aid stringlengths 9 15 | mid stringlengths 7 10 | abstract stringlengths 78 2.56k | related_work stringlengths 92 1.77k | ref_abstract dict |
|---|---|---|---|---|
1709.01630 | 2751270710 | We present a first-person method for cooperative basketball intention prediction: we predict with whom the camera wearer will cooperate in the near future from unlabeled first-person images. This is a challenging task that requires inferring the camera wearer's visual attention, and decoding the social cues of other players. Our key observation is that a first-person view provides strong cues to infer the camera wearer's momentary visual attention, and his her intentions. We exploit this observation by proposing a new cross-model EgoSupervision learning scheme that allows us to predict with whom the camera wearer will cooperate in the near future, without using manually labeled intention labels. Our cross-model EgoSupervision operates by transforming the outputs of a pretrained pose-estimation network, into pseudo ground truth labels, which are then used as a supervisory signal to train a new network for a cooperative intention task. We evaluate our method, and show that it achieves similar or even better accuracy than the fully supervised methods do. | In the past, most first-person methods have focused on first-person object detection @cite_45 @cite_19 @cite_43 @cite_33 @cite_26 , or activity recognition @cite_1 @cite_21 @cite_36 @cite_22 @cite_11 @cite_17 . Several methods have employed first-person videos to summarize videos @cite_45 @cite_47 while recently the work in @cite_24 proposed to predict the camera wearer's engagement detection from first-person videos. The work in @cite_42 used a group of people wearing first-person cameras to infer their social interactions such as monologues, dialogues, or discussions. The method in @cite_15 predicted physical forces experienced by the camera wearer, while the work in @cite_14 recognized the activities performed in various extreme sports. Several recent methods @cite_44 @cite_4 also predicted the camera wearer's movement trajectories. Finally, first-person cameras have also been used for various robotics applications @cite_48 @cite_34 | {
"cite_N": [
"@cite_22",
"@cite_36",
"@cite_42",
"@cite_44",
"@cite_43",
"@cite_15",
"@cite_4",
"@cite_48",
"@cite_21",
"@cite_17",
"@cite_26",
"@cite_19",
"@cite_34",
"@cite_14",
"@cite_33",
"@cite_1",
"@cite_24",
"@cite_45",
"@cite_47",
"@cite_11"
],
"mid": [
"1947050545",
"",
"",
"",
"",
"",
"",
"",
"2432964524",
"2149276562",
"2296893412",
"",
"",
"",
"2031688197",
"2253138976",
"2338803139",
"2071711566",
"2120645068",
""
],
"abstract": [
"We address the challenging problem of recognizing the camera wearer's actions from videos captured by an egocentric camera. Egocentric videos encode a rich set of signals regarding the camera wearer, including head movement, hand pose and gaze information. We propose to utilize these mid-level egocentric cues for egocentric action recognition. We present a novel set of egocentric features and show how they can be combined with motion and object features. The result is a compact representation with superior performance. In addition, we provide the first systematic evaluation of motion, object and egocentric cues in egocentric action recognition. Our benchmark leads to several surprising findings. These findings uncover the best practices for egocentric actions, with a significant performance boost over all previous state-of-the-art methods on three publicly available datasets.",
"",
"",
"",
"",
"",
"",
"",
"We focus on the problem of wearer's action recognition in first person a.k.a. egocentric videos. This problem is more challenging than third person activity recognition due to unavailability of wearer's pose and sharp movements in the videos caused by the natural head motion of the wearer. Carefully crafted features based on hands and objects cues for the problem have been shown to be successful for limited targeted datasets. We propose convolutional neural networks (CNNs) for end to end learning and classification of wearer's actions. The proposed network makes use of egocentric cues by capturing hand pose, head motion and saliency map. It is compact. It can also be trained from relatively small number of labeled egocentric videos that are available. We show that the proposed network can generalize and give state of the art performance on various disparate egocentric action datasets.",
"We present a method to analyze daily activities, such as meal preparation, using video from an egocentric camera. Our method performs inference about activities, actions, hands, and objects. Daily activities are a challenging domain for activity recognition which are well-suited to an egocentric approach. In contrast to previous activity recognition methods, our approach does not require pre-trained detectors for objects and hands. Instead we demonstrate the ability to learn a hierarchical model of an activity by exploiting the consistent appearance of objects, hands, and actions that results from the egocentric context. We show that joint modeling of activities, actions, and objects leads to superior performance in comparison to the case where they are considered independently. We introduce a novel representation of actions based on object-hand interactions and experimentally demonstrate the superior performance of our representation in comparison to standard activity representations such as bag of words.",
