of profit- and welfare-optimal two-sided pricing schemes and reveal that 1) the elasticity of system throughput plays a crucial role in determining the structures of optimal pricing, 2) the changes of optimal pricing under varying AP's capacity and users' congestion sensitivity are largely driven by the type of data traffic, e.g., text or video, and 3) APs and regulators will be incentivized to shift from one-sided to two-sided pricing when APs' capacities and user demand for video traffic grow. Our results can help APs design optimal two-sided pricing and guide regulators to legislate desirable policies. | Several works have studied two-sided pricing in the Internet markets. @cite_5 showed that two-sided pricing could help APs extract higher profits and maintain higher investment levels than one-sided pricing. @cite_12 analyzed the impacts of CP's revenue models, either subscription or advertisement, on AP's pricing strategies. Choi and Kim @cite_11 found that expanding capacity will decrease the CP-side price. In @cite_23 , concluded that two-sided pricing is more favorable in terms of social welfare than one-sided pricing when the ratio between CPs' advertising rates and user price sensitivity are extreme. All of the above works do not consider the impact of network congestion on user throughput, which strongly influences the AP's pricing strategy. In this paper, we characterize the interactions among network congestion, throughput, and price, based on which we analyze both the profit-optimal and welfare-optimal pricing. | {
"cite_N": [
"@cite_5",
"@cite_23",
"@cite_12",
"@cite_11"
],
"mid": [
"2000244192",
"2006549632",
"2165616869",
"1971741818"
],
"abstract": [
"This paper develops a game-theoretic model based on a two-sided market framework to compare Internet service providers’ (ISPs) investment incentives, content providers’ (CPs) participation, and social welfare between neutral and non-neutral network regimes. We find that ISPs’ investments are driven by the trade-off between softening consumer price competition and increasing revenues from CPs. Specifically, investments are higher in the non-neutral regime because it is easier to extract revenue through appropriate CP pricing. On the other hand, participation of CPs may be reduced in a non-neutral network due to higher prices. The net impact of non-neutrality on social welfare is determined by which of these two effects is dominant. Overall, we find that the non-neutral network is always welfare superior in a “walled-gardens” model, while the neutral network is superior in a “priority lanes” model when CP-quality heterogeneity is large. These results provide useful insights that inform the net-neutrality debate.",
"We address whether local ISPs should be allowed to charge content providers, who derive advertising revenue, for the right to access end-users. We compare two-sided pricing where such charges are allowed to one-sided pricing where they are prohibited. By deriving provider equilibrium actions (prices and investments), we determine which regime is welfare-superior as a function of a few key parameters. We find that two-sided pricing is more favorable when the ratio between parameters characterizing advertising rates and end-user price sensitivity is either low or high.",
"This paper studies economic utilities and quality of service (QoS) in a two-sided non-neutral market where Internet service providers (ISPs) charge content providers (CPs) for the content delivery. We propose new models that involve a CP, an ISP, end users and advertisers. The CP may have either a subscription revenue model (charging end users) or an advertisement revenue model (charging advertisers). We formulate the interactions between the ISP and the CP as a noncooperative game for the former and an optimization problem for the latter. Our analysis shows that the revenue model of the CP plays a significant role in a non-neutral Internet. With the subscription model, both the ISP and the CP receive better (or worse) utilities as well as QoS in the presence of the side payment at the same time. With the advertisement model, the side payment impedes the CP from investing on its contents.",
"This article analyzes the effects of net neutrality regulation on investment incentives for Internet service providers (ISPs) and content providers (CPs), and their implications for social welfare. Concerning the ISPs' investment incentives, we find that capacity expansion decreases the sale price of the priority right under the discriminatory regime. Thus, contrary to ISPs' claims that net neutrality regulations would have a chilling effect on their incentive to invest, we cannot dismiss the possibility of the opposite. A discriminatory regime can also weaken CPs' investment incentives because of CPs' concern that the ISP would expropriate some of the investment benefits."
]
} |
1704.03641 | 2952691609 | Traditionally, Internet Access Providers (APs) only charge end-users for Internet access services; however, to recoup infrastructure costs and increase revenues, some APs have recently adopted two-sided pricing schemes under which both end-users and content providers are charged. Meanwhile, with the rapid growth of traffic, network congestion could seriously degrade user experiences and influence providers' utility. To optimize profit and social welfare, APs and regulators need to design appropriate pricing strategies and regulatory policies that take the effects of network congestion into consideration. In this paper, we model two-sided networks under which users' traffic demands are influenced by exogenous pricing and endogenous congestion parameters and derive the system congestion under an equilibrium. We characterize the structures and sensitivities of profit- and welfare-optimal two-sided pricing schemes and reveal that 1) the elasticity of system throughput plays a crucial role in determining the structures of optimal pricing, 2) the changes of optimal pricing under varying AP's capacity and users' congestion sensitivity are largely driven by the type of data traffic, e.g., text or video, and 3) APs and regulators will be incentivized to shift from one-sided to two-sided pricing when APs' capacities and user demand for video traffic grow. Our results can help APs design optimal two-sided pricing and guide regulators to legislate desirable policies. | Whether APs should be allowed to charge CPs for their content traffic has been a focus of debate on net neutrality @cite_16 . To sidestep this debate and extract revenue from CPs, some APs, e.g., AT &T, have recently provided sponsored data plans, which allow CPs to partially or fully subsidize users the fees induced by their data traffic for increasing market share. Since the sponsorship offers a way for CPs to subsidize their users and indirectly transfer value to APs, it could be seen as a variant of the two-sided pricing model. Some work @cite_0 @cite_4 @cite_6 have studied this variant and showed that it benefits both CPs and APs. Because APs charge users different prices for content traffic from different CPs, the sponsored data plan is also considered as a type of price discrimination in disguise and has raised concerns from the FCC who says that they will be monitoring and prepared to intervene if necessary @cite_17 . In this paper, rather than pursuing the price differentiation, we focus on the two-sided pricing under which users are charged entirely based on their traffic volumes. | {
"cite_N": [
"@cite_4",
"@cite_6",
"@cite_0",
"@cite_16",
"@cite_17"
],
"mid": [
"2949398319",
"2058615235",
"2314643314",
"160103087",
""
],
"abstract": [
"Unlike telephone operators, which pay termination fees to reach the users of another network, Internet Content Providers (CPs) do not pay the Internet Service Providers (ISPs) of users they reach. While the consequent cross subsidization to CPs has nurtured content innovations at the edge of the Internet, it reduces the investment incentives for the access ISPs to expand capacity. As potential charges for terminating CPs' traffic are criticized under the net neutrality debate, we propose to allow CPs to voluntarily subsidize the usagebased fees induced by their content traffic for end-users. We model the regulated subsidization competition among CPs under a neutral network and show how deregulation of subsidization could increase an access ISP's utilization and revenue, strengthening its investment incentives. Although the competition might harm certain CPs, we find that the main cause comes from high access prices rather than the existence of subsidization. Our results suggest that subsidization competition will increase the competitiveness and welfare of the Internet content market; however, regulators might need to regulate access prices if the access ISP market is not competitive enough. We envision that subsidization competition could become a viable model for the future Internet.",
"Data traffic demand over the Internet is increasing rapidly, and it is changing the pricing model between Internet service providers (ISPs), content providers (CPs) and end users. One recent pricing proposal is sponsored data plan, i.e., when accessing contents from a particular CP, end users do not need to pay for that volume of traffic consumed, but the CP will sponsor for this data consumption. In this paper, our goal is to understand the rationale behind this new pricing model, as well as its impacts to the wireless data market, in particular, who will benefit and who will be hurt from this scheme. We build a two-class service model to analyze the consumers' traffic demand under the sponsored data plan with consideration of QoS. We use a two-stage Stackelberg game to characterize the interaction between CPs and the ISP and reveal a number of important findings. Our conclusions include: 1) When the ISP's capacity is sufficient, the sponsored data plan benefits consumers and CPs in the short run, but the ISP does not have incentives to further improve its service in the long run. 2) When ISP's capacity is insufficient, the ISP and end users may achieve a win- win trade, while the ISP and CPs always compete for the revenue. 3) The sponsored data plan may enlarge the un- balance in revenue distribution between different CPs; CPs with higher unit income and poorer technology support are more likely to prefer the sponsored data plan.",
"The interaction of a content provider with end users on an infrastructure platform built and maintained by a service provider can be viewed as a two-sided market. Content sponsoring, i.e., charging the content provider instead of viewers for resources consumed in viewing the content, can benefit all parties involved. Without being charged directly or having it counted against their monthly data quotas, end users will view more content, allowing the content provider to generate more advertising revenue, extracted by the service provider to subsidize its investment and operation of the network infrastructure. However, realizing such gains requires a proper contractual relationship between the service provider and content provider. We consider the determination of this contract through a Stackelberg game. The service provider sets a pricing schedule for sponsoring and the content provider responds by deciding how much content to sponsor. We analyze the best strategies for the content provider and service provider in the event that the underlying demand for the content is uncertain. Two separate settings are defined. In the first, end users can be charged for non-sponsored views on a per-byte basis. In the second we extend the model to the more common case in which end users purchase data quotas on a periodic basis. Our main conclusion is that a coordinating contract can be designed that maximizes total system profit. Moreover, the additional profit due to sponsoring can be split between the content provider and service provider in an arbitrary manner.",
"This paper examines the the concept of network neutrality in telecommunications policy and its relationship to Darwinian theories of innovation. It also considers the record of broadband discrimination practiced by broadband operators in the early 2000s.",
""
]
} |
1704.03641 | 2952691609 | Traditionally, Internet Access Providers (APs) only charge end-users for Internet access services; however, to recoup infrastructure costs and increase revenues, some APs have recently adopted two-sided pricing schemes under which both end-users and content providers are charged. Meanwhile, with the rapid growth of traffic, network congestion could seriously degrade user experiences and influence providers' utility. To optimize profit and social welfare, APs and regulators need to design appropriate pricing strategies and regulatory policies that take the effects of network congestion into consideration. In this paper, we model two-sided networks under which users' traffic demands are influenced by exogenous pricing and endogenous congestion parameters and derive the system congestion under an equilibrium. We characterize the structures and sensitivities of profit- and welfare-optimal two-sided pricing schemes and reveal that 1) the elasticity of system throughput plays a crucial role in determining the structures of optimal pricing, 2) the changes of optimal pricing under varying AP's capacity and users' congestion sensitivity are largely driven by the type of data traffic, e.g., text or video, and 3) APs and regulators will be incentivized to shift from one-sided to two-sided pricing when APs' capacities and user demand for video traffic grow. Our results can help APs design optimal two-sided pricing and guide regulators to legislate desirable policies. | From the perspective of modeling and analysis, our model extends Rochet and Tirole @cite_14 , in which the two-sided network platforms do not incur congestion. To capture the endogenous network congestion and its effect in Internet markets, we introduce a system congestion under a market equilibrium and use a gain function to model the decline degrees of network throughput under different congestion levels. Besides, we also analyze the sensitivities of two-sided pricing under the varying network environments, which are instructive and meaningful for APs and regulators to adjust pricing schemes and regulatory policies with the evolution of the Internet. Ma @cite_10 and Chander and Leruth @cite_7 also consider the service markets with congestion externalities. Ma @cite_10 analyzed the pay-as-you-go pricing and competition among multiple ISPs. Chander and Leruth @cite_7 studied the quality differentiation strategy of a monopoly provider. Both of them focused on the one-sided markets while we consider the more general two-sided markets. | {
"cite_N": [
"@cite_14",
"@cite_10",
"@cite_7"
],
"mid": [
"2085225290",
"2087607694",
"2135483574"
],
"abstract": [
"Many if not most markets with network externalities are two-sided. To succeed, platforms in industries such as software, portals and media, payment systems and the Internet, must \"get both sides of the market on board \". Accordingly, platforms devote much attention to their business model, that is to how they court each side while making money overall. The paper builds a model of platform competition with two-sided markets. It unveils the determinants of price allocation and enduser surplus for different governance structures (profit-maximizing platforms and not-for-profit joint undertakings), and compares the outcomes with those under an integrated monopolist and a Ramsey planner.",
"As Internet traffic grows exponentially due to the pervasive Internet accesses via mobile devices and increasing adoptions of cloud-based applications, broadband providers start to shift from flat-rate to usage-based pricing, which has gained support from regulators such as the FCC. We consider generic congestion-prone network services, including cloud services, and study the pay-as-you-go type of usage-based pricing of service providers under market competition. Based on a novel model that captures users' preferences over usage price and congestion alternatives, we derive the induced congestion and market share of the service providers under a market equilibrium and design algorithms to calculate them. By analyzing different market structures, we reveal how users' value on usage and sensitivity to congestion influence the optimal price, revenue, and competition of service providers, as well as the social welfare. We also obtain the conditions under which monopolistic providers have strong incentives to implement service differentiation via Paris Metro Pricing and whether regulators should encourage such practices.",
"The paper develops a model of product differentiation in which the quality of a product may be negatively affected by the number of consumers buying it, as it is the case for any good affected by congestion. It is shown that for any positive degree of heterogeneity among the consumers, a monopolist will always find it more profitable to differentiate, i.e., to sell more than one quality of the product at different prices."
]
} |
1704.03615 | 2606461407 | While deep feature learning has revolutionized techniques for static-image understanding, the same does not quite hold for video processing. Architectures and optimization techniques used for video are largely based off those for static images, potentially underutilizing rich video information. In this work, we rethink both the underlying network architecture and the stochastic learning paradigm for temporal data. To do so, we draw inspiration from classic theory on linear dynamic systems for modeling time series. By extending such models to include nonlinear mappings, we derive a series of novel recurrent neural networks that sequentially make top-down predictions about the future and then correct those predictions with bottom-up observations. Predictive-corrective networks have a number of desirable properties: (1) they can adaptively focus computation on surprising frames where predictions require large corrections, (2) they simplify learning in that only residual-like corrective terms need to be learned over time and (3) they naturally decorrelate an input data stream in a hierarchical fashion, producing a more reliable signal for learning at each layer of a network. We provide an extensive analysis of our lightweight and interpretable framework, and demonstrate that our model is competitive with the two-stream network on three challenging datasets without the need for computationally expensive optical flow. | Action recognition There is a vast literature on action recognition from videos: to name a few, @cite_53 @cite_47 explore fusing image-based convolutional networks over time, @cite_58 use RGB pixel information together with optical flow to capture motion, @cite_50 @cite_16 @cite_24 extend image-based convolutional networks into 3D convolutional networks that operate on video volumes'' consisting of a fixed number of video frames. In contrast to these works, we focus on the more challenging task of temporal action localization. | {
"cite_N": [
"@cite_53",
"@cite_16",
"@cite_24",
"@cite_50",
"@cite_47",
"@cite_58"
],
"mid": [
"2016053056",
"1522734439",
"2951864506",
"1983364832",
"1923404803",
"2156303437"
],
"abstract": [
"Convolutional Neural Networks (CNNs) have been established as a powerful class of models for image recognition problems. Encouraged by these results, we provide an extensive empirical evaluation of CNNs on large-scale video classification using a new dataset of 1 million YouTube videos belonging to 487 classes. We study multiple approaches for extending the connectivity of a CNN in time domain to take advantage of local spatio-temporal information and suggest a multiresolution, foveated architecture as a promising way of speeding up the training. Our best spatio-temporal networks display significant performance improvements compared to strong feature-based baselines (55.3 to 63.9 ), but only a surprisingly modest improvement compared to single-frame models (59.3 to 60.9 ). We further study the generalization performance of our best model by retraining the top layers on the UCF-101 Action Recognition dataset and observe significant performance improvements compared to the UCF-101 baseline model (63.3 up from 43.9 ).",
"We propose a simple, yet effective approach for spatiotemporal feature learning using deep 3-dimensional convolutional networks (3D ConvNets) trained on a large scale supervised video dataset. Our findings are three-fold: 1) 3D ConvNets are more suitable for spatiotemporal feature learning compared to 2D ConvNets, 2) A homogeneous architecture with small 3x3x3 convolution kernels in all layers is among the best performing architectures for 3D ConvNets, and 3) Our learned features, namely C3D (Convolutional 3D), with a simple linear classifier outperform state-of-the-art methods on 4 different benchmarks and are comparable with current best methods on the other 2 benchmarks. In addition, the features are compact: achieving 52.8 accuracy on UCF101 dataset with only 10 dimensions and also very efficient to compute due to the fast inference of ConvNets. Finally, they are conceptually very simple and easy to train and use.",
"Typical human actions last several seconds and exhibit characteristic spatio-temporal structure. Recent methods attempt to capture this structure and learn action representations with convolutional neural networks. Such representations, however, are typically learned at the level of a few video frames failing to model actions at their full temporal extent. In this work we learn video representations using neural networks with long-term temporal convolutions (LTC). We demonstrate that LTC-CNN models with increased temporal extents improve the accuracy of action recognition. We also study the impact of different low-level representations, such as raw values of video pixels and optical flow vector fields and demonstrate the importance of high-quality optical flow estimation for learning accurate action models. We report state-of-the-art results on two challenging benchmarks for human action recognition UCF101 (92.7 ) and HMDB51 (67.2 ).",
"We consider the automated recognition of human actions in surveillance videos. Most current methods build classifiers based on complex handcrafted features computed from the raw inputs. Convolutional neural networks (CNNs) are a type of deep model that can act directly on the raw inputs. However, such models are currently limited to handling 2D inputs. In this paper, we develop a novel 3D CNN model for action recognition. This model extracts features from both the spatial and the temporal dimensions by performing 3D convolutions, thereby capturing the motion information encoded in multiple adjacent frames. The developed model generates multiple channels of information from the input frames, and the final feature representation combines information from all channels. To further boost the performance, we propose regularizing the outputs with high-level features and combining the predictions of a variety of different models. We apply the developed models to recognize human actions in the real-world environment of airport surveillance videos, and they achieve superior performance in comparison to baseline methods.",
"Convolutional neural networks (CNNs) have been extensively applied for image recognition problems giving state-of-the-art results on recognition, detection, segmentation and retrieval. In this work we propose and evaluate several deep neural network architectures to combine image information across a video over longer time periods than previously attempted. We propose two methods capable of handling full length videos. The first method explores various convolutional temporal feature pooling architectures, examining the various design choices which need to be made when adapting a CNN for this task. The second proposed method explicitly models the video as an ordered sequence of frames. For this purpose we employ a recurrent neural network that uses Long Short-Term Memory (LSTM) cells which are connected to the output of the underlying CNN. Our best networks exhibit significant performance improvements over previously published results on the Sports 1 million dataset (73.1 vs. 60.9 ) and the UCF-101 datasets with (88.6 vs. 88.0 ) and without additional optical flow information (82.6 vs. 73.0 ).",
"We investigate architectures of discriminatively trained deep Convolutional Networks (ConvNets) for action recognition in video. The challenge is to capture the complementary information on appearance from still frames and motion between frames. We also aim to generalise the best performing hand-crafted features within a data-driven learning framework. Our contribution is three-fold. First, we propose a two-stream ConvNet architecture which incorporates spatial and temporal networks. Second, we demonstrate that a ConvNet trained on multi-frame dense optical flow is able to achieve very good performance in spite of limited training data. Finally, we show that multitask learning, applied to two different action classification datasets, can be used to increase the amount of training data and improve the performance on both. Our architecture is trained and evaluated on the standard video actions benchmarks of UCF-101 and HMDB-51, where it is competitive with the state of the art. It also exceeds by a large margin previous attempts to use deep nets for video classification."
]
} |
1704.03615 | 2606461407 | While deep feature learning has revolutionized techniques for static-image understanding, the same does not quite hold for video processing. Architectures and optimization techniques used for video are largely based off those for static images, potentially underutilizing rich video information. In this work, we rethink both the underlying network architecture and the stochastic learning paradigm for temporal data. To do so, we draw inspiration from classic theory on linear dynamic systems for modeling time series. By extending such models to include nonlinear mappings, we derive a series of novel recurrent neural networks that sequentially make top-down predictions about the future and then correct those predictions with bottom-up observations. Predictive-corrective networks have a number of desirable properties: (1) they can adaptively focus computation on surprising frames where predictions require large corrections, (2) they simplify learning in that only residual-like corrective terms need to be learned over time and (3) they naturally decorrelate an input data stream in a hierarchical fashion, producing a more reliable signal for learning at each layer of a network. We provide an extensive analysis of our lightweight and interpretable framework, and demonstrate that our model is competitive with the two-stream network on three challenging datasets without the need for computationally expensive optical flow. | Temporal action localization A common way to extend action recognition models to temporal detection is through the sliding windows paradigm @cite_51 @cite_45 @cite_21 @cite_4 @cite_36 . However, this is both computationally inefficient and prevents the model from leveraging memory over the video. Classical temporal models, on the other hand, can leverage information from the past as well as the future. These models generally rely on chain structured models that admit efficient inference, such as HMMs @cite_6 @cite_54 and CRFs @cite_48 . More recent approaches for reasoning about memory generally focus on Recurrent Neural Networks (RNNs), which sequentially @cite_31 or sporadically @cite_34 process video frames and maintain an explicit memory of the previously observed frames. @cite_32 develop a Clockwork RNN'' that maintains memory states that evolve at different speeds while processing a sequence; @cite_7 extend this model to convolutional networks for semantic segmentation in videos. Our model follows similar intuition for temporal action detection. | {
"cite_N": [
"@cite_4",
"@cite_7",
"@cite_36",
"@cite_48",
"@cite_54",
"@cite_21",
"@cite_32",
"@cite_6",
"@cite_45",
"@cite_31",
"@cite_34",
"@cite_51"
],
"mid": [
"2187772033",
"2963866581",
"2566933410",
"2145835757",
"1978511849",
"",
"2138660131",
"2014914041",
"",
"2952835694",
"2963321993",
"2105101328"
],
"abstract": [
"We describe the submission of the INRIA LEAR team to the THU-MOS workshop in conjunction with ECCV 2014. Our system is based on Fisher vector (FV) encoding of dense trajectory features (DTF), which we also used in our 2013 submission. This year's submission additionally incorporates static-image features (SIFT, Color, and CNN) and audio features (ASR and MFCC) for the classification task. For the detection task, we combine scores from the clas-sification task with FV-DTF features extracted from video slices. We found that these additional visual and audio feature significantly improve the classification results. For localization we found that using the classification scores as a contex-tual feature besides local motion features leads to significant improvements.",
"Recent years have seen tremendous progress in still-image segmentation; however the naive application of these state-of-the-art algorithms to every video frame requires considerable computation and ignores the temporal continuity inherent in video. We propose a video recognition framework that relies on two key observations: (1) while pixels may change rapidly from frame to frame, the semantic content of a scene evolves more slowly, and (2) execution can be viewed as an aspect of architecture, yielding purpose-fit computation schedules for networks. We define a novel family of “clockwork” convnets driven by fixed or adaptive clock signals that schedule the processing of different layers at different update rates according to their semantic stability. We design a pipeline schedule to reduce latency for real-time recognition and a fixed-rate schedule to reduce overall computation. Finally, we extend clockwork scheduling to adaptive video processing by incorporating data-driven clocks that can be tuned on unlabeled video. The accuracy and efficiency of clockwork convnets are evaluated on the Youtube-Objects, NYUD, and Cityscapes video datasets.",
"This notebook paper describes our approaches for the action recognition and temporal localization tasks of the THUMOS Challenge 2015. For the action recognition task, we use the subsequence-score distribution (SSD) framework. We use the Improved Fisher Vectors (IFVs) encoding of the Improved Dense Trajectories (IDTs) to capture motion, as well as a VGG-16 deep net model to extract 4096 dimension feature vector to capture the context information. A linear SVM is trained for classification of 101 categories' action video clips. For the temporal localization task, we use the IFV encoding at 9 different temporal scales, and apply the above SVM to obtain a pyramid score descriptor. The score features are used for generating action labels at frame level, and by proper post processing we are able to detect the 20 class actions in given videos.",
"We introduce a discriminative hidden-state approach for the recognition of human gestures. Gesture sequences often have a complex underlying structure, and models that can incorporate hidden structures have proven to be advantageous for recognition tasks. Most existing approaches to gesture recognition with hidden states employ a Hidden Markov Model or suitable variant (e.g., a factored or coupled state model) to model gesture streams; a significant limitation of these models is the requirement of conditional independence of observations. In addition, hidden states in a generative model are selected to maximize the likelihood of generating all the examples of a given gesture class, which is not necessarily optimal for discriminating the gesture class against other gestures. Previous discriminative approaches to gesture sequence recognition have shown promising results, but have not incorporated hidden states nor addressed the problem of predicting the label of an entire sequence. In this paper, we derive a discriminative sequence model with a hidden state structure, and demonstrate its utility both in a detection and in a multi-way classification formulation. We evaluate our method on the task of recognizing human arm and head gestures, and compare the performance of our method to both generative hidden state and discriminative fully-observable models.",
"Automatic video segmentation and action recognition has been a long-standing problem in computer vision. Much work in the literature treats video segmentation and action recognition as two independent problems; while segmentation is often done without a temporal model of the activity, action recognition is usually performed on pre-segmented clips. In this paper we propose a novel method that avoids the limitations of the above approaches by jointly performing video segmentation and action recognition. Unlike standard approaches based on extensions of dynamic Bayesian networks, our method is based on a discriminative temporal extension of the spatial bag-of-words model that has been very popular in object recognition. The classification is performed robustly within a multi-class SVM framework whereas the inference over the segments is done efficiently with dynamic programming. Experimental results on honeybee, Weizmann, and Hollywood datasets illustrate the benefits of our approach compared to state-of-the-art methods.",
"",
"Sequence prediction and classification are ubiquitous and challenging problems in machine learning that can require identifying complex dependencies between temporally distant inputs. Recurrent Neural Networks (RNNs) have the ability, in theory, to cope with these temporal dependencies by virtue of the short-term memory implemented by their recurrent (feedback) connections. However, in practice they are difficult to train successfully when long-term memory is required. This paper introduces a simple, yet powerful modification to the simple RNN (SRN) architecture, the Clockwork RNN (CW-RNN), in which the hidden layer is partitioned into separate modules, each processing inputs at its own temporal granularity, making computations only at its prescribed clock rate. Rather than making the standard RNN models more complex, CW-RNN reduces the number of SRN parameters, improves the performance significantly in the tasks tested, and speeds up the network evaluation. The network is demonstrated in preliminary experiments involving three tasks: audio signal generation, TIMIT spoken word classification, where it outperforms both SRN and LSTM networks, and online handwriting recognition, where it outperforms SRNs.",
"A challenging problem in human action understanding is to jointly segment and recognize human actions from an unseen video sequence, where one person performs a sequence of continuous actions. In this paper, we propose a discriminative semi-Markov model approach, and define a set of features over boundary frames, segments, as well as neighboring segments. This enable us to conveniently capture a combination of local and global features that best represent each specific action type. To efficiently solve the inference problem of simultaneous segmentation and recognition, a Viterbi-like dynamic programming algorithm is utilized, which in practice is able to process 20 frames per second. Moreover, the model is discriminatively learned from large margin principle, and is formulated as an optimization problem with exponentially many constraints. To solve it efficiently, we present two different optimization algorithms, namely cutting plane method and bundle method, and demonstrate that each can be alternatively deployed in a \"plug and play\" fashion. From its theoretical aspect, we also analyze the generalization error of the proposed approach and provide a PAC-Bayes bound. The proposed approach is evaluated on a variety of datasets, and is shown to perform competitively to the state-of-the-art methods. For example, on KTH dataset, it achieves 95.0 recognition accuracy, where the best known result on this dataset is 93.4 (Reddy and Shah in ICCV, 2009).",
"",
"Every moment counts in action recognition. A comprehensive understanding of human activity in video requires labeling every frame according to the actions occurring, placing multiple labels densely over a video sequence. To study this problem we extend the existing THUMOS dataset and introduce MultiTHUMOS, a new dataset of dense labels over unconstrained internet videos. Modeling multiple, dense labels benefits from temporal relations within and across classes. We define a novel variant of long short-term memory (LSTM) deep networks for modeling these temporal relations via multiple input and output connections. We show that this model improves action labeling accuracy and further enables deeper understanding tasks ranging from structured retrieval to action prediction.",
"In this work we introduce a fully end-to-end approach for action detection in videos that learns to directly predict the temporal bounds of actions. Our intuition is that the process of detecting actions is naturally one of observation and refinement: observing moments in video, and refining hypotheses about when an action is occurring. Based on this insight, we formulate our model as a recurrent neural network-based agent that interacts with a video over time. The agent observes video frames and decides both where to look next and when to emit a prediction. Since backpropagation is not adequate in this non-differentiable setting, we use REINFORCE to learn the agent's decision policy. Our model achieves state-of-the-art results on the THUMOS'14 and ActivityNet datasets while observing only a fraction (2 or less) of the video frames.",
"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."
]
} |
1704.03615 | 2606461407 | While deep feature learning has revolutionized techniques for static-image understanding, the same does not quite hold for video processing. Architectures and optimization techniques used for video are largely based off those for static images, potentially underutilizing rich video information. In this work, we rethink both the underlying network architecture and the stochastic learning paradigm for temporal data. To do so, we draw inspiration from classic theory on linear dynamic systems for modeling time series. By extending such models to include nonlinear mappings, we derive a series of novel recurrent neural networks that sequentially make top-down predictions about the future and then correct those predictions with bottom-up observations. Predictive-corrective networks have a number of desirable properties: (1) they can adaptively focus computation on surprising frames where predictions require large corrections, (2) they simplify learning in that only residual-like corrective terms need to be learned over time and (3) they naturally decorrelate an input data stream in a hierarchical fashion, producing a more reliable signal for learning at each layer of a network. We provide an extensive analysis of our lightweight and interpretable framework, and demonstrate that our model is competitive with the two-stream network on three challenging datasets without the need for computationally expensive optical flow. | Predictive models It has been shown that leveraging global contextual information can be used to improve image @cite_1 or video @cite_38 understanding. Recent work has examined predicting the appearance and semantic content of future video frames @cite_37 @cite_44 @cite_0 @cite_57 @cite_41 @cite_19 @cite_35 . Recently @cite_20 train a recurrent neural network in an encoder-decoder fashion to predict future frames, and demonstrates that the learned video representation improves action recognition accuracy. However, to the best of our knowledge these insights have not been used for designing accurate end-to-end action localization models. | {
"cite_N": [
"@cite_38",
"@cite_37",
"@cite_35",
"@cite_41",
"@cite_1",
"@cite_44",
"@cite_0",
"@cite_57",
"@cite_19",
"@cite_20"
],
"mid": [
"2163292664",
"2422305492",
"2963125871",
"2963402657",
"2159564241",
"2520707650",
"",
"2963547393",
"2964151830",
"2952453038"
],
"abstract": [
"This paper exploits the context of natural dynamic scenes for human action recognition in video. Human actions are frequently constrained by the purpose and the physical properties of scenes and demonstrate high correlation with particular scene classes. For example, eating often happens in a kitchen while running is more common outdoors. The contribution of this paper is three-fold: (a) we automatically discover relevant scene classes and their correlation with human actions, (b) we show how to learn selected scene classes from video without manual supervision and (c) we develop a joint framework for action and scene recognition and demonstrate improved recognition of both in natural video. We use movie scripts as a means of automatic supervision for training. For selected action classes we identify correlated scene classes in text and then retrieve video samples of actions and scenes for training using script-to-video alignment. Our visual models for scenes and actions are formulated within the bag-of-features framework and are combined in a joint scene-action SVM-based classifier. We report experimental results and validate the method on a new large dataset with twelve action classes and ten scene classes acquired from 69 movies.",
"Anticipating actions and objects before they start or appear is a difficult problem in computer vision with several real-world applications. This task is challenging partly because it requires leveraging extensive knowledge of the world that is difficult to write down. We believe that a promising resource for efficiently learning this knowledge is through readily available unlabeled video. We present a framework that capitalizes on temporal structure in unlabeled video to learn to anticipate human actions and objects. The key idea behind our approach is that we can train deep networks to predict the visual representation of images in the future. Visual representations are a promising prediction target because they encode images at a higher semantic level than pixels yet are automatic to compute. We then apply recognition algorithms on our predicted representation to anticipate objects and actions. We experimentally validate this idea on two datasets, anticipating actions one second in the future and objects five seconds in the future.",
"Abstract: Learning to predict future images from a video sequence involves the construction of an internal representation that models the image evolution accurately, and therefore, to some degree, its content and dynamics. This is why pixel-space video prediction may be viewed as a promising avenue for unsupervised feature learning. In addition, while optical flow has been a very studied problem in computer vision for a long time, future frame prediction is rarely approached. Still, many vision applications could benefit from the knowledge of the next frames of videos, that does not require the complexity of tracking every pixel trajectories. In this work, we train a convolutional network to generate future frames given an input sequence. To deal with the inherently blurry predictions obtained from the standard Mean Squared Error (MSE) loss function, we propose three different and complementary feature learning strategies: a multi-scale architecture, an adversarial training method, and an image gradient difference loss function. We compare our predictions to different published results based on recurrent neural networks on the UCF101 dataset",
"A core challenge for an agent learning to interact with the world is to predict how its actions affect objects in its environment. Many existing methods for learning the dynamics of physical interactions require labeled object information. However, to scale real-world interaction learning to a variety of scenes and objects, acquiring labeled data becomes increasingly impractical. To learn about physical object motion without labels, we develop an action-conditioned video prediction model that explicitly models pixel motion, by predicting a distribution over pixel motion from previous frames. Because our model explicitly predicts motion, it is partially invariant to object appearance, enabling it to generalize to previously unseen objects. To explore video prediction for real-world interactive agents, we also introduce a dataset of 59,000 robot interactions involving pushing motions, including a test set with novel objects. In this dataset, accurate prediction of videos conditioned on the robot's future actions amounts to learning a \"visual imagination\" of different futures based on different courses of action. Our experiments show that our proposed method produces more accurate video predictions both quantitatively and qualitatively, when compared to prior methods.",
"In the real world, objects never occur in isolation; they co-vary with other objects and particular environments, providing a rich source of contextual associations to be exploited by the visual system. A natural way of representing the context of an object is in terms of its relationship to other objects. Alternately, recent work has shown that a statistical summary of the scene provides a complementary and effective source of information for contextual inference, which enables humans to quickly guide their attention and eyes to regions of interest in natural scenes. A better understanding of how humans build such scene representations, and of the mechanisms of contextual analysis, will lead to a new generation of computer vision systems.",
"We capitalize on large amounts of unlabeled video in order to learn a model of scene dynamics for both video recognition tasks (e.g. action classification) and video generation tasks (e.g. future prediction). We propose a generative adversarial network for video with a spatio-temporal convolutional architecture that untangles the scene's foreground from the background. Experiments suggest this model can generate tiny videos up to a second at full frame rate better than simple baselines, and we show its utility at predicting plausible futures of static images. Moreover, experiments and visualizations show the model internally learns useful features for recognizing actions with minimal supervision, suggesting scene dynamics are a promising signal for representation learning. We believe generative video models can impact many applications in video understanding and simulation.",
"",
"We study the problem of synthesizing a number of likely future frames from a single input image. In contrast to traditional methods, which have tackled this problem in a deterministic or non-parametric way, we propose to model future frames in a probabilistic manner. Our probabilistic model makes it possible for us to sample and synthesize many possible future frames from a single input image. To synthesize realistic movement of objects, we propose a novel network structure, namely a Cross Convolutional Network; this network encodes image and motion information as feature maps and convolutional kernels, respectively. In experiments, our model performs well on synthetic data, such as 2D shapes and animated game sprites, as well as on real-world video frames. We also show that our model can be applied to visual analogy-making, and present an analysis of the learned network representations.",
"While great strides have been made in using deep learning algorithms to solve supervised learning tasks, the problem of unsupervised learning - leveraging unlabeled examples to learn about the structure of a domain - remains a difficult unsolved challenge. Here, we explore prediction of future frames in a video sequence as an unsupervised learning rule for learning about the structure of the visual world. We describe a predictive neural network (\"PredNet\") architecture that is inspired by the concept of \"predictive coding\" from the neuroscience literature. These networks learn to predict future frames in a video sequence, with each layer in the network making local predictions and only forwarding deviations from those predictions to subsequent network layers. We show that these networks are able to robustly learn to predict the movement of synthetic (rendered) objects, and that in doing so, the networks learn internal representations that are useful for decoding latent object parameters (e.g. pose) that support object recognition with fewer training views. We also show that these networks can scale to complex natural image streams (car-mounted camera videos), capturing key aspects of both egocentric movement and the movement of objects in the visual scene, and the representation learned in this setting is useful for estimating the steering angle. These results suggest that prediction represents a powerful framework for unsupervised learning, allowing for implicit learning of object and scene structure.",
"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."