"Unlike traditional third-person cameras mounted on robots, a first-person camera, captures a person's visual sensorimotor object interactions from up close. In this paper, we study the tight interplay between our momentary visual attention and motor action with objects from a first-person camera. We propose a concept of action-objects---the objects that capture person's conscious visual (watching a TV) or tactile (taking a cup) interactions. Action-objects may be task-dependent but since many tasks share common person-object spatial configurations, action-objects exhibit a characteristic 3D spatial distance and orientation with respect to the person. We design a predictive model that detects action-objects using EgoNet, a joint two-stream network that holistically integrates visual appearance (RGB) and 3D spatial layout (depth and height) cues to predict per-pixel likelihood of action-objects. Our network also incorporates a first-person coordinate embedding, which is designed to learn a spatial distribution of the action-objects in the first-person data. We demonstrate EgoNet's predictive power, by showing that it consistently outperforms previous baseline approaches. Furthermore, EgoNet also exhibits a strong generalization ability, i.e., it predicts semantically meaningful objects in novel first-person datasets. Our method's ability to effectively detect action-objects could be used to improve robots' understanding of human-object interactions.",
"",
"",
"",
"This paper addresses the problem of learning object models from egocentric video of household activities, using extremely weak supervision. For each activity sequence, we know only the names of the objects which are present within it, and have no other knowledge regarding the appearance or location of objects. The key to our approach is a robust, unsupervised bottom up segmentation method, which exploits the structure of the egocentric domain to partition each frame into hand, object, and background categories. By using Multiple Instance Learning to match object instances across sequences, we discover and localize object occurrences. Object representations are refined through transduction and object-level classifiers are trained. We demonstrate encouraging results in detecting novel object instances using models produced by weakly-supervised learning.",
"In this paper, we study the problem of recognizing human actions in the presence of a single egocentric camera and multiple static cameras. Some actions are better presented in static cameras, where the whole body of an actor and the context of actions are visible. Some other actions are better recognized in egocentric cameras, where subtle movements of hands and complex object interactions are visible. In this paper, we introduce a model that can benefit from the best of both worlds by learning to predict the importance of each camera in recognizing actions in each frame. By joint discriminative learning of latent camera importance variables and action classifiers, our model achieves successful results in the challenging CMU-MMAC dataset. Our experimental results show significant gain in learning to use the cameras according to their predicted importance. The learned latent variables provide a level of understanding of a scene that enables automatic cinematography by smoothly switching between cameras in order to maximize the amount of relevant information in each frame.",
"In a wearable camera video, we see what the camera wearer sees. While this makes it easy to know roughly Open image in new window , it does not immediately reveal Open image in new window . Specifically, at what moments did his focus linger, as he paused to gather more information about something he saw? Knowing this answer would benefit various applications in video summarization and augmented reality, yet prior work focuses solely on the “what” question (estimating saliency, gaze) without considering the “when” (engagement). We propose a learning-based approach that uses long-term egomotion cues to detect engagement, specifically in browsing scenarios where one frequently takes in new visual information (e.g., shopping, touring). We introduce a large, richly annotated dataset for ego-engagement that is the first of its kind. Our approach outperforms a wide array of existing methods. We show engagement can be detected well independent of both scene appearance and the camera wearer’s identity.",
"We present a video summarization approach for egocentric or \"wearable\" camera data. Given hours of video, the proposed method produces a compact storyboard summary of the camera wearer's day. In contrast to traditional keyframe selection techniques, the resulting summary focuses on the most important objects and people with which the camera wearer interacts. To accomplish this, we develop region cues indicative of high-level saliency in egocentric video--such as the nearness to hands, gaze, and frequency of occurrence--and learn a regressor to predict the relative importance of any new region based on these cues. Using these predictions and a simple form of temporal event detection, our method selects frames for the storyboard that reflect the key object-driven happenings. We adjust the compactness of the final summary given either an importance selection criterion or a length budget; for the latter, we design an efficient dynamic programming solution that accounts for importance, visual uniqueness, and temporal displacement. Critically, the approach is neither camera-wearer-specific nor object-specific; that means the learned importance metric need not be trained for a given user or context, and it can predict the importance of objects and people that have never been seen previously. Our results on two egocentric video datasets show the method's promise relative to existing techniques for saliency and summarization.",