]
} |
1704.03615 | 2606461407 | While deep feature learning has revolutionized techniques for static-image understanding, the same does not quite hold for video processing. Architectures and optimization techniques used for video are largely based off those for static images, potentially underutilizing rich video information. In this work, we rethink both the underlying network architecture and the stochastic learning paradigm for temporal data. To do so, we draw inspiration from classic theory on linear dynamic systems for modeling time series. By extending such models to include nonlinear mappings, we derive a series of novel recurrent neural networks that sequentially make top-down predictions about the future and then correct those predictions with bottom-up observations. Predictive-corrective networks have a number of desirable properties: (1) they can adaptively focus computation on surprising frames where predictions require large corrections, (2) they simplify learning in that only residual-like corrective terms need to be learned over time and (3) they naturally decorrelate an input data stream in a hierarchical fashion, producing a more reliable signal for learning at each layer of a network. We provide an extensive analysis of our lightweight and interpretable framework, and demonstrate that our model is competitive with the two-stream network on three challenging datasets without the need for computationally expensive optical flow. | Accelerating learning Recurrent neural nets are notoriously difficult to train because of the exploding gradients encountered during SGD @cite_26 . We refer the reader to @cite_25 for an excellent introduction to general SGD learning. Though naturally a sequential algorithm that processes one data example at a time, much recent work focuses on mini-batch methods that can exploit parallelism in GPU architectures or clusters @cite_10 . One general theme is efficient online approximation of second-order methods @cite_17 , which can model correlations between input features. Batch normalization @cite_56 computes correlation statistics between samples in a batch, speeding up convergence. Predictive-corrective networks naturally de-correlate batch statistics without needing expensive second-order computations (Fig. ). | {
"cite_N": [
"@cite_26",
"@cite_56",
"@cite_10",
"@cite_25",
"@cite_17"
],
"mid": [
"1815076433",
"1836465849",
"2168231600",
"114517082",
"2137515395"
],
"abstract": [
"There are two widely known issues with properly training recurrent neural networks, the vanishing and the exploding gradient problems detailed in (1994). In this paper we attempt to improve the understanding of the underlying issues by exploring these problems from an analytical, a geometric and a dynamical systems perspective. Our analysis is used to justify a simple yet effective solution. We propose a gradient norm clipping strategy to deal with exploding gradients and a soft constraint for the vanishing gradients problem. We validate empirically our hypothesis and proposed solutions in the experimental section.",
"Training Deep Neural Networks is complicated by the fact that the distribution of each layer's inputs changes during training, as the parameters of the previous layers change. This slows down the training by requiring lower learning rates and careful parameter initialization, and makes it notoriously hard to train models with saturating nonlinearities. We refer to this phenomenon as internal covariate shift, and address the problem by normalizing layer inputs. Our method draws its strength from making normalization a part of the model architecture and performing the normalization for each training mini-batch. Batch Normalization allows us to use much higher learning rates and be less careful about initialization, and in some cases eliminates the need for Dropout. Applied to a state-of-the-art image classification model, Batch Normalization achieves the same accuracy with 14 times fewer training steps, and beats the original model by a significant margin. Using an ensemble of batch-normalized networks, we improve upon the best published result on ImageNet classification: reaching 4.82 top-5 test error, exceeding the accuracy of human raters.",
"Recent work in unsupervised feature learning and deep learning has shown that being able to train large models can dramatically improve performance. In this paper, we consider the problem of training a deep network with billions of parameters using tens of thousands of CPU cores. We have developed a software framework called DistBelief that can utilize computing clusters with thousands of machines to train large models. Within this framework, we have developed two algorithms for large-scale distributed training: (i) Downpour SGD, an asynchronous stochastic gradient descent procedure supporting a large number of model replicas, and (ii) Sandblaster, a framework that supports a variety of distributed batch optimization procedures, including a distributed implementation of L-BFGS. Downpour SGD and Sandblaster L-BFGS both increase the scale and speed of deep network training. We have successfully used our system to train a deep network 30x larger than previously reported in the literature, and achieves state-of-the-art performance on ImageNet, a visual object recognition task with 16 million images and 21k categories. We show that these same techniques dramatically accelerate the training of a more modestly- sized deep network for a commercial speech recognition service. Although we focus on and report performance of these methods as applied to training large neural networks, the underlying algorithms are applicable to any gradient-based machine learning algorithm.",
"During the last decade, the data sizes have grown faster than the speed of processors. In this context, the capabilities of statistical machine learning methods is limited by the computing time rather than the sample size. A more precise analysis uncovers qualitatively different tradeoffs for the case of small-scale and large-scale learning problems. The large-scale case involves the computational complexity of the underlying optimization algorithm in non-trivial ways. Unlikely optimization algorithms such as stochastic gradient descent show amazing performance for large-scale problems. In particular, second order stochastic gradient and averaged stochastic gradient are asymptotically efficient after a single pass on the training set.",
"The SGD-QN algorithm is a stochastic gradient descent algorithm that makes careful use of second-order information and splits the parameter update into independently scheduled components. Thanks to this design, SGD-QN iterates nearly as fast as a first-order stochastic gradient descent but requires less iterations to achieve the same accuracy. This algorithm won the \"Wild Track\" of the first PASCAL Large Scale Learning Challenge (, 2008)."
]
} |
1704.03615 | 2606461407 | While deep feature learning has revolutionized techniques for static-image understanding, the same does not quite hold for video processing. Architectures and optimization techniques used for video are largely based off those for static images, potentially underutilizing rich video information. In this work, we rethink both the underlying network architecture and the stochastic learning paradigm for temporal data. To do so, we draw inspiration from classic theory on linear dynamic systems for modeling time series. By extending such models to include nonlinear mappings, we derive a series of novel recurrent neural networks that sequentially make top-down predictions about the future and then correct those predictions with bottom-up observations. Predictive-corrective networks have a number of desirable properties: (1) they can adaptively focus computation on surprising frames where predictions require large corrections, (2) they simplify learning in that only residual-like corrective terms need to be learned over time and (3) they naturally decorrelate an input data stream in a hierarchical fashion, producing a more reliable signal for learning at each layer of a network. We provide an extensive analysis of our lightweight and interpretable framework, and demonstrate that our model is competitive with the two-stream network on three challenging datasets without the need for computationally expensive optical flow. | Interpretable models Understanding the inner workings of models is important for diagnosing and correcting mistakes @cite_22 @cite_14 . However, despite some recent progress on this front @cite_8 , recurrent neural networks largely remain a mystery. By introducing a lightweight interpretable recurrent model we aim to gain some insight into the critical components of accurate and efficient video processing. | {
"cite_N": [
"@cite_14",
"@cite_22",
"@cite_8"
],
"mid": [
"1849277567",
"",
"1951216520"
],
"abstract": [
"Large Convolutional Network models have recently demonstrated impressive classification performance on the ImageNet benchmark [18]. However there is no clear understanding of why they perform so well, or how they might be improved. In this paper we explore both issues. We introduce a novel visualization technique that gives insight into the function of intermediate feature layers and the operation of the classifier. Used in a diagnostic role, these visualizations allow us to find model architectures that outperform on the ImageNet classification benchmark. We also perform an ablation study to discover the performance contribution from different model layers. We show our ImageNet model generalizes well to other datasets: when the softmax classifier is retrained, it convincingly beats the current state-of-the-art results on Caltech-101 and Caltech-256 datasets.",
"",
"Recurrent Neural Networks (RNNs), and specifically a variant with Long Short-Term Memory (LSTM), are enjoying renewed interest as a result of successful applications in a wide range of machine learning problems that involve sequential data. However, while LSTMs provide exceptional results in practice, the source of their performance and their limitations remain rather poorly understood. Using character-level language models as an interpretable testbed, we aim to bridge this gap by providing an analysis of their representations, predictions and error types. In particular, our experiments reveal the existence of interpretable cells that keep track of long-range dependencies such as line lengths, quotes and brackets. Moreover, our comparative analysis with finite horizon n-gram models traces the source of the LSTM improvements to long-range structural dependencies. Finally, we provide analysis of the remaining errors and suggests areas for further study."
]
} |
1704.03607 | 2952749830 | Attribute-based recognition models, due to their impressive performance and their ability to generalize well on novel categories, have been widely adopted for many computer vision applications. However, usually both the attribute vocabulary and the class-attribute associations have to be provided manually by domain experts or large number of annotators. This is very costly and not necessarily optimal regarding recognition performance, and most importantly, it limits the applicability of attribute-based models to large scale data sets. To tackle this problem, we propose an end-to-end unsupervised attribute learning approach. We utilize online text corpora to automatically discover a salient and discriminative vocabulary that correlates well with the human concept of semantic attributes. Moreover, we propose a deep convolutional model to optimize class-attribute associations with a linguistic prior that accounts for noise and missing data in text. In a thorough evaluation on ImageNet, we demonstrate that our model is able to efficiently discover and learn semantic attributes at a large scale. Furthermore, we demonstrate that our model outperforms the state-of-the-art in zero-shot learning on three data sets: ImageNet, Animals with Attributes and aPascal aYahoo. Finally, we enable attribute-based learning on ImageNet and will share the attributes and associations for future research. | There were few attempts in the literature to automatically obtain an attribute vocabulary. @cite_21 @cite_41 mine attributes by crawling the WordNet @cite_33 ontology. Specifically, they track the has-part'' relations in WordNet to extract part'' attributes. On the other hand, @cite_17 @cite_44 @cite_7 use the top ranked images returned by web search engines queried with a certain vocabulary to estimate the visualness'' of words. @cite_30 samples pairs of (image, description) from the Internet to automatically find a set of visual attributes. Similarly, @cite_8 uses both image-based textual descriptions as well as a set of image tags provided by users in social media to identify the attribute vocabulary. Different from previous work, our approach does not require images aligned with textual descriptions or tags. Furthermore, we do not rely on a predefined ontology such as WordNet or target only a specific type of attributes like parts''. Instead, we use textual description at the category level in form of encyclopedia entries to extract a salient and diverse set of attributes. | {
"cite_N": [
"@cite_30",
"@cite_33",
"@cite_7",
"@cite_8",
"@cite_41",
"@cite_21",
"@cite_44",
"@cite_17"
],
"mid": [
"1528802670",
"2081580037",
"2081613070",
"2482018248",
"2077071968",
"1992454046",
"1964763677",
"2125560515"
],
"abstract": [
"It is common to use domain specific terminology - attributes - to describe the visual appearance of objects. In order to scale the use of these describable visual attributes to a large number of categories, especially those not well studied by psychologists or linguists, it will be necessary to find alternative techniques for identifying attribute vocabularies and for learning to recognize attributes without hand labeled training data. We demonstrate that it is possible to accomplish both these tasks automatically by mining text and image data sampled from the Internet. The proposed approach also characterizes attributes according to their visual representation: global or local, and type: color, texture, or shape. This work focuses on discovering attributes and their visual appearance, and is as agnostic as possible about the textual description.",
"Because meaningful sentences are composed of meaningful words, any system that hopes to process natural languages as people do must have information about words and their meanings. This information is traditionally provided through dictionaries, and machine-readable dictionaries are now widely available. But dictionary entries evolved for the convenience of human readers, not for machines. WordNet 1 provides a more effective combination of traditional lexicographic information and modern computing. WordNet is an online lexical database designed for use under program control. English nouns, verbs, adjectives, and adverbs are organized into sets of synonyms, each representing a lexicalized concept. Semantic relations link the synonym sets [4].",
"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.",
"How can a machine learn to recognize visual attributes emerging out of online community without a definitive supervised dataset? This paper proposes an automatic approach to discover and analyze visual attributes from a noisy collection of image-text data on the Web. Our approach is based on the relationship between attributes and neural activations in the deep network. We characterize the visual property of the attribute word as a divergence within weakly-annotated set of images. We show that the neural activations are useful for discovering and learning a classifier that well agrees with human perception from the noisy real-world Web data. The empirical study suggests the layered structure of the deep neural networks also gives us insights into the perceptual depth of the given word. Finally, we demonstrate that we can utilize highly-activating neurons for finding semantically relevant regions.",
"While knowledge transfer (KT) between object classes has been accepted as a promising route towards scalable recognition, most experimental KT studies are surprisingly limited in the number of object classes considered. To support claims of KT w.r.t. scalability we thus advocate to evaluate KT in a large-scale setting. To this end, we provide an extensive evaluation of three popular approaches to KT on a recently proposed large-scale data set, the ImageNet Large Scale Visual Recognition Competition 2010 data set. In a first setting they are directly compared to one-vs-all classification often neglected in KT papers and in a second setting we evaluate their ability to enable zero-shot learning. While none of the KT methods can improve over one-vs-all classification they prove valuable for zero-shot learning, especially hierarchical and direct similarity based KT. We also propose and describe several extensions of the evaluated approaches that are necessary for this large-scale study.",
"Remarkable performance has been reported to recognize single object classes. Scalability to large numbers of classes however remains an important challenge for today's recognition methods. Several authors have promoted knowledge transfer between classes as a key ingredient to address this challenge. However, in previous work the decision which knowledge to transfer has required either manual supervision or at least a few training examples limiting the scalability of these approaches. In this work we explicitly address the question of how to automatically decide which information to transfer between classes without the need of any human intervention. For this we tap into linguistic knowledge bases to provide the semantic link between sources (what) and targets (where) of knowledge transfer. We provide a rigorous experimental evaluation of different knowledge bases and state-of-the-art techniques from Natural Language Processing which goes far beyond the limited use of language in related work. We also give insights into the applicability (why) of different knowledge sources and similarity measures for knowledge transfer.",
"We propose NEIL (Never Ending Image Learner), a computer program that runs 24 hours per day and 7 days per week to automatically extract visual knowledge from Internet data. NEIL uses a semi-supervised learning algorithm that jointly discovers common sense relationships (e.g., \"Corolla is a kind of looks similar to Car\", \"Wheel is a part of Car\") and labels instances of the given visual categories. It is an attempt to develop the world's largest visual structured knowledge base with minimum human labeling effort. As of 10th October 2013, NEIL has been continuously running for 2.5 months on 200 core cluster (more than 350K CPU hours) and has an ontology of 1152 object categories, 1034 scene categories and 87 attributes. During this period, NEIL has discovered more than 1700 relationships and has labeled more than 400K visual instances.",
"We present a probabilistic generative model of visual attributes, together with an efficient learning algorithm. Attributes are visual qualities of objects, such as 'red', 'striped', or 'spotted'. The model sees attributes as patterns of image segments, repeatedly sharing some characteristic properties. These can be any combination of appearance, shape, or the layout of segments within the pattern. Moreover, attributes with general appearance are taken into account, such as the pattern of alternation of any two colors which is characteristic for stripes. To enable learning from unsegmented training images, the model is learnt discriminatively, by optimizing a likelihood ratio. As demonstrated in the experimental evaluation, our model can learn in a weakly supervised setting and encompasses a broad range of attributes. We show that attributes can be learnt starting from a text query to Google image search, and can then be used to recognize the attribute and determine its spatial extent in novel real-world images."
]
} |
1704.03607 | 2952749830 | Attribute-based recognition models, due to their impressive performance and their ability to generalize well on novel categories, have been widely adopted for many computer vision applications. However, usually both the attribute vocabulary and the class-attribute associations have to be provided manually by domain experts or large number of annotators. This is very costly and not necessarily optimal regarding recognition performance, and most importantly, it limits the applicability of attribute-based models to large scale data sets. To tackle this problem, we propose an end-to-end unsupervised attribute learning approach. We utilize online text corpora to automatically discover a salient and discriminative vocabulary that correlates well with the human concept of semantic attributes. Moreover, we propose a deep convolutional model to optimize class-attribute associations with a linguistic prior that accounts for noise and missing data in text. In a thorough evaluation on ImageNet, we demonstrate that our model is able to efficiently discover and learn semantic attributes at a large scale. Furthermore, we demonstrate that our model outperforms the state-of-the-art in zero-shot learning on three data sets: ImageNet, Animals with Attributes and aPascal aYahoo. Finally, we enable attribute-based learning on ImageNet and will share the attributes and associations for future research. | In a different direction, other approaches focused on predicting the class-attribute associations automatically. @cite_21 @cite_3 estimate the associations strength from web-based co-occurrence statistics. However, web-based hit counts estimations are noisy since it does not take into consideration the context or the specific relation sought between the category and the attribute. @cite_49 uses WordNet hierarchy to transfer the attribute associations of an unseen class from its parent in the ontology. Recently, @cite_36 proposes to predict attribute associations using semantic relations in a tensor factorization approach. However, both @cite_49 and @cite_36 assume the availability of training associations. Here, we propose a deep model to estimate the class-attribute associations from scratch. Our model takes advantage of an initial linguistic prior over the associations from textual description and improves the estimations in a joint optimization framework of object and attributes predictions. | {
"cite_N": [
"@cite_36",
"@cite_21",
"@cite_49",
"@cite_3"
],
"mid": [
"2949729859",
"1992454046",
"2005285092",
""
],
"abstract": [
"Collecting training images for all visual categories is not only expensive but also impractical. Zero-shot learning (ZSL), especially using attributes, offers a pragmatic solution to this problem. However, at test time most attribute-based methods require a full description of attribute associations for each unseen class. Providing these associations is time consuming and often requires domain specific knowledge. In this work, we aim to carry out attribute-based zero-shot classification in an unsupervised manner. We propose an approach to learn relations that couples class embeddings with their corresponding attributes. Given only the name of an unseen class, the learned relationship model is used to automatically predict the class-attribute associations. Furthermore, our model facilitates transferring attributes across data sets without additional effort. Integrating knowledge from multiple sources results in a significant additional improvement in performance. We evaluate on two public data sets: Animals with Attributes and aPascal aYahoo. Our approach outperforms state-of-the-art methods in both predicting class-attribute associations and unsupervised ZSL by a large margin.",
"Remarkable performance has been reported to recognize single object classes. Scalability to large numbers of classes however remains an important challenge for today's recognition methods. Several authors have promoted knowledge transfer between classes as a key ingredient to address this challenge. However, in previous work the decision which knowledge to transfer has required either manual supervision or at least a few training examples limiting the scalability of these approaches. In this work we explicitly address the question of how to automatically decide which information to transfer between classes without the need of any human intervention. For this we tap into linguistic knowledge bases to provide the semantic link between sources (what) and targets (where) of knowledge transfer. We provide a rigorous experimental evaluation of different knowledge bases and state-of-the-art techniques from Natural Language Processing which goes far beyond the limited use of language in related work. We also give insights into the applicability (why) of different knowledge sources and similarity measures for knowledge transfer.",
"Attribute based knowledge transfer has proven very successful in visual object analysis and learning previously unseen classes. However, the common approach learns and transfers attributes without taking into consideration the embedded structure between the categories in the source set. Such information provides important cues on the intraattribute variations. We propose to capture these variations in a hierarchical model that expands the knowledge source with additional abstraction levels of attributes. We also provide a novel transfer approach that can choose the appropriate attributes to be shared with an unseen class. We evaluate our approach on three public datasets: a Pascal, Animals with Attributes and CUB-200-2011 Birds. The experiments demonstrate the effectiveness of our model with significant improvement over state-of-the-art.",
""
]
} |
1704.03607 | 2952749830 | Attribute-based recognition models, due to their impressive performance and their ability to generalize well on novel categories, have been widely adopted for many computer vision applications. However, usually both the attribute vocabulary and the class-attribute associations have to be provided manually by domain experts or large number of annotators. This is very costly and not necessarily optimal regarding recognition performance, and most importantly, it limits the applicability of attribute-based models to large scale data sets. To tackle this problem, we propose an end-to-end unsupervised attribute learning approach. We utilize online text corpora to automatically discover a salient and discriminative vocabulary that correlates well with the human concept of semantic attributes. Moreover, we propose a deep convolutional model to optimize class-attribute associations with a linguistic prior that accounts for noise and missing data in text. In a thorough evaluation on ImageNet, we demonstrate that our model is able to efficiently discover and learn semantic attributes at a large scale. Furthermore, we demonstrate that our model outperforms the state-of-the-art in zero-shot learning on three data sets: ImageNet, Animals with Attributes and aPascal aYahoo. Finally, we enable attribute-based learning on ImageNet and will share the attributes and associations for future research. | Semantic attributes -- with their ability to be shared across categories -- have shown impressive performance in tasks like ZSL. However, due to their limited scalability, there is an increasing interest in conducting ZSL by tapping to an alternative knowledge source, for example by exploiting lexical hierarchies to transfer visual models between the categories @cite_41 or learning a hierarchical embedding @cite_26 . A different direction leverage powerful word embeddings @cite_43 @cite_40 to establish the semantic link between seen and unseen categories @cite_42 @cite_48 . More closely to our work are the ones from @cite_24 @cite_10 and @cite_0 . These approaches use article embeddings to directly predict the classifier weights of the novel category in a domain adaption framework. However, most of the unsupervised ZSL approaches do not result in good discriminative classifiers when compared to their attribute-based counterpart @cite_36 . We show in the evaluation, that our unsupervised deep attribute model can predict novel categories with high accuracy and it outperforms state-of-the-art in unsupervised ZSL with a significant margin. | {
"cite_N": [
"@cite_26",
"@cite_41",
"@cite_48",
"@cite_36",
"@cite_42",
"@cite_24",
"@cite_43",
"@cite_40",
"@cite_0",
"@cite_10"
],
"mid": [
"",
"2077071968",
"2950700180",
"2949729859",
"",
"",
"2164019165",
"1614298861",
"2341047870",
"870084106"
],
"abstract": [
"",
"While knowledge transfer (KT) between object classes has been accepted as a promising route towards scalable recognition, most experimental KT studies are surprisingly limited in the number of object classes considered. To support claims of KT w.r.t. scalability we thus advocate to evaluate KT in a large-scale setting. To this end, we provide an extensive evaluation of three popular approaches to KT on a recently proposed large-scale data set, the ImageNet Large Scale Visual Recognition Competition 2010 data set. In a first setting they are directly compared to one-vs-all classification often neglected in KT papers and in a second setting we evaluate their ability to enable zero-shot learning. While none of the KT methods can improve over one-vs-all classification they prove valuable for zero-shot learning, especially hierarchical and direct similarity based KT. We also propose and describe several extensions of the evaluated approaches that are necessary for this large-scale study.",
"Several recent publications have proposed methods for mapping images into continuous semantic embedding spaces. In some cases the embedding space is trained jointly with the image transformation. In other cases the semantic embedding space is established by an independent natural language processing task, and then the image transformation into that space is learned in a second stage. Proponents of these image embedding systems have stressed their advantages over the traditional classification framing of image understanding, particularly in terms of the promise for zero-shot learning -- the ability to correctly annotate images of previously unseen object categories. In this paper, we propose a simple method for constructing an image embedding system from any existing image classifier and a semantic word embedding model, which contains the @math class labels in its vocabulary. Our method maps images into the semantic embedding space via convex combination of the class label embedding vectors, and requires no additional training. We show that this simple and direct method confers many of the advantages associated with more complex image embedding schemes, and indeed outperforms state of the art methods on the ImageNet zero-shot learning task.",
"Collecting training images for all visual categories is not only expensive but also impractical. Zero-shot learning (ZSL), especially using attributes, offers a pragmatic solution to this problem. However, at test time most attribute-based methods require a full description of attribute associations for each unseen class. Providing these associations is time consuming and often requires domain specific knowledge. In this work, we aim to carry out attribute-based zero-shot classification in an unsupervised manner. We propose an approach to learn relations that couples class embeddings with their corresponding attributes. Given only the name of an unseen class, the learned relationship model is used to automatically predict the class-attribute associations. Furthermore, our model facilitates transferring attributes across data sets without additional effort. Integrating knowledge from multiple sources results in a significant additional improvement in performance. We evaluate on two public data sets: Animals with Attributes and aPascal aYahoo. Our approach outperforms state-of-the-art methods in both predicting class-attribute associations and unsupervised ZSL by a large margin.",
"",
"",
"Unsupervised word representations are very useful in NLP tasks both as inputs to learning algorithms and as extra word features in NLP systems. However, most of these models are built with only local context and one representation per word. This is problematic because words are often polysemous and global context can also provide useful information for learning word meanings. We present a new neural network architecture which 1) learns word embeddings that better capture the semantics of words by incorporating both local and global document context, and 2) accounts for homonymy and polysemy by learning multiple embeddings per word. We introduce a new dataset with human judgments on pairs of words in sentential context, and evaluate our model on it, showing that our model outperforms competitive baselines and other neural language models.",
"",
"Classifying a visual concept merely from its associated online textual source, such as a Wikipedia article, is an attractive research topic in zero-shot learning because it alleviates the burden of manually collecting semantic attributes. Several recent works have pursued this approach by exploring various ways of connecting the visual and text domains. This paper revisits this idea by stepping further to consider one important factor: the textual representation is usually too noisy for the zero-shot learning application. This consideration motivates us to design a simple-but-effective zero-shot learning method capable of suppressing noise in the text. More specifically, we propose an @math -norm based objective function which can simultaneously suppress the noisy signal in the text and learn a function to match the text document and visual features. We also develop an optimization algorithm to efficiently solve the resulting problem. By conducting experiments on two large datasets, we demonstrate that the proposed method significantly outperforms the competing methods which rely on online information sources but without explicit noise suppression. We further make an in-depth analysis of the proposed method and provide insight as to what kind of information in documents is useful for zero-shot learning.",
"One of the main challenges in Zero-Shot Learning of visual categories is gathering semantic attributes to accompany images. Recent work has shown that learning from textual descriptions, such as Wikipedia articles, avoids the problem of having to explicitly define these attributes. We present a new model that can classify unseen categories from their textual description. Specifically, we use text features to predict the output weights of both the convolutional and the fully connected layers in a deep convolutional neural network (CNN). We take advantage of the architecture of CNNs and learn features at different layers, rather than just learning an embedding space for both modalities, as is common with existing approaches. The proposed model also allows us to automatically generate a list of pseudo- attributes for each visual category consisting of words from Wikipedia articles. We train our models end-to-end us- ing the Caltech-UCSD bird and flower datasets and evaluate both ROC and Precision-Recall curves. Our empirical results show that the proposed model significantly outperforms previous methods."
]
} |
1704.03012 | 2606433045 | Deep reinforcement learning has achieved many impressive results in recent years. However, tasks with sparse rewards or long horizons continue to pose significant challenges. To tackle these important problems, we propose a general framework that first learns useful skills in a pre-training environment, and then leverages the acquired skills for learning faster in downstream tasks. Our approach brings together some of the strengths of intrinsic motivation and hierarchical methods: the learning of useful skill is guided by a single proxy reward, the design of which requires very minimal domain knowledge about the downstream tasks. Then a high-level policy is trained on top of these skills, providing a significant improvement of the exploration and allowing to tackle sparse rewards in the downstream tasks. To efficiently pre-train a large span of skills, we use Stochastic Neural Networks combined with an information-theoretic regularizer. Our experiments show that this combination is effective in learning a wide span of interpretable skills in a sample-efficient way, and can significantly boost the learning performance uniformly across a wide range of downstream tasks. | One of the main appealing aspects of hierarchical reinforcement learning (HRL) is to use skills to reduce the search complexity of the problem . However, specifying a good hierarchy by hand requires domain-specific knowledge and careful engineering, hence motivating the need for learning skills automatically. Prior work on automatic learning of skills has largely focused on learning skills in discrete domains . A popular approach there is to use statistics about state transitions to identify bottleneck states . It is however not clear how these techniques can be applied to settings with high-dimensional continuous spaces. More recently, @cite_4 propose a DRAW-like recurrent neural network architecture that can learn temporally extended macro actions. However, the learning needs to be guided by external rewards as supervisory signals, and hence the algorithm cannot be straightforwardly applied in sparse reward settings. | {
"cite_N": [
"@cite_4"
],
"mid": [
"2442341664"
],
"abstract": [
"We present a novel deep recurrent neural network architecture that learns to build implicit plans in an end-to-end manner by purely interacting with an environment in reinforcement learning setting. The network builds an internal plan, which is continuously updated upon observation of the next input from the environment. It can also partition this internal representation into contiguous sub- sequences by learning for how long the plan can be committed to - i.e. followed without re-planing. Combining these properties, the proposed model, dubbed STRategic Attentive Writer (STRAW) can learn high-level, temporally abstracted macro- actions of varying lengths that are solely learnt from data without any prior information. These macro-actions enable both structured exploration and economic computation. We experimentally demonstrate that STRAW delivers strong improvements on several ATARI games by employing temporally extended planning strategies (e.g. Ms. Pacman and Frostbite). It is at the same time a general algorithm that can be applied on any sequence data. To that end, we also show that when trained on text prediction task, STRAW naturally predicts frequent n-grams (instead of macro-actions), demonstrating the generality of the approach."
]
} |
1704.03012 | 2606433045 | Deep reinforcement learning has achieved many impressive results in recent years. However, tasks with sparse rewards or long horizons continue to pose significant challenges. To tackle these important problems, we propose a general framework that first learns useful skills in a pre-training environment, and then leverages the acquired skills for learning faster in downstream tasks. Our approach brings together some of the strengths of intrinsic motivation and hierarchical methods: the learning of useful skill is guided by a single proxy reward, the design of which requires very minimal domain knowledge about the downstream tasks. Then a high-level policy is trained on top of these skills, providing a significant improvement of the exploration and allowing to tackle sparse rewards in the downstream tasks. To efficiently pre-train a large span of skills, we use Stochastic Neural Networks combined with an information-theoretic regularizer. Our experiments show that this combination is effective in learning a wide span of interpretable skills in a sample-efficient way, and can significantly boost the learning performance uniformly across a wide range of downstream tasks. | There has also been work on learning skills in tasks with continuous actions . These methods extract useful skills from successful trajectories for the same (or closely related) tasks, and hence require first solving a comparably challenging task or demonstrations. Guided Policy Search (GPS) @cite_3 leverages access to training in a simpler setting. GPS trains with iLQG @cite_6 in state space, and in parallel trains a neural net that only receives raw sensory inputs that has to agree with the iLQG controller. The neural net policy is then able to generalize to new situations. | {
"cite_N": [
"@cite_6",
"@cite_3"
],
"mid": [
"2167856595",
"2964161785"
],
"abstract": [
"We present an iterative linear-quadratic-Gaussian method for locally-optimal feedback control of nonlinear stochastic systems subject to control constraints. Previously, similar methods have been restricted to deterministic unconstrained problems with quadratic costs. The new method constructs an affine feedback control law, obtained by minimizing a novel quadratic approximation to the optimal cost-to-go function. Global convergence is guaranteed through a Levenberg-Marquardt method; convergence in the vicinity of a local minimum is quadratic. Performance is illustrated on a limited-torque inverted pendulum problem, as well as a complex biomechanical control problem involving a stochastic model of the human arm, with 10 state dimensions and 6 muscle actuators. A Matlab implementation of the new algorithm is availabe at www.cogsci.ucsd.edu spl sim todorov.",
"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."