"We present a video summarization approach that discovers the story of an egocentric video. Given a long input video, our method selects a short chain of video sub shots depicting the essential events. Inspired by work in text analysis that links news articles over time, we define a random-walk based metric of influence between sub shots that reflects how visual objects contribute to the progression of events. Using this influence metric, we define an objective for the optimal k-subs hot summary. Whereas traditional methods optimize a summary's diversity or representative ness, ours explicitly accounts for how one sub-event \"leads to\" another-which, critically, captures event connectivity beyond simple object co-occurrence. As a result, our summaries provide a better sense of story. We apply our approach to over 12 hours of daily activity video taken from 23 unique camera wearers, and systematically evaluate its quality compared to multiple baselines with 34 human subjects.",
""
]
} |
1709.01630 | 2751270710 | We present a first-person method for cooperative basketball intention prediction: we predict with whom the camera wearer will cooperate in the near future from unlabeled first-person images. This is a challenging task that requires inferring the camera wearer's visual attention, and decoding the social cues of other players. Our key observation is that a first-person view provides strong cues to infer the camera wearer's momentary visual attention, and his her intentions. We exploit this observation by proposing a new cross-model EgoSupervision learning scheme that allows us to predict with whom the camera wearer will cooperate in the near future, without using manually labeled intention labels. Our cross-model EgoSupervision operates by transforming the outputs of a pretrained pose-estimation network, into pseudo ground truth labels, which are then used as a supervisory signal to train a new network for a cooperative intention task. We evaluate our method, and show that it achieves similar or even better accuracy than the fully supervised methods do. | With the introduction of supervised CNN models @cite_2 , there has been a lot of interest in adapting generic set of features @cite_28 for different tasks at hand @cite_23 @cite_20 @cite_7 @cite_39 @cite_3 @cite_27 . Recently, generic image classification features were successfully used for the tasks such as edge detection @cite_20 @cite_39 , object detection @cite_7 @cite_3 @cite_27 , and semantic segmentation @cite_46 @cite_37 @cite_6 @cite_38 . More related to our work, a recent line of research investigated how to transfer knowledge across different models by a combination of parameter updates @cite_13 @cite_0 @cite_35 , transformation learning @cite_8 @cite_10 , network distillation @cite_5 or cross-model supervision @cite_30 @cite_12 . The most similar to our work are the methods in @cite_30 @cite_12 that use cross-model supervision to transfer knowledge from one model to another. | {
"cite_N": [
"@cite_30",
"@cite_35",
"@cite_3",
"@cite_2",
"@cite_5",
"@cite_10",
"@cite_20",
"@cite_38",
"@cite_8",
"@cite_39",
"@cite_23",
"@cite_46",
"@cite_37",
"@cite_7",
"@cite_28",
"@cite_6",
"@cite_27",
"@cite_12",
"@cite_0",
"@cite_13"
],
"mid": [
"2463402750",
"",
"2613718673",
"",
"1821462560",
"",
"2949192504",
"1923697677",
"2090923791",
"",
"",
"",
"",
"2102605133",
"2953360861",
"2952632681",
"1487583988",
"2951874610",
"2953265577",
"1988348003"
],
"abstract": [
"We present a modality hallucination architecture for training an RGB object detection model which incorporates depth side information at training time. Our convolutional hallucination network learns a new and complementary RGB image representation which is taught to mimic convolutional mid-level features from a depth network. At test time images are processed jointly through the RGB and hallucination networks to produce improved detection performance. Thus, our method transfers information commonly extracted from depth training data to a network which can extract that information from the RGB counterpart. We present results on the standard NYUDv2 dataset and report improvement on the RGB detection task.",
"",
"State-of-the-art object detection networks depend on region proposal algorithms to hypothesize object locations. Advances like SPPnet [7] and Fast R-CNN [5] 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. RPNs are trained end-to-end to generate high-quality region proposals, which are used by Fast R-CNN for detection. With a simple alternating optimization, RPN and Fast R-CNN can be trained to share convolutional features. For the very deep VGG-16 model [19], 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 (73.2 mAP) and 2012 (70.4 mAP) using 300 proposals per image. Code is available at https: github.com ShaoqingRen faster_rcnn.",
"",
"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.",
"",