]
} |
1704.03012 | 2606433045 | Deep reinforcement learning has achieved many impressive results in recent years. However, tasks with sparse rewards or long horizons continue to pose significant challenges. To tackle these important problems, we propose a general framework that first learns useful skills in a pre-training environment, and then leverages the acquired skills for learning faster in downstream tasks. Our approach brings together some of the strengths of intrinsic motivation and hierarchical methods: the learning of useful skill is guided by a single proxy reward, the design of which requires very minimal domain knowledge about the downstream tasks. Then a high-level policy is trained on top of these skills, providing a significant improvement of the exploration and allowing to tackle sparse rewards in the downstream tasks. To efficiently pre-train a large span of skills, we use Stochastic Neural Networks combined with an information-theoretic regularizer. Our experiments show that this combination is effective in learning a wide span of interpretable skills in a sample-efficient way, and can significantly boost the learning performance uniformly across a wide range of downstream tasks. | Another line of work on skill discovery particularly relevant to our approach is the HiREPS algorithm by @cite_2 where, instead of having a mutual information bonus like in our proposed approach, they introduce a constraint on an equivalent metric. The solution approach is nevertheless very different as they cannot use policy gradients. Furthermore, although they do achieve multimodality like us, they only tried the episodic case where a single option is active during the rollout. Hence the hierarchical use of the learned skills is less clear. Similarly, the Option-critic architecture can learn interpretable skills, but whether they can be reuse across complex tasks is still an open question. Recently, @cite_0 have independently proposed to learn a range of skills in a pre-training environment that will be useful for the downstream tasks, which is similar to our framework. However, their pre-training setup requires a set of goals to be specified. In comparison, we use proxy rewards as the only signal to the agent during the pre-training phase, the construction of which only requires minimal domain knowledge or instrumentation of the environment. | {
"cite_N": [
"@cite_0",
"@cite_2"
],
"mid": [
"2534060593",
"2111967991"
],
"abstract": [
"We study a novel architecture and training procedure for locomotion tasks. A high-frequency, low-level \"spinal\" network with access to proprioceptive sensors learns sensorimotor primitives by training on simple tasks. This pre-trained module is fixed and connected to a low-frequency, high-level \"cortical\" network, with access to all sensors, which drives behavior by modulating the inputs to the spinal network. Where a monolithic end-to-end architecture fails completely, learning with a pre-trained spinal module succeeds at multiple high-level tasks, and enables the effective exploration required to learn from sparse rewards. We test our proposed architecture on three simulated bodies: a 16-dimensional swimming snake, a 20-dimensional quadruped, and a 54-dimensional humanoid. Our results are illustrated in the accompanying video at this https URL",
"Many reinforcement learning (RL) tasks, especially in robotics, consist of multiple sub-tasks that are strongly structured. Such task structures can be exploited by incorporating hierarchical policies that consist of gating networks and sub-policies. However, this concept has only been partially explored for real world settings and complete methods, derived from first principles, are needed. Real world settings are challenging due to large and continuous state-action spaces that are prohibitive for exhaustive sampling methods. We define the problem of learning sub-policies in continuous state action spaces as finding a hierarchical policy that is composed of a high-level gating policy to select the low-level sub-policies for execution by the agent. In order to efficiently share experience with all sub-policies, also called inter-policy learning, we treat these sub-policies as latent variables which allows for distribution of the update information between the sub-policies. We present three different variants of our algorithm, designed to be suitable for a wide variety of real world robot learning tasks and evaluate our algorithms in two real robot learning scenarios as well as several simulations and comparisons."
]
} |
1704.03144 | 2613146443 | This work introduces the concept of parametric Gaussian processes (PGPs), which is built upon the seemingly self-contradictory idea of making Gaussian processes parametric. Parametric Gaussian processes, by construction, are designed to operate in "big data" regimes where one is interested in quantifying the uncertainty associated with noisy data. The proposed methodology circumvents the well-established need for stochastic variational inference, a scalable algorithm for approximating posterior distributions. The effectiveness of the proposed approach is demonstrated using an illustrative example with simulated data and a benchmark dataset in the airline industry with approximately 6 million records. | Despite some subtle differences, it is generally safe to recognize the input-output pairs @math (see equation )) as the so called inducing points'', a frequently used term in the literature on sparse approximations to Gaussian process priors (see e.g., @cite_3 for a compressive review). However, it is not advisable to interpret @math and @math (see equation )) as variational parameters @cite_5 since no (stochastic) variational inference is carried out in the current work. Furthermore, to highlight the subtle differences between inducing points'' and what this work calls ), it is worth observing that in the literature on sparse approximations to Gaussian processes it turns out that @math and @math . Under these assumptions and using equations ) and ), one obtains @math and @math . In other words, in the sparse Gaussian processes framework, @math and @math are essentially identical; i.e., @math . In contrast, this work treats @math and @math as parameters of the model responsible for encoding the history of observed data. In this regard, the current work is similar to @cite_5 . However, unlike @cite_5 , the parameter @math and @math are not variational parameters of some variational distribution. | {
"cite_N": [
"@cite_5",
"@cite_3"
],
"mid": [
"2952677397",
"1571870753"
],
"abstract": [
"We introduce stochastic variational inference for Gaussian process models. This enables the application of Gaussian process (GP) models to data sets containing millions of data points. We show how GPs can be vari- ationally decomposed to depend on a set of globally relevant inducing variables which factorize the model in the necessary manner to perform variational inference. Our ap- proach is readily extended to models with non-Gaussian likelihoods and latent variable models based around Gaussian processes. We demonstrate the approach on a simple toy problem and two real world data sets.",
"We provide a new unifying view, including all existing proper probabilistic sparse approximations for Gaussian process regression. Our approach relies on expressing the effective prior which the methods are using. This allows new insights to be gained, and highlights the relationship between existing methods. It also allows for a clear theoretically justified ranking of the closeness of the known approximations to the corresponding full GPs. Finally we point directly to designs of new better sparse approximations, combining the best of the existing strategies, within attractive computational constraints."
]
} |
1704.02998 | 2608258611 | We explore the power of spatial context as a self-supervisory signal for learning visual representations. In particular, we propose spatial context networks that learn to predict a representation of one image patch from another image patch, within the same image, conditioned on their real-valued relative spatial offset. Unlike auto-encoders, that aim to encode and reconstruct original image patches, our network aims to encode and reconstruct intermediate representations of the spatially offset patches. As such, the network learns a spatially conditioned contextual representation. By testing performance with various patch selection mechanisms we show that focusing on object-centric patches is important, and that using object proposal as a patch selection mechanism leads to the highest improvement in performance. Further, unlike auto-encoders, context encoders [21], or other forms of unsupervised feature learning, we illustrate that contextual supervision (with pre-trained model initialization) can improve on existing pre-trained model performance. We build our spatial context networks on top of standard VGG_19 and CNN_M architectures and, among other things, show that we can achieve improvements (with no additional explicit supervision) over the original ImageNet pre-trained VGG_19 and CNN_M models in object categorization and detection on VOC2007. | Unsupervised Learning . Auto-encoders @cite_35 are among the earliest models for unsupervised deep learning. They typically learn a representation by employing an encoder-decoder architecture, which are inverses of one another; the encoder encodes the image (or patch) into a compact hidden state representation and the decoder reconstructs it back to a full image. De-noising auto-encoders @cite_12 reconstruct images (or patches) subject to local corruptions. The most extreme variant of de-noising auto-encoders are the context encoders @cite_23 , which aim to reconstruct a large hole (patch) given its surrounding spatial context. | {
"cite_N": [
"@cite_35",
"@cite_23",
"@cite_12"
],
"mid": [
"",
"2963420272",
"2025768430"
],
"abstract": [
"",
"We present an unsupervised visual feature learning algorithm driven by context-based pixel prediction. By analogy with auto-encoders, we propose Context Encoders – a convolutional neural network trained to generate the contents of an arbitrary image region conditioned on its surroundings. In order to succeed at this task, context encoders need to both understand the content of the entire image, as well as produce a plausible hypothesis for the missing part(s). When training context encoders, we have experimented with both a standard pixel-wise reconstruction loss, as well as a reconstruction plus an adversarial loss. The latter produces much sharper results because it can better handle multiple modes in the output. We found that a context encoder learns a representation that captures not just appearance but also the semantics of visual structures. We quantitatively demonstrate the effectiveness of our learned features for CNN pre-training on classification, detection, and segmentation tasks. Furthermore, context encoders can be used for semantic inpainting tasks, either stand-alone or as initialization for non-parametric methods.",
"Previous work has shown that the difficulties in learning deep generative or discriminative models can be overcome by an initial unsupervised learning step that maps inputs to useful intermediate representations. We introduce and motivate a new training principle for unsupervised learning of a representation based on the idea of making the learned representations robust to partial corruption of the input pattern. This approach can be used to train autoencoders, and these denoising autoencoders can be stacked to initialize deep architectures. The algorithm can be motivated from a manifold learning and information theoretic perspective or from a generative model perspective. Comparative experiments clearly show the surprising advantage of corrupting the input of autoencoders on a pattern classification benchmark suite."
]
} |
1704.02998 | 2608258611 | We explore the power of spatial context as a self-supervisory signal for learning visual representations. In particular, we propose spatial context networks that learn to predict a representation of one image patch from another image patch, within the same image, conditioned on their real-valued relative spatial offset. Unlike auto-encoders, that aim to encode and reconstruct original image patches, our network aims to encode and reconstruct intermediate representations of the spatially offset patches. As such, the network learns a spatially conditioned contextual representation. By testing performance with various patch selection mechanisms we show that focusing on object-centric patches is important, and that using object proposal as a patch selection mechanism leads to the highest improvement in performance. Further, unlike auto-encoders, context encoders [21], or other forms of unsupervised feature learning, we illustrate that contextual supervision (with pre-trained model initialization) can improve on existing pre-trained model performance. We build our spatial context networks on top of standard VGG_19 and CNN_M architectures and, among other things, show that we can achieve improvements (with no additional explicit supervision) over the original ImageNet pre-trained VGG_19 and CNN_M models in object categorization and detection on VOC2007. | A number of papers proposed to learn representations by converting the generative auto-encoder-like objectives to discriminative classification counterparts, where CNNs have been shown to learn effectively. For example, @cite_22 proposed an idea of surrogate classes that are formed by applying a variety of transformations to randomly sampled image patches. Classification into these surrogate classes is used as a supervisory signal to learn image representations. Alternatively, in @cite_27 , neighboring patches are used in Siamese-like networks to predict the relative discrete ( , to the top-right, bottom-left, ) location of patches. Related, is also @cite_17 that attempts to learn a similarity function across patches using various deep learning architectures, including center-surround (similar to @cite_23 ) and forms of Siamese networks. Goodfellow al @cite_15 proposed Generative Adversarial Networks (GAN) that contain a generative model and discriminative model. Pathak al @cite_23 built upon GANs to model context through inpainting missing patches. | {
"cite_N": [
"@cite_22",
"@cite_27",
"@cite_23",
"@cite_15",
"@cite_17"
],
"mid": [
"2148349024",
"2950187998",
"2963420272",
"2099471712",
"2949213045"
],
"abstract": [
"Current methods for training convolutional neural networks depend on large amounts of labeled samples for supervised training. In this paper we present an approach for training a convolutional neural network using only unlabeled data. We train the network to discriminate between a set of surrogate classes. Each surrogate class is formed by applying a variety of transformations to a randomly sampled 'seed' image patch. We find that this simple feature learning algorithm is surprisingly successful when applied to visual object recognition. The feature representation learned by our algorithm achieves classification results matching or outperforming the current state-of-the-art for unsupervised learning on several popular datasets (STL-10, CIFAR-10, Caltech-101).",
"This work explores the use of spatial context as a source of free and plentiful supervisory signal for training a rich visual representation. Given only a large, unlabeled image collection, we extract random pairs of patches from each image and train a convolutional neural net to predict the position of the second patch relative to the first. We argue that doing well on this task requires the model to learn to recognize objects and their parts. We demonstrate that the feature representation learned using this within-image context indeed captures visual similarity across images. For example, this representation allows us to perform unsupervised visual discovery of objects like cats, people, and even birds from the Pascal VOC 2011 detection dataset. Furthermore, we show that the learned ConvNet can be used in the R-CNN framework and provides a significant boost over a randomly-initialized ConvNet, resulting in state-of-the-art performance among algorithms which use only Pascal-provided training set annotations.",
"We present an unsupervised visual feature learning algorithm driven by context-based pixel prediction. By analogy with auto-encoders, we propose Context Encoders – a convolutional neural network trained to generate the contents of an arbitrary image region conditioned on its surroundings. In order to succeed at this task, context encoders need to both understand the content of the entire image, as well as produce a plausible hypothesis for the missing part(s). When training context encoders, we have experimented with both a standard pixel-wise reconstruction loss, as well as a reconstruction plus an adversarial loss. The latter produces much sharper results because it can better handle multiple modes in the output. We found that a context encoder learns a representation that captures not just appearance but also the semantics of visual structures. We quantitatively demonstrate the effectiveness of our learned features for CNN pre-training on classification, detection, and segmentation tasks. Furthermore, context encoders can be used for semantic inpainting tasks, either stand-alone or as initialization for non-parametric methods.",
"We propose a new framework for estimating generative models via an adversarial process, in which we simultaneously train two models: a generative model G that captures the data distribution, and a discriminative model D that estimates the probability that a sample came from the training data rather than G. The training procedure for G is to maximize the probability of D making a mistake. This framework corresponds to a minimax two-player game. In the space of arbitrary functions G and D, a unique solution exists, with G recovering the training data distribution and D equal to ½ everywhere. In the case where G and D are defined by multilayer perceptrons, the entire system can be trained with backpropagation. There is no need for any Markov chains or unrolled approximate inference networks during either training or generation of samples. Experiments demonstrate the potential of the framework through qualitative and quantitative evaluation of the generated samples.",
"In this paper we show how to learn directly from image data (i.e., without resorting to manually-designed features) a general similarity function for comparing image patches, which is a task of fundamental importance for many computer vision problems. To encode such a function, we opt for a CNN-based model that is trained to account for a wide variety of changes in image appearance. To that end, we explore and study multiple neural network architectures, which are specifically adapted to this task. We show that such an approach can significantly outperform the state-of-the-art on several problems and benchmark datasets."
]
} |
1704.02998 | 2608258611 | We explore the power of spatial context as a self-supervisory signal for learning visual representations. In particular, we propose spatial context networks that learn to predict a representation of one image patch from another image patch, within the same image, conditioned on their real-valued relative spatial offset. Unlike auto-encoders, that aim to encode and reconstruct original image patches, our network aims to encode and reconstruct intermediate representations of the spatially offset patches. As such, the network learns a spatially conditioned contextual representation. By testing performance with various patch selection mechanisms we show that focusing on object-centric patches is important, and that using object proposal as a patch selection mechanism leads to the highest improvement in performance. Further, unlike auto-encoders, context encoders [21], or other forms of unsupervised feature learning, we illustrate that contextual supervision (with pre-trained model initialization) can improve on existing pre-trained model performance. We build our spatial context networks on top of standard VGG_19 and CNN_M architectures and, among other things, show that we can achieve improvements (with no additional explicit supervision) over the original ImageNet pre-trained VGG_19 and CNN_M models in object categorization and detection on VOC2007. | Pre-trained Models . Pre-trained CNN models have been shown to generalize to a large number of different tasks @cite_36 @cite_16 . However, their transferability, as was noted in @cite_37 , is affected by specialization of higher layer neurons to the original task (often ImageNet categorization). By taking a network pre-trained on the ImageNet task and using its intermediate representation as target for our decoder, we make use of the knowledge distilled in the network @cite_3 while attempting to improve it using spatial context. Works like @cite_28 and @cite_19 attempt to similarly re-use lower layers @cite_28 of the pre-trained network and fine-tune, typically, fully-connected layers to specific tasks ( , object detection). However, such models assume some labeled data in the target domain, if not for classes of interest @cite_28 , then for related ones @cite_19 . In our case, we assume no supervision of this form. Instead, we just assume that there exists a process that can generate category agnostic object-like proposal patches. Our work is similar to @cite_32 that also attempts to improve the performance of pre-trained models. While they augment existing networks with reconstructive decoding pathways for image reconstruction, our model focuses on exploiting contextual relationships in images. | {
"cite_N": [
"@cite_37",
"@cite_36",
"@cite_28",
"@cite_32",
"@cite_3",
"@cite_19",
"@cite_16"
],
"mid": [
"2949667497",
"2953360861",
"2161381512",
"",
"1821462560",
"2951084305",
"2953391683"
],
"abstract": [
"Many deep neural networks trained on natural images exhibit a curious phenomenon in common: on the first layer they learn features similar to Gabor filters and color blobs. Such first-layer features appear not to be specific to a particular dataset or task, but general in that they are applicable to many datasets and tasks. Features must eventually transition from general to specific by the last layer of the network, but this transition has not been studied extensively. In this paper we experimentally quantify the generality versus specificity of neurons in each layer of a deep convolutional neural network and report a few surprising results. Transferability is negatively affected by two distinct issues: (1) the specialization of higher layer neurons to their original task at the expense of performance on the target task, which was expected, and (2) optimization difficulties related to splitting networks between co-adapted neurons, which was not expected. In an example network trained on ImageNet, we demonstrate that either of these two issues may dominate, depending on whether features are transferred from the bottom, middle, or top of the network. We also document that the transferability of features decreases as the distance between the base task and target task increases, but that transferring features even from distant tasks can be better than using random features. A final surprising result is that initializing a network with transferred features from almost any number of layers can produce a boost to generalization that lingers even after fine-tuning to the target dataset.",
"We evaluate whether features extracted from the activation of a deep convolutional network trained in a fully supervised fashion on a large, fixed set of object recognition tasks can be re-purposed to novel generic tasks. Our generic tasks may differ significantly from the originally trained tasks and there may be insufficient labeled or unlabeled data to conventionally train or adapt a deep architecture to the new tasks. We investigate and visualize the semantic clustering of deep convolutional features with respect to a variety of such tasks, including scene recognition, domain adaptation, and fine-grained recognition challenges. We compare the efficacy of relying on various network levels to define a fixed feature, and report novel results that significantly outperform the state-of-the-art on several important vision challenges. We are releasing DeCAF, an open-source implementation of these deep convolutional activation features, along with all associated network parameters to enable vision researchers to be able to conduct experimentation with deep representations across a range of visual concept learning paradigms.",
"Convolutional neural networks (CNN) have recently shown outstanding image classification performance in the large- scale visual recognition challenge (ILSVRC2012). The suc- cess of CNNs is attributed to their ability to learn rich mid- level image representations as opposed to hand-designed low-level features used in other image classification meth- ods. Learning CNNs, however, amounts to estimating mil- lions of parameters and requires a very large number of annotated image samples. This property currently prevents application of CNNs to problems with limited training data. In this work we show how image representations learned with CNNs on large-scale annotated datasets can be effi- ciently transferred to other visual recognition tasks with limited amount of training data. We design a method to reuse layers trained on the ImageNet dataset to compute mid-level image representation for images in the PASCAL VOC dataset. We show that despite differences in image statistics and tasks in the two datasets, the transferred rep- resentation leads to significantly improved results for object and action classification, outperforming the current state of the art on Pascal VOC 2007 and 2012 datasets. We also show promising results for object and action localization.",
"",
"A very simple way to improve the performance of almost any machine learning algorithm is to train many different models on the same data and then to average their predictions. Unfortunately, making predictions using a whole ensemble of models is cumbersome and may be too computationally expensive to allow deployment to a large number of users, especially if the individual models are large neural nets. Caruana and his collaborators have shown that it is possible to compress the knowledge in an ensemble into a single model which is much easier to deploy and we develop this approach further using a different compression technique. We achieve some surprising results on MNIST and we show that we can significantly improve the acoustic model of a heavily used commercial system by distilling the knowledge in an ensemble of models into a single model. We also introduce a new type of ensemble composed of one or more full models and many specialist models which learn to distinguish fine-grained classes that the full models confuse. Unlike a mixture of experts, these specialist models can be trained rapidly and in parallel.",
"A major challenge in scaling object detection is the difficulty of obtaining labeled images for large numbers of categories. Recently, deep convolutional neural networks (CNNs) have emerged as clear winners on object classification benchmarks, in part due to training with 1.2M+ labeled classification images. Unfortunately, only a small fraction of those labels are available for the detection task. It is much cheaper and easier to collect large quantities of image-level labels from search engines than it is to collect detection data and label it with precise bounding boxes. In this paper, we propose Large Scale Detection through Adaptation (LSDA), an algorithm which learns the difference between the two tasks and transfers this knowledge to classifiers for categories without bounding box annotated data, turning them into detectors. Our method has the potential to enable detection for the tens of thousands of categories that lack bounding box annotations, yet have plenty of classification data. Evaluation on the ImageNet LSVRC-2013 detection challenge demonstrates the efficacy of our approach. This algorithm enables us to produce a >7.6K detector by using available classification data from leaf nodes in the ImageNet tree. We additionally demonstrate how to modify our architecture to produce a fast detector (running at 2fps for the 7.6K detector). Models and software are available at",
"Recent results indicate that the generic descriptors extracted from the convolutional neural networks are very powerful. This paper adds to the mounting evidence that this is indeed the case. We report on a series of experiments conducted for different recognition tasks using the publicly available code and model of the network which was trained to perform object classification on ILSVRC13. We use features extracted from the network as a generic image representation to tackle the diverse range of recognition tasks of object image classification, scene recognition, fine grained recognition, attribute detection and image retrieval applied to a diverse set of datasets. We selected these tasks and datasets as they gradually move further away from the original task and data the network was trained to solve. Astonishingly, we report consistent superior results compared to the highly tuned state-of-the-art systems in all the visual classification tasks on various datasets. For instance retrieval it consistently outperforms low memory footprint methods except for sculptures dataset. The results are achieved using a linear SVM classifier (or @math distance in case of retrieval) applied to a feature representation of size 4096 extracted from a layer in the net. The representations are further modified using simple augmentation techniques e.g. jittering. The results strongly suggest that features obtained from deep learning with convolutional nets should be the primary candidate in most visual recognition tasks."
]
} |
1704.02998 | 2608258611 | We explore the power of spatial context as a self-supervisory signal for learning visual representations. In particular, we propose spatial context networks that learn to predict a representation of one image patch from another image patch, within the same image, conditioned on their real-valued relative spatial offset. Unlike auto-encoders, that aim to encode and reconstruct original image patches, our network aims to encode and reconstruct intermediate representations of the spatially offset patches. As such, the network learns a spatially conditioned contextual representation. By testing performance with various patch selection mechanisms we show that focusing on object-centric patches is important, and that using object proposal as a patch selection mechanism leads to the highest improvement in performance. Further, unlike auto-encoders, context encoders [21], or other forms of unsupervised feature learning, we illustrate that contextual supervision (with pre-trained model initialization) can improve on existing pre-trained model performance. We build our spatial context networks on top of standard VGG_19 and CNN_M architectures and, among other things, show that we can achieve improvements (with no additional explicit supervision) over the original ImageNet pre-trained VGG_19 and CNN_M models in object categorization and detection on VOC2007. | Weakly-supervised and Self-supervised Learning . Weakly-supervised and self-supervised learning attempt to achieve similar performance to fully supervised models with limited use of annotated labels. A typical setting is to, for example, use image-level annotations to learn an object detection model @cite_38 @cite_18 @cite_0 @cite_6 @cite_25 @cite_33 . However, such models typically rely on latent variables and appearance regularities present within individual object class. In addition, researchers also utilized motion coherence (tracked patches @cite_21 or ego-motion from sensors @cite_13 ) in videos as supervisory signals to train networks. Zhang al @cite_14 generated a color version of a grayscale photo through a CNN model, which could further serve as an auxiliary task for feature learning. Noroozi al learned features by solving jigsaw puzzles @cite_9 . Different from these works, we experiment with (category-independent) object proposals as a way to tokenize an image into more semantically meaningful parts. This can be thought of as (perhaps) a very weak form of supervision, but unlike any that we are aware has been used before. | {
"cite_N": [
"@cite_38",
"@cite_18",
"@cite_14",
"@cite_33",
"@cite_9",
"@cite_21",
"@cite_6",
"@cite_0",
"@cite_13",
"@cite_25"
],
"mid": [
"2016016818",
"1994488211",
"2326925005",
"318792885",
"2321533354",
"219040644",
"2952072685",
"2079267072",
"2951590555",
""
],
"abstract": [
"Object category localization is a challenging problem in computer vision. Standard supervised training requires bounding box annotations of object instances. This time-consuming annotation process is sidestepped in weakly supervised learning. In this case, the supervised information is restricted to binary labels that indicate the absence presence of object instances in the image, without their locations. We follow a multiple-instance learning approach that iteratively trains the detector and infers the object locations in the positive training images. Our main contribution is a multi-fold multiple instance learning procedure, which prevents training from prematurely locking onto erroneous object locations. This procedure is particularly important when high-dimensional representations, such as the Fisher vectors, are used. We present a detailed experimental evaluation using the PASCAL VOC 2007 dataset. Compared to state-of-the-art weakly supervised detectors, our approach better localizes objects in the training images, which translates into improved detection performance.",
"Successful methods for visual object recognition typically rely on training datasets containing lots of richly annotated images. Detailed image annotation, e.g. by object bounding boxes, however, is both expensive and often subjective. We describe a weakly supervised convolutional neural network (CNN) for object classification that relies only on image-level labels, yet can learn from cluttered scenes containing multiple objects. We quantify its object classification and object location prediction performance on the Pascal VOC 2012 (20 object classes) and the much larger Microsoft COCO (80 object classes) datasets. We find that the network (i) outputs accurate image-level labels, (ii) predicts approximate locations (but not extents) of objects, and (iii) performs comparably to its fully-supervised counterparts using object bounding box annotation for training.",
"Given a grayscale photograph as input, this paper attacks the problem of hallucinating a plausible color version of the photograph. This problem is clearly underconstrained, so previous approaches have either relied on significant user interaction or resulted in desaturated colorizations. We propose a fully automatic approach that produces vibrant and realistic colorizations. We embrace the underlying uncertainty of the problem by posing it as a classification task and use class-rebalancing at training time to increase the diversity of colors in the result. The system is implemented as a feed-forward pass in a CNN at test time and is trained on over a million color images. We evaluate our algorithm using a “colorization Turing test,” asking human participants to choose between a generated and ground truth color image. Our method successfully fools humans on 32 of the trials, significantly higher than previous methods. Moreover, we show that colorization can be a powerful pretext task for self-supervised feature learning, acting as a cross-channel encoder. This approach results in state-of-the-art performance on several feature learning benchmarks.",
"Localizing objects in cluttered backgrounds is a challenging task in weakly supervised localization. Due to large object variations in cluttered images, objects have large ambiguity with backgrounds. However, backgrounds contain useful latent information, e.g., the sky for aeroplanes. If we can learn this latent information, object-background ambiguity can be reduced to suppress the background. In this paper, we propose the latent category learning (LCL), which is an unsupervised learning problem given only image-level class labels. Firstly, inspired by the latent semantic discovery, we use the typical probabilistic Latent Semantic Analysis (pLSA) to learn the latent categories, which can represent objects, object parts or backgrounds. Secondly, to determine which category contains the target object, we propose a category selection method evaluating each category’s discrimination. We evaluate the method on the PASCAL VOC 2007 database and ILSVRC 2013 detection challenge. On VOC 2007, the proposed method yields the annotation accuracy of 48 , which outperforms previous results by 10 . More importantly, we achieve the detection average precision of 30.9 , which improves previous results by 8 and can be competitive with the supervised deformable part model (DPM) 5.0 baseline 33.7 . On ILSVRC 2013 detection, the method yields the precision of 6.0 , which is also competitive with the DPM 5.0.",
"We propose a novel unsupervised learning approach to build features suitable for object detection and classification. The features are pre-trained on a large dataset without human annotation and later transferred via fine-tuning on a different, smaller and labeled dataset. The pre-training consists of solving jigsaw puzzles of natural images. To facilitate the transfer of features to other tasks, we introduce the context-free network (CFN), a siamese-ennead convolutional neural network. The features correspond to the columns of the CFN and they process image tiles independently (i.e., free of context). The later layers of the CFN then use the features to identify their geometric arrangement. Our experimental evaluations show that the learned features capture semantically relevant content. We pre-train the CFN on the training set of the ILSVRC2012 dataset and transfer the features on the combined training and validation set of Pascal VOC 2007 for object detection (via fast RCNN) and classification. These features outperform all current unsupervised features with (51.8 , ) for detection and (68.6 , ) for classification, and reduce the gap with supervised learning ( (56.5 , ) and (78.2 , ) respectively).",
"Is strong supervision necessary for learning a good visual representation? Do we really need millions of semantically-labeled images to train a Convolutional Neural Network (CNN)? In this paper, we present a simple yet surprisingly powerful approach for unsupervised learning of CNN. Specifically, we use hundreds of thousands of unlabeled videos from the web to learn visual representations. Our key idea is that visual tracking provides the supervision. That is, two patches connected by a track should have similar visual representation in deep feature space since they probably belong to same object or object part. We design a Siamese-triplet network with a ranking loss function to train this CNN representation. Without using a single image from ImageNet, just using 100K unlabeled videos and the VOC 2012 dataset, we train an ensemble of unsupervised networks that achieves 52 mAP (no bounding box regression). This performance comes tantalizingly close to its ImageNet-supervised counterpart, an ensemble which achieves a mAP of 54.4 . We also show that our unsupervised network can perform competitively in other tasks such as surface-normal estimation.",
"Learning to localize objects with minimal supervision is an important problem in computer vision, since large fully annotated datasets are extremely costly to obtain. In this paper, we propose a new method that achieves this goal with only image-level labels of whether the objects are present or not. Our approach combines a discriminative submodular cover problem for automatically discovering a set of positive object windows with a smoothed latent SVM formulation. The latter allows us to leverage efficient quasi-Newton optimization techniques. Our experiments demonstrate that the proposed approach provides a 50 relative improvement in mean average precision over the current state-of-the-art on PASCAL VOC 2007 detection.",
"We address the problem of localisation of objects as bounding boxes in images with weak labels. This weakly supervised object localisation problem has been tackled in the past using discriminative models where each object class is localised independently from other classes. We propose a novel framework based on Bayesian joint topic modelling. Our framework has three distinctive advantages over previous works: (1) All object classes and image backgrounds are modelled jointly together in a single generative model so that \"explaining away\" inference can resolve ambiguity and lead to better learning and localisation. (2) The Bayesian formulation of the model enables easy integration of prior knowledge about object appearance to compensate for limited supervision. (3) Our model can be learned with a mixture of weakly labelled and unlabelled data, allowing the large volume of unlabelled images on the Internet to be exploited for learning. Extensive experiments on the challenging VOC dataset demonstrate that our approach outperforms the state-of-the-art competitors.",
"The dominant paradigm for feature learning in computer vision relies on training neural networks for the task of object recognition using millions of hand labelled images. Is it possible to learn useful features for a diverse set of visual tasks using any other form of supervision? In biology, living organisms developed the ability of visual perception for the purpose of moving and acting in the world. Drawing inspiration from this observation, in this work we investigate if the awareness of egomotion can be used as a supervisory signal for feature learning. As opposed to the knowledge of class labels, information about egomotion is freely available to mobile agents. We show that given the same number of training images, features learnt using egomotion as supervision compare favourably to features learnt using class-label as supervision on visual tasks of scene recognition, object recognition, visual odometry and keypoint matching.",
""
]
} |
1704.03173 | 2951110097 | Given a convolutional neural network (CNN) that is pre-trained for object classification, this paper proposes to use active question-answering to semanticize neural patterns in conv-layers of the CNN and mine part concepts. For each part concept, we mine neural patterns in the pre-trained CNN, which are related to the target part, and use these patterns to construct an And-Or graph (AOG) to represent a four-layer semantic hierarchy of the part. As an interpretable model, the AOG associates different CNN units with different explicit object parts. We use an active human-computer communication to incrementally grow such an AOG on the pre-trained CNN as follows. We allow the computer to actively identify objects, whose neural patterns cannot be explained by the current AOG. Then, the computer asks human about the unexplained objects, and uses the answers to automatically discover certain CNN patterns corresponding to the missing knowledge. We incrementally grow the AOG to encode new knowledge discovered during the active-learning process. In experiments, our method exhibits high learning efficiency. Our method uses about 1 6-1 3 of the part annotations for training, but achieves similar or better part-localization performance than fast-RCNN methods. | | | In order to explore the hidden semantics in the CNN, many studies visualized and analyzed patterns of CNN units @cite_15 @cite_30 @cite_26 @cite_46 @cite_44 . | {
"cite_N": [
"@cite_30",
"@cite_26",
"@cite_46",
"@cite_44",
"@cite_15"
],
"mid": [
"2949987032",
"2962851944",
"1661149683",
"",
"2952186574"
],
"abstract": [
"Image representations, from SIFT and Bag of Visual Words to Convolutional Neural Networks (CNNs), are a crucial component of almost any image understanding system. Nevertheless, our understanding of them remains limited. In this paper we conduct a direct analysis of the visual information contained in representations by asking the following question: given an encoding of an image, to which extent is it possible to reconstruct the image itself? To answer this question we contribute a general framework to invert representations. We show that this method can invert representations such as HOG and SIFT more accurately than recent alternatives while being applicable to CNNs too. We then use this technique to study the inverse of recent state-of-the-art CNN image representations for the first time. Among our findings, we show that several layers in CNNs retain photographically accurate information about the image, with different degrees of geometric and photometric invariance.",
"This paper addresses the visualisation of image classification models, learnt using deep Convolutional Networks (ConvNets). We consider two visualisation techniques, based on computing the gradient of the class score with respect to the input image. The first one generates an image, which maximises the class score [5], thus visualising the notion of the class, captured by a ConvNet. The second technique computes a class saliency map, specific to a given image and class. We show that such maps can be employed for weakly supervised object segmentation using classification ConvNets. Finally, we establish the connection between the gradient-based ConvNet visualisation methods and deconvolutional networks [13].",
"We introduce an approach for analyzing the variation of features generated by convolutional neural networks (CNNs) trained on large image datasets with respect to scene factors that occur in natural images. Such factors may include object style, 3D viewpoint, color, and scene lighting configuration. Our approach analyzes CNN feature responses with respect to different scene factors by controlling for them via rendering using a large database of 3D CAD models. The rendered images are presented to a trained CNN and responses for different layers are studied with respect to the input scene factors. We perform a linear decomposition of the responses based on knowledge of the input scene factors and analyze the resulting components. In particular, we quantify their relative importance in the CNN responses and visualize them using principal component analysis. We show qualitative and quantitative results of our study on three trained CNNs: AlexNet [18], Places [43], and Oxford VGG [8]. We observe important differences across the different networks and CNN layers with respect to different scene factors and object categories. Finally, we demonstrate that our analysis based on computer-generated imagery translates to the network representation of natural images.",
"",
"Large Convolutional Network models have recently demonstrated impressive classification performance on the ImageNet benchmark. However there is no clear understanding of why they perform so well, or how they might be improved. In this paper we address both issues. We introduce a novel visualization technique that gives insight into the function of intermediate feature layers and the operation of the classifier. We also perform an ablation study to discover the performance contribution from different model layers. This enables us to find model architectures that outperform Krizhevsky al on the ImageNet classification benchmark. We show our ImageNet model generalizes well to other datasets: when the softmax classifier is retrained, it convincingly beats the current state-of-the-art results on Caltech-101 and Caltech-256 datasets."