"Contour detection has been a fundamental component in many image segmentation and object detection systems. Most previous work utilizes low-level features such as texture or saliency to detect contours and then use them as cues for a higher-level task such as object detection. However, we claim that recognizing objects and predicting contours are two mutually related tasks. Contrary to traditional approaches, we show that we can invert the commonly established pipeline: instead of detecting contours with low-level cues for a higher-level recognition task, we exploit object-related features as high-level cues for contour detection. We achieve this goal by means of a multi-scale deep network that consists of five convolutional layers and a bifurcated fully-connected sub-network. The section from the input layer to the fifth convolutional layer is fixed and directly lifted from a pre-trained network optimized over a large-scale object classification task. This section of the network is applied to four different scales of the image input. These four parallel and identical streams are then attached to a bifurcated sub-network consisting of two independently-trained branches. One branch learns to predict the contour likelihood (with a classification objective) whereas the other branch is trained to learn the fraction of human labelers agreeing about the contour presence at a given point (with a regression criterion). We show that without any feature engineering our multi-scale deep learning approach achieves state-of-the-art results in contour detection.",
"Deep Convolutional Neural Networks (DCNNs) have recently shown state of the art performance in high level vision tasks, such as image classification and object detection. This work brings together methods from DCNNs and probabilistic graphical models for addressing the task of pixel-level classification (also called \"semantic image segmentation\"). We show that responses at the final layer of DCNNs are not sufficiently localized for accurate object segmentation. This is due to the very invariance properties that make DCNNs good for high level tasks. We overcome this poor localization property of deep networks by combining the responses at the final DCNN layer with a fully connected Conditional Random Field (CRF). Qualitatively, our \"DeepLab\" system is able to localize segment boundaries at a level of accuracy which is beyond previous methods. Quantitatively, our method sets the new state-of-art at the PASCAL VOC-2012 semantic image segmentation task, reaching 71.6 IOU accuracy in the test set. We show how these results can be obtained efficiently: Careful network re-purposing and a novel application of the 'hole' algorithm from the wavelet community allow dense computation of neural net responses at 8 frames per second on a modern GPU.",
"In real-world applications, “what you saw” during training is often not “what you get” during deployment: the distribution and even the type and dimensionality of features can change from one dataset to the next. In this paper, we address the problem of visual domain adaptation for transferring object models from one dataset or visual domain to another. We introduce ARC-t, a flexible model for supervised learning of non-linear transformations between domains. Our method is based on a novel theoretical result demonstrating that such transformations can be learned in kernel space. Unlike existing work, our model is not restricted to symmetric transformations, nor to features of the same type and dimensionality, making it applicable to a significantly wider set of adaptation scenarios than previous methods. Furthermore, the method can be applied to categories that were not available during training. We demonstrate the ability of our method to adapt object recognition models under a variety of situations, such as differing imaging conditions, feature types and codebooks.",
"",
"",
"",
"",
"Object detection performance, as measured on the canonical PASCAL VOC dataset, has plateaued in the last few years. The best-performing methods are complex ensemble systems that typically combine multiple low-level image features with high-level context. In this paper, we propose a simple and scalable detection algorithm that improves mean average precision (mAP) by more than 30 relative to the previous best result on VOC 2012 -- achieving a mAP of 53.3 . Our approach combines two key insights: (1) one can apply high-capacity convolutional neural networks (CNNs) to bottom-up region proposals in order to localize and segment objects and (2) when labeled training data is scarce, supervised pre-training for an auxiliary task, followed by domain-specific fine-tuning, yields a significant performance boost. Since we combine region proposals with CNNs, we call our method R-CNN: Regions with CNN features. We also present experiments that provide insight into what the network learns, revealing a rich hierarchy of image features. Source code for the complete system is available at http: www.cs.berkeley.edu rbg rcnn.",
"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 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 an integrated framework for using Convolutional Networks for classification, localization and detection. We show how a multiscale and sliding window approach can be efficiently implemented within a ConvNet. We also introduce a novel deep learning approach to localization by learning to predict object boundaries. Bounding boxes are then accumulated rather than suppressed in order to increase detection confidence. We show that different tasks can be learned simultaneously using a single shared network. This integrated framework is the winner of the localization task of the ImageNet Large Scale Visual Recognition Challenge 2013 (ILSVRC2013) and obtained very competitive results for the detection and classifications tasks. In post-competition work, we establish a new state of the art for the detection task. Finally, we release a feature extractor from our best model called OverFeat.",