]
} |
1704.03173 | 2951110097 | Given a convolutional neural network (CNN) that is pre-trained for object classification, this paper proposes to use active question-answering to semanticize neural patterns in conv-layers of the CNN and mine part concepts. For each part concept, we mine neural patterns in the pre-trained CNN, which are related to the target part, and use these patterns to construct an And-Or graph (AOG) to represent a four-layer semantic hierarchy of the part. As an interpretable model, the AOG associates different CNN units with different explicit object parts. We use an active human-computer communication to incrementally grow such an AOG on the pre-trained CNN as follows. We allow the computer to actively identify objects, whose neural patterns cannot be explained by the current AOG. Then, the computer asks human about the unexplained objects, and uses the answers to automatically discover certain CNN patterns corresponding to the missing knowledge. We incrementally grow the AOG to encode new knowledge discovered during the active-learning process. In experiments, our method exhibits high learning efficiency. Our method uses about 1 6-1 3 of the part annotations for training, but achieves similar or better part-localization performance than fast-RCNN methods. | | | Many methods have been developed to learn object models in an unsupervised or weakly supervised manner. Methods of @cite_35 @cite_19 @cite_2 @cite_40 learned with image-level annotations without labeling object bounding boxes. @cite_0 @cite_27 did not require any annotations during the learning process. @cite_25 collected training data online from videos to incrementally learn models. @cite_3 @cite_22 discovered objects and identified actions from language Instructions and videos. Inspired by active learning @cite_43 @cite_37 @cite_45 , the idea of learning from question-answering has been used to learn object models @cite_20 @cite_36 @cite_12 . Branson @cite_17 used human-computer interactions to label object parts to learn part models. Instead of directly building new models from active QA, our method uses the QA to semanticize the CNN and transfer the hidden knowledge to the AOG. | {
"cite_N": [
"@cite_35",
"@cite_37",
"@cite_22",
"@cite_36",
"@cite_3",
"@cite_0",
"@cite_19",
"@cite_40",
"@cite_27",
"@cite_43",
"@cite_45",
"@cite_2",
"@cite_12",
"@cite_25",
"@cite_20",
"@cite_17"
],
"mid": [
"2950931866",
"2027953712",
"2579317665",
"1908985308",
"2099501835",
"2148349024",
"2952072685",
"2949820118",
"2950926078",
"1956202674",
"2201054263",
"2210113682",
"2951614005",
"2087044413",
"",
"2122686738"
],
"abstract": [
"We present an approach to utilize large amounts of web data for learning CNNs. Specifically inspired by curriculum learning, we present a two-step approach for CNN training. First, we use easy images to train an initial visual representation. We then use this initial CNN and adapt it to harder, more realistic images by leveraging the structure of data and categories. We demonstrate that our two-stage CNN outperforms a fine-tuned CNN trained on ImageNet on Pascal VOC 2012. We also demonstrate the strength of webly supervised learning by localizing objects in web images and training a R-CNN style detector. It achieves the best performance on VOC 2007 where no VOC training data is used. Finally, we show our approach is quite robust to noise and performs comparably even when we use image search results from March 2013 (pre-CNN image search era).",
"Active learning and crowdsourcing are promising ways to efficiently build up training sets for object recognition, but thus far techniques are tested in artificially controlled settings. Typically the vision researcher has already determined the dataset's scope, the labels “actively” obtained are in fact already known, and or the crowd-sourced collection process is iteratively fine-tuned. We present an approach for live learning of object detectors, in which the system autonomously refines its models by actively requesting crowd-sourced annotations on images crawled from the Web. To address the technical issues such a large-scale system entails, we introduce a novel part-based detector amenable to linear classifiers, and show how to identify its most uncertain instances in sub-linear time with a hashing-based solution. We demonstrate the approach with experiments of unprecedented scale and autonomy, and show it successfully improves the state-of-the-art for the most challenging objects in the PASCAL benchmark. In addition, we show our detector competes well with popular nonlinear classifiers that are much more expensive to train.",
"Advances in video technology and data storage have made large scale video data collections of complex activities readily accessible. An increasingly popular approach for automatically inferring the details of a video is to associate the spatio-temporal segments in a video with its natural language descriptions. Most algorithms for connecting natural language with video rely on pre-aligned supervised training data. Recently, several models have been shown to be effective for unsupervised alignment of objects in video with language. However, it remains difficult to generate good spatio-temporal video segments for actions that align well with language. This paper presents a framework that extracts higher level representations of low-level action features through hyperfeature coding from video and aligns them with language. We propose a two-step process that creates a high-level action feature codebook with temporally consistent motions, and then applies an unsupervised alignment algorithm over the action codewords and verbs in the language to identify individual activities. We show an improvement over previous alignment models of objects and nouns on videos of biological experiments, and also evaluate our system on a larger scale collection of videos involving kitchen activities.",
"The long-standing goal of localizing every object in an image remains elusive. Manually annotating objects is quite expensive despite crowd engineering innovations. Current state-of-the-art automatic object detectors can accurately detect at most a few objects per image. This paper brings together the latest advancements in object detection and in crowd engineering into a principled framework for accurately and efficiently localizing objects in images. The input to the system is an image to annotate and a set of annotation constraints: desired precision, utility and or human cost of the labeling. The output is a set of object annotations, informed by human feedback and computer vision. Our model seamlessly integrates multiple computer vision models with multiple sources of human input in a Markov Decision Process. We empirically validate the effectiveness of our human-in-the-loop labeling approach on the ILSVRC2014 object detection dataset.",
"We propose a visual event recognition framework for consumer videos by leveraging a large amount of loosely labeled web videos (e.g., from YouTube). Observing that consumer videos generally contain large intraclass variations within the same type of events, we first propose a new method, called Aligned Space-Time Pyramid Matching (ASTPM), to measure the distance between any two video clips. Second, we propose a new transfer learning method, referred to as Adaptive Multiple Kernel Learning (A-MKL), in order to 1) fuse the information from multiple pyramid levels and features (i.e., space-time features and static SIFT features) and 2) cope with the considerable variation in feature distributions between videos from two domains (i.e., web video domain and consumer video domain). For each pyramid level and each type of local features, we first train a set of SVM classifiers based on the combined training set from two domains by using multiple base kernels from different kernel types and parameters, which are then fused with equal weights to obtain a prelearned average classifier. In A-MKL, for each event class we learn an adapted target classifier based on multiple base kernels and the prelearned average classifiers from this event class or all the event classes by minimizing both the structural risk functional and the mismatch between data distributions of two domains. Extensive experiments demonstrate the effectiveness of our proposed framework that requires only a small number of labeled consumer videos by leveraging web data. We also conduct an in-depth investigation on various aspects of the proposed method A-MKL, such as the analysis on the combination coefficients on the prelearned classifiers, the convergence of the learning algorithm, and the performance variation by using different proportions of labeled consumer videos. Moreover, we show that A-MKL using the prelearned classifiers from all the event classes leads to better performance when compared with A-MKL using the prelearned classifiers only from each individual event class.",
"Current methods for training convolutional neural networks depend on large amounts of labeled samples for supervised training. In this paper we present an approach for training a convolutional neural network using only unlabeled data. We train the network to discriminate between a set of surrogate classes. Each surrogate class is formed by applying a variety of transformations to a randomly sampled 'seed' image patch. We find that this simple feature learning algorithm is surprisingly successful when applied to visual object recognition. The feature representation learned by our algorithm achieves classification results matching or outperforming the current state-of-the-art for unsupervised learning on several popular datasets (STL-10, CIFAR-10, Caltech-101).",
"Learning to localize objects with minimal supervision is an important problem in computer vision, since large fully annotated datasets are extremely costly to obtain. In this paper, we propose a new method that achieves this goal with only image-level labels of whether the objects are present or not. Our approach combines a discriminative submodular cover problem for automatically discovering a set of positive object windows with a smoothed latent SVM formulation. The latter allows us to leverage efficient quasi-Newton optimization techniques. Our experiments demonstrate that the proposed approach provides a 50 relative improvement in mean average precision over the current state-of-the-art on PASCAL VOC 2007 detection.",
"Part models of object categories are essential for challenging recognition tasks, where differences in categories are subtle and only reflected in appearances of small parts of the object. We present an approach that is able to learn part models in a completely unsupervised manner, without part annotations and even without given bounding boxes during learning. The key idea is to find constellations of neural activation patterns computed using convolutional neural networks. In our experiments, we outperform existing approaches for fine-grained recognition on the CUB200-2011, NA birds, Oxford PETS, and Oxford Flowers dataset in case no part or bounding box annotations are available and achieve state-of-the-art performance for the Stanford Dog dataset. We also show the benefits of neural constellation models as a data augmentation technique for fine-tuning. Furthermore, our paper unites the areas of generic and fine-grained classification, since our approach is suitable for both scenarios. The source code of our method is available online at this http URL",
"This paper addresses unsupervised discovery and localization of dominant objects from a noisy image collection with multiple object classes. The setting of this problem is fully unsupervised, without even image-level annotations or any assumption of a single dominant class. This is far more general than typical colocalization, cosegmentation, or weakly-supervised localization tasks. We tackle the discovery and localization problem using a part-based region matching approach: We use off-the-shelf region proposals to form a set of candidate bounding boxes for objects and object parts. These regions are efficiently matched across images using a probabilistic Hough transform that evaluates the confidence for each candidate correspondence considering both appearance and spatial consistency. Dominant objects are discovered and localized by comparing the scores of candidate regions and selecting those that stand out over other regions containing them. Extensive experimental evaluations on standard benchmarks demonstrate that the proposed approach significantly outperforms the current state of the art in colocalization, and achieves robust object discovery in challenging mixed-class datasets.",
"This paper studies active learning in structured probabilistic models such as Conditional Random Fields (CRFs). This is a challenging problem because unlike unstructured prediction problems such as binary or multi-class classification, structured prediction problems involve a distribution with an exponentially-large support, for instance, over the space of all possible segmentations of an image. Thus, the entropy of such models is typically intractable to compute. We propose a crude yet surprisingly effective histogram approximation to the Gibbs distribution, which replaces the exponentially-large support with a coarsened distribution that may be viewed as a histogram over M bins. We show that our approach outperforms a number of baselines and results in a 90 -reduction in the number of annotations needed to achieve nearly the same accuracy as learning from the entire dataset.",
"Active learning is an effective way to relieve the tedious work of manual annotation in many applications of visual recognition. However, less research attention has been focused on multi-class active learning. In this paper, we propose a novel Gaussian process classifier model with multiple annotators for multi-class visual recognition. Expectation propagation (EP) is adopted for efficient approximate Bayesian inference of our probabilistic model for classification. Based on the EP approximation inference, a generalized Expectation Maximization (GEM) algorithm is derived to estimate both the parameters for instances and the quality of each individual annotator. Also, we incorporate the idea of reinforcement learning to actively select both the informative samples and the high-quality annotators, which better explores the trade-off between exploitation and exploration. The experiments clearly demonstrate the efficacy of the proposed model.",
"This paper reformulates the theory of graph mining on the technical basis of graph matching, and extends its scope of applications to computer vision. Given a set of attributed relational graphs (ARGs), we propose to use a hierarchical And-Or Graph (AoG) to model the pattern of maximal-size common subgraphs embedded in the ARGs, and we develop a general method to mine the AoG model from the unlabeled ARGs. This method provides a general solution to the problem of mining hierarchical models from unannotated visual data without exhaustive search of objects. We apply our method to RGB RGB-D images and videos to demonstrate its generality and the wide range of applicability. The code will be available at https: sites.google.com site quanshizhang mining-and-or-graphs.",
"We propose a framework for parsing video and text jointly for understanding events and answering user queries. Our 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 our 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. We 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, we 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: Firstly, we aim at deep semantic parsing of video and text that goes beyond the traditional bag-of-words approaches; Secondly, we perform parsing and reasoning across the spatial, temporal and causal dimensions based on the joint S T C-AOG representation; Thirdly, we 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. We empirically evaluated our system based on comparison against ground-truth as well as accuracy of query answering and obtained satisfactory results.",
"Conventional visual recognition systems usually train an image classifier in a bath mode with all training data provided in advance. However, in many practical applications, only a small amount of training samples are available in the beginning and many more would come sequentially during online recognition. Because the image data characteristics could change over time, it is important for the classifier to adapt to the new data incrementally. In this paper, we present an online metric learning method to address the online scene recognition problem via adaptive similarity measurement. Given a number of labeled data followed by a sequential input of unseen testing samples, the similarity metric is learned to maximize the margin of the distance among different classes of samples. By considering the low rank constraint, our online metric learning model not only can provide competitive performance compared with the state-of-the-art methods, but also guarantees convergence. A bi-linear graph is also defined to model the pair-wise similarity, and an unseen sample is labeled depending on the graph-based label propagation, while the model can also self-update using the more confident new samples. With the ability of online learning, our methodology can well handle the large-scale streaming video data with the ability of incremental self-updating. We evaluate our model to online scene categorization and experiments on various benchmark datasets and comparisons with state-of-the-art methods demonstrate the effectiveness and efficiency of our algorithm.",
"",
"We propose a framework for large scale learning and annotation of structured models. The system interleaves interactive labeling (where the current model is used to semi-automate the labeling of a new example) and online learning (where a newly labeled example is used to update the current model parameters). This framework is scalable to large datasets and complex image models and is shown to have excellent theoretical and practical properties in terms of train time, optimality guarantees, and bounds on the amount of annotation effort per image. We apply this framework to part-based detection, and introduce a novel algorithm for interactive labeling of deformable part models. The labeling tool updates and displays in real-time the maximum likelihood location of all parts as the user clicks and drags the location of one or more parts. We demonstrate that the system can be used to efficiently and robustly train part and pose detectors on the CUB Birds-200-a challenging dataset of birds in unconstrained pose and environment."
]
} |
1704.03173 | 2951110097 | Given a convolutional neural network (CNN) that is pre-trained for object classification, this paper proposes to use active question-answering to semanticize neural patterns in conv-layers of the CNN and mine part concepts. For each part concept, we mine neural patterns in the pre-trained CNN, which are related to the target part, and use these patterns to construct an And-Or graph (AOG) to represent a four-layer semantic hierarchy of the part. As an interpretable model, the AOG associates different CNN units with different explicit object parts. We use an active human-computer communication to incrementally grow such an AOG on the pre-trained CNN as follows. We allow the computer to actively identify objects, whose neural patterns cannot be explained by the current AOG. Then, the computer asks human about the unexplained objects, and uses the answers to automatically discover certain CNN patterns corresponding to the missing knowledge. We incrementally grow the AOG to encode new knowledge discovered during the active-learning process. In experiments, our method exhibits high learning efficiency. Our method uses about 1 6-1 3 of the part annotations for training, but achieves similar or better part-localization performance than fast-RCNN methods. | | | In the scope of unsupervised learning and or weakly-supervised learning, modeling parts is usually more challenging than modeling entire objects. Given image-level labels (without object bounding boxes), object discovery @cite_48 @cite_28 @cite_39 and co-segmentation @cite_38 can be achieved by identifying common foreground patterns from complex background. In addition, there are some strong prior knowledges for object discovery, such as closed boundaries and common object structures. | {
"cite_N": [
"@cite_28",
"@cite_48",
"@cite_38",
"@cite_39"
],
"mid": [
"",
"1994488211",
"2114542651",
"2952004933"
],
"abstract": [
"",
"Successful methods for visual object recognition typically rely on training datasets containing lots of richly annotated images. Detailed image annotation, e.g. by object bounding boxes, however, is both expensive and often subjective. We describe a weakly supervised convolutional neural network (CNN) for object classification that relies only on image-level labels, yet can learn from cluttered scenes containing multiple objects. We quantify its object classification and object location prediction performance on the Pascal VOC 2012 (20 object classes) and the much larger Microsoft COCO (80 object classes) datasets. We find that the network (i) outputs accurate image-level labels, (ii) predicts approximate locations (but not extents) of objects, and (iii) performs comparably to its fully-supervised counterparts using object bounding box annotation for training.",
"We present an algorithm for Interactive Co-segmentation of a foreground object from a group of related images. While previous works in co-segmentation have focussed on unsupervised co-segmentation, we use successful ideas from the interactive object-cutout literature. We develop an algorithm that allows users to decide what foreground is, and then guide the output of the co-segmentation algorithm towards it via scribbles. Interestingly, keeping a user in the loop leads to simpler and highly parallelizable energy functions, allowing us to work with significantly more images per group. However, unlike the interactive single-image counterpart, a user cannot be expected to exhaustively examine all cutouts (from tens of images) returned by the system to make corrections. Hence, we propose iCoseg, an automatic recommendation system that intelligently recommends where the user should scribble next. We introduce and make publicly available the largest co-segmentation dataset yet, the CMU-Cornell iCoseg dataset, with 38 groups, 643 images, and pixelwise hand-annotated groundtruth. Through machine experiments and real user studies with our developed interface, we show that iCoseg can intelligently recommend regions to scribble on, and users following these recommendations can achieve good quality cutouts with significantly lower time and effort than exhaustively examining all cutouts.",
"We present an approach to learn a dense pixel-wise labeling from image-level tags. Each image-level tag imposes constraints on the output labeling of a Convolutional Neural Network (CNN) classifier. We propose Constrained CNN (CCNN), a method which uses a novel loss function to optimize for any set of linear constraints on the output space (i.e. predicted label distribution) of a CNN. Our loss formulation is easy to optimize and can be incorporated directly into standard stochastic gradient descent optimization. The key idea is to phrase the training objective as a biconvex optimization for linear models, which we then relax to nonlinear deep networks. Extensive experiments demonstrate the generality of our new learning framework. The constrained loss yields state-of-the-art results on weakly supervised semantic image segmentation. We further demonstrate that adding slightly more supervision can greatly improve the performance of the learning algorithm."