"In this work we propose a technique that transfers supervision between images from different modalities. We use learned representations from a large labeled modality as a supervisory signal for training representations for a new unlabeled paired modality. Our method enables learning of rich representations for unlabeled modalities and can be used as a pre-training procedure for new modalities with limited labeled data. We show experimental results where we transfer supervision from labeled RGB images to unlabeled depth and optical flow images and demonstrate large improvements for both these cross modal supervision transfers. Code, data and pre-trained models are available at this https URL",
"We propose a new learning method for heterogeneous domain adaptation (HDA), in which the data from the source domain and the target domain are represented by heterogeneous features with different dimensions. Using two different projection matrices, we first transform the data from two domains into a common subspace in order to measure the similarity between the data from two domains. We then propose two new feature mapping functions to augment the transformed data with their original features and zeros. The existing learning methods (e.g., SVM and SVR) can be readily incorporated with our newly proposed augmented feature representations to effectively utilize the data from both domains for HDA. Using the hinge loss function in SVM as an example, we introduce the detailed objective function in our method called Heterogeneous Feature Augmentation (HFA) for a linear case and also describe its kernelization in order to efficiently cope with the data with very high dimensions. Moreover, we also develop an alternating optimization algorithm to effectively solve the nontrivial optimization problem in our HFA method. Comprehensive experiments on two benchmark datasets clearly demonstrate that HFA outperforms the existing HDA methods.",
"Our objective is transfer training of a discriminatively trained object category detector, in order to reduce the number of training images required. To this end we propose three transfer learning formulations where a template learnt previously for other categories is used to regularize the training of a new category. All the formulations result in convex optimization problems. Experiments (on PASCAL VOC) demonstrate significant performance gains by transfer learning from one class to another (e.g. motorbike to bicycle), including one-shot learning, specialization from class to a subordinate class (e.g. from quadruped to horse) and transfer using multiple components. In the case of multiple training samples it is shown that a detection performance approaching that of the state of the art can be achieved with substantially fewer training samples."
]
} |
1709.01727 | 2751748110 | Scene text recognition has attracted great interests from the computer vision and pattern recognition community in recent years. State-of-the-art methods use concolutional neural networks (CNNs), recurrent neural networks with long short-term memory (RNN-LSTM) or the combination of them. In this paper, we investigate the intrinsic characteristics of text recognition, and inspired by human cognition mechanisms in reading texts, we propose a scene text recognition method with character models on convolutional feature map. The method simultaneously detects and recognizes characters by sliding the text line image with character models, which are learned end-to-end on text line images labeled with text transcripts. The character classifier outputs on the sliding windows are normalized and decoded with Connectionist Temporal Classification (CTC) based algorithm. Compared to previous methods, our method has a number of appealing properties: (1) It avoids the difficulty of character segmentation which hinders the performance of segmentation-based recognition methods; (2) The model can be trained simply and efficiently because it avoids gradient vanishing exploding in training RNN-LSTM based models; (3) It bases on character models trained free of lexicon, and can recognize unknown words. (4) The recognition process is highly parallel and enables fast recognition. Our experiments on several challenging English and Chinese benchmarks, including the IIIT-5K, SVT, ICDAR03 13 and TRW15 datasets, demonstrate that the proposed method yields superior or comparable performance to state-of-the-art methods while the model size is relatively small. | In general, explicit segmentation methods consists of character segmentation and word recognition. The recognition performance largely relies on character segmentation. The existing segmentation methods roughly fall in two categories: binarization based and detection based. Binarization based methods find segmentation points after binarization. Niblack's adaptive binarization and Extremal Regions (ERs) are two typical binarization based methods, which are employed in @cite_20 and @cite_32 , respectively. However, since text in natural scene image suffers from uneven illumination