]
} |
1704.03173 | 2951110097 | Given a convolutional neural network (CNN) that is pre-trained for object classification, this paper proposes to use active question-answering to semanticize neural patterns in conv-layers of the CNN and mine part concepts. For each part concept, we mine neural patterns in the pre-trained CNN, which are related to the target part, and use these patterns to construct an And-Or graph (AOG) to represent a four-layer semantic hierarchy of the part. As an interpretable model, the AOG associates different CNN units with different explicit object parts. We use an active human-computer communication to incrementally grow such an AOG on the pre-trained CNN as follows. We allow the computer to actively identify objects, whose neural patterns cannot be explained by the current AOG. Then, the computer asks human about the unexplained objects, and uses the answers to automatically discover certain CNN patterns corresponding to the missing knowledge. We incrementally grow the AOG to encode new knowledge discovered during the active-learning process. In experiments, our method exhibits high learning efficiency. Our method uses about 1 6-1 3 of the part annotations for training, but achieves similar or better part-localization performance than fast-RCNN methods. | In contrast, to the best of our knowledge, there is no mechanism to distinguish a certain part concept from other parts of the same object. It is because 1) all the parts represent common foreground patterns among objects; 2) some parts ( the abdomen) do not have shape boundaries to identify their shape extent. Thus, up to now, people mainly extract implicit middle-level part patches @cite_8 , but it is difficult to capture explicit semantic meanings of these parts. | {
"cite_N": [
"@cite_8"
],
"mid": [
"2951702175"
],
"abstract": [
"The goal of this paper is to discover a set of discriminative patches which can serve as a fully unsupervised mid-level visual representation. The desired patches need to satisfy two requirements: 1) to be representative, they need to occur frequently enough in the visual world; 2) to be discriminative, they need to be different enough from the rest of the visual world. The patches could correspond to parts, objects, \"visual phrases\", etc. but are not restricted to be any one of them. We pose this as an unsupervised discriminative clustering problem on a huge dataset of image patches. We use an iterative procedure which alternates between clustering and training discriminative classifiers, while applying careful cross-validation at each step to prevent overfitting. The paper experimentally demonstrates the effectiveness of discriminative patches as an unsupervised mid-level visual representation, suggesting that it could be used in place of visual words for many tasks. Furthermore, discriminative patches can also be used in a supervised regime, such as scene classification, where they demonstrate state-of-the-art performance on the MIT Indoor-67 dataset."
]
} |
1704.03058 | 2605455247 | This work is about recognizing human activities occurring in videos at distinct semantic levels, including individual actions, interactions, and group activities. The recognition is realized using a two-level hierarchy of Long Short-Term Memory (LSTM) networks, forming a feed-forward deep architecture, which can be trained end-to-end. In comparison with existing architectures of LSTMs, we make two key contributions giving the name to our approach as Confidence-Energy Recurrent Network -- CERN. First, instead of using the common softmax layer for prediction, we specify a novel energy layer (EL) for estimating the energy of our predictions. Second, rather than finding the common minimum-energy class assignment, which may be numerically unstable under uncertainty, we specify that the EL additionally computes the p-values of the solutions, and in this way estimates the most confident energy minimum. The evaluation on the Collective Activity and Volleyball datasets demonstrates: (i) advantages of our two contributions relative to the common softmax and energy-minimization formulations and (ii) a superior performance relative to the state-of-the-art approaches. | Group activity recognition . Group activity recognition often requires the explicit representation of spatiotemporal structures of group activities defined in terms of individual actions and pairwise interactions. Previous work typically used graphical models @cite_7 @cite_1 @cite_16 @cite_19 @cite_22 or AND-OR grammar models @cite_14 @cite_0 to learn the structures grounded on hand-crafted features. Recent methods learn a graphical model, typically MRF @cite_5 @cite_11 or CRF @cite_8 @cite_2 @cite_24 , using recurrent neural networks (RNNs). Also, work on group activity recognition @cite_3 @cite_32 has demonstrated many advantages of using deep architectures of RNNs over the mentioned non-deep approaches. Our approach extends this work by replacing the RNN's softmax layer with a new energy layer, and by specifying a new energy-based model that takes into account p-values of the network's predictions. | {
"cite_N": [
"@cite_14",
"@cite_22",
"@cite_7",
"@cite_8",
"@cite_1",
"@cite_32",
"@cite_3",
"@cite_0",
"@cite_19",
"@cite_24",
"@cite_2",
"@cite_5",
"@cite_16",
"@cite_11"
],
"mid": [
"2171544105",
"",
"2047499569",
"",
"2057067088",
"2952563226",
"2930494921",
"2952390756",
"",
"2951729963",
"",
"2950774024",
"1989004008",
""
],
"abstract": [
"This paper addresses a new problem, that of multiscale activity recognition. Our goal is to detect and localize a wide range of activities, including individual actions and group activities, which may simultaneously co-occur in high-resolution video. The video resolution allows for digital zoom-in (or zoom-out) for examining fine details (or coarser scales), as needed for recognition. The key challenge is how to avoid running a multitude of detectors at all spatiotemporal scales, and yet arrive at a holistically consistent video interpretation. To this end, we use a three-layered AND-OR graph to jointly model group activities, individual actions, and participating objects. The AND-OR graph allows a principled formulation of efficient, cost-sensitive inference via an explore-exploit strategy. Our inference optimally schedules the following computational processes: 1) direct application of activity detectors --- called α process; 2) bottom-up inference based on detecting activity parts --- called β process; and 3) top-down inference based on detecting activity context --- called γ process. The scheduling iteratively maximizes the log-posteriors of the resulting parse graphs. For evaluation, we have compiled and benchmarked a new dataset of high-resolution videos of group and individual activities co-occurring in a courtyard of the UCLA campus.",
"",
"In this paper, we go beyond recognizing the actions of individuals and focus on group activities. This is motivated from the observation that human actions are rarely performed in isolation; the contextual information of what other people in the scene are doing provides a useful cue for understanding high-level activities. We propose a novel framework for recognizing group activities which jointly captures the group activity, the individual person actions, and the interactions among them. Two types of contextual information, group-person interaction and person-person interaction, are explored in a latent variable framework. In particular, we propose three different approaches to model the person-person interaction. One approach is to explore the structures of person-person interaction. Differently from most of the previous latent structured models, which assume a predefined structure for the hidden layer, e.g., a tree structure, we treat the structure of the hidden layer as a latent variable and implicitly infer it during learning and inference. The second approach explores person-person interaction in the feature level. We introduce a new feature representation called the action context (AC) descriptor. The AC descriptor encodes information about not only the action of an individual person in the video, but also the behavior of other people nearby. The third approach combines the above two. Our experimental results demonstrate the benefit of using contextual information for disambiguating group activities.",
"",
"We present a hierarchical model for human activity recognition in entire multi-person scenes. Our model describes human behaviour at multiple levels of detail, ranging from low-level actions through to high-level events. We also include a model of social roles, the expected behaviours of certain people, or groups of people, in a scene. The hierarchical model includes these varied representations, and various forms of interactions between people present in a scene. The model is trained in a discriminative max-margin framework. Experimental results demonstrate that this model can improve performance at all considered levels of detail, on two challenging datasets.",
"Rich semantic relations are important in a variety of visual recognition problems. As a concrete example, group activity recognition involves the interactions and relative spatial relations of a set of people in a scene. State of the art recognition methods center on deep learning approaches for training highly effective, complex classifiers for interpreting images. However, bridging the relatively low-level concepts output by these methods to interpret higher-level compositional scenes remains a challenge. Graphical models are a standard tool for this task. In this paper, we propose a method to integrate graphical models and deep neural networks into a joint framework. Instead of using a traditional inference method, we use a sequential inference modeled by a recurrent neural network. Beyond this, the appropriate structure for inference can be learned by imposing gates on edges between nodes. Empirical results on group activity recognition demonstrate the potential of this model to handle highly structured learning tasks.",
"In this paper we present an approach for classifying the activity performed by a group of people in a video sequence. This problem of group activity recognition can be addressed by examining individual person actions and their relations. Temporal dynamics exist both at the level of individual person actions as well as at the level of group activity. Given a video sequence as input, methods can be developed to capture these dynamics at both person-level and group-level detail. We build a deep model to capture these dynamics based on LSTM (long short-term memory) models. In order to model both person-level and group-level dynamics, we present a 2-stage deep temporal model for the group activity recognition problem. In our approach, one LSTM model is designed to represent action dynamics of individual people in a video sequence and another LSTM model is designed to aggregate person-level information for group activity recognition. We collected a new dataset consisting of volleyball videos labeled with individual and group activities in order to evaluate our method. Experimental results on this new Volleyball Dataset and the standard benchmark Collective Activity Dataset demonstrate the efficacy of the proposed models.",
"With the advent of drones, aerial video analysis becomes increasingly important; yet, it has received scant attention in the literature. This paper addresses a new problem of parsing low-resolution aerial videos of large spatial areas, in terms of 1) grouping, 2) recognizing events and 3) assigning roles to people engaged in events. We propose a novel framework aimed at conducting joint inference of the above tasks, as reasoning about each in isolation typically fails in our setting. Given noisy tracklets of people and detections of large objects and scene surfaces (e.g., building, grass), we use a spatiotemporal AND-OR graph to drive our joint inference, using Markov Chain Monte Carlo and dynamic programming. We also introduce a new formalism of spatiotemporal templates characterizing latent sub-events. For evaluation, we have collected and released a new aerial videos dataset using a hex-rotor flying over picnic areas rich with group events. Our results demonstrate that we successfully address above inference tasks under challenging conditions.",
"",
"By taking the semantic object parsing task as an exemplar application scenario, we propose the Graph Long Short-Term Memory (Graph LSTM) network, which is the generalization of LSTM from sequential data or multi-dimensional data to general graph-structured data. Particularly, instead of evenly and fixedly dividing an image to pixels or patches in existing multi-dimensional LSTM structures (e.g., Row, Grid and Diagonal LSTMs), we take each arbitrary-shaped superpixel as a semantically consistent node, and adaptively construct an undirected graph for each image, where the spatial relations of the superpixels are naturally used as edges. Constructed on such an adaptive graph topology, the Graph LSTM is more naturally aligned with the visual patterns in the image (e.g., object boundaries or appearance similarities) and provides a more economical information propagation route. Furthermore, for each optimization step over Graph LSTM, we propose to use a confidence-driven scheme to update the hidden and memory states of nodes progressively till all nodes are updated. In addition, for each node, the forgets gates are adaptively learned to capture different degrees of semantic correlation with neighboring nodes. Comprehensive evaluations on four diverse semantic object parsing datasets well demonstrate the significant superiority of our Graph LSTM over other state-of-the-art solutions.",
"",
"Many problems in real-world applications involve predicting several random variables which are statistically related. Markov random fields (MRFs) are a great mathematical tool to encode such relationships. The goal of this paper is to combine MRFs with deep learning algorithms to estimate complex representations while taking into account the dependencies between the output random variables. Towards this goal, we propose a training algorithm that is able to learn structured models jointly with deep features that form the MRF potentials. Our approach is efficient as it blends learning and inference and makes use of GPU acceleration. We demonstrate the effectiveness of our algorithm in the tasks of predicting words from noisy images, as well as multi-class classification of Flickr photographs. We show that joint learning of the deep features and the MRF parameters results in significant performance gains.",
"We deal with the problem of recognizing social roles played by people in an event. Social roles are governed by human interactions, and form a fundamental component of human event description. We focus on a weakly supervised setting, where we are provided different videos belonging to an event class, without training role labels. Since social roles are described by the interaction between people in an event, we propose a Conditional Random Field to model the inter-role interactions, along with person specific social descriptors. We develop tractable variational inference to simultaneously infer model weights, as well as role assignment to all people in the videos. We also present a novel YouTube social roles dataset with ground truth role annotations, and introduce annotations on a subset of videos from the TRECVID-MED11 [1] event kits for evaluation purposes. The performance of the model is compared against different baseline methods on these datasets.",
""
]
} |
1704.03058 | 2605455247 | This work is about recognizing human activities occurring in videos at distinct semantic levels, including individual actions, interactions, and group activities. The recognition is realized using a two-level hierarchy of Long Short-Term Memory (LSTM) networks, forming a feed-forward deep architecture, which can be trained end-to-end. In comparison with existing architectures of LSTMs, we make two key contributions giving the name to our approach as Confidence-Energy Recurrent Network -- CERN. First, instead of using the common softmax layer for prediction, we specify a novel energy layer (EL) for estimating the energy of our predictions. Second, rather than finding the common minimum-energy class assignment, which may be numerically unstable under uncertainty, we specify that the EL additionally computes the p-values of the solutions, and in this way estimates the most confident energy minimum. The evaluation on the Collective Activity and Volleyball datasets demonstrates: (i) advantages of our two contributions relative to the common softmax and energy-minimization formulations and (ii) a superior performance relative to the state-of-the-art approaches. | Energy-based learning . While energy-based formulations of inference and learning are common in non-deep group activity recognition @cite_16 @cite_19 @cite_22 @cite_0 , they are seldom used for deep architectures. Recently, a few approaches have tried to learn an energy-based model @cite_27 @cite_18 using deep neural networks @cite_26 @cite_9 . They have demonstrated that energy-based objectives have great potential in improving the performance of structured predictions, especially when training data are limited. Our approach extends this work by regularizing the energy-based objective such that it additionally accounts for the confidence of predictions. | {
"cite_N": [
"@cite_18",
"@cite_26",
"@cite_22",
"@cite_9",
"@cite_0",
"@cite_19",
"@cite_27",
"@cite_16"
],
"mid": [
"2161914416",
"2276303159",
"",
"2521028896",
"2952390756",
"",
"10021998",
"1989004008"
],
"abstract": [
"Energy-Based Models (EBMs) capture dependencies between variables by associating a scalar energy to each configuration of the variab les. Inference consists in clamping the value of observed variables and finding config urations of the remaining variables that minimize the energy. Learning consists in finding an energy function in which observed configurations of the variables a re given lower energies than unobserved ones. The EBM approach provides a common theoretical framework for many learning models, including traditional discr iminative and generative approaches, as well as graph-transformer networks, co nditional random fields, maximum margin Markov networks, and several manifold learning methods. Probabilistic models must be properly normalized, which sometimes requires evaluating intractable integrals over the space of all poss ible variable configurations. Since EBMs have no requirement for proper normalization, this problem is naturally circumvented. EBMs can be viewed as a form of non-probabilistic factor graphs, and they provide considerably more flexibility in th e design of architectures and training criteria than probabilistic approaches .",
"We introduce structured prediction energy networks (SPENs), a flexible framework for structured prediction. A deep architecture is used to define an energy function of candidate labels, and then predictions are produced by using back-propagation to iteratively optimize the energy with respect to the labels. This deep architecture captures dependencies between labels that would lead to intractable graphical models, and performs structure learning by automatically learning discriminative features of the structured output. One natural application of our technique is multi-label classification, which traditionally has required strict prior assumptions about the interactions between labels to ensure tractable learning and prediction. We are able to apply SPENs to multi-label problems with substantially larger label sets than previous applications of structured prediction, while modeling high-order interactions using minimal structural assumptions. Overall, deep learning provides remarkable tools for learning features of the inputs to a prediction problem, and this work extends these techniques to learning features of structured outputs. Our experiments provide impressive performance on a variety of benchmark multi-label classification tasks, demonstrate that our technique can be used to provide interpretable structure learning, and illuminate fundamental trade-offs between feed-forward and iterative structured prediction.",
"",
"We introduce the \"Energy-based Generative Adversarial Network\" model (EBGAN) which views the discriminator as an energy function that attributes low energies to the regions near the data manifold and higher energies to other regions. Similar to the probabilistic GANs, a generator is seen as being trained to produce contrastive samples with minimal energies, while the discriminator is trained to assign high energies to these generated samples. Viewing the discriminator as an energy function allows to use a wide variety of architectures and loss functionals in addition to the usual binary classifier with logistic output. Among them, we show one instantiation of EBGAN framework as using an auto-encoder architecture, with the energy being the reconstruction error, in place of the discriminator. We show that this form of EBGAN exhibits more stable behavior than regular GANs during training. We also show that a single-scale architecture can be trained to generate high-resolution images.",
"With the advent of drones, aerial video analysis becomes increasingly important; yet, it has received scant attention in the literature. This paper addresses a new problem of parsing low-resolution aerial videos of large spatial areas, in terms of 1) grouping, 2) recognizing events and 3) assigning roles to people engaged in events. We propose a novel framework aimed at conducting joint inference of the above tasks, as reasoning about each in isolation typically fails in our setting. Given noisy tracklets of people and detections of large objects and scene surfaces (e.g., building, grass), we use a spatiotemporal AND-OR graph to drive our joint inference, using Markov Chain Monte Carlo and dynamic programming. We also introduce a new formalism of spatiotemporal templates characterizing latent sub-events. For evaluation, we have collected and released a new aerial videos dataset using a hex-rotor flying over picnic areas rich with group events. Our results demonstrate that we successfully address above inference tasks under challenging conditions.",
"",
"",
"We deal with the problem of recognizing social roles played by people in an event. Social roles are governed by human interactions, and form a fundamental component of human event description. We focus on a weakly supervised setting, where we are provided different videos belonging to an event class, without training role labels. Since social roles are described by the interaction between people in an event, we propose a Conditional Random Field to model the inter-role interactions, along with person specific social descriptors. We develop tractable variational inference to simultaneously infer model weights, as well as role assignment to all people in the videos. We also present a novel YouTube social roles dataset with ground truth role annotations, and introduce annotations on a subset of videos from the TRECVID-MED11 [1] event kits for evaluation purposes. The performance of the model is compared against different baseline methods on these datasets."
]
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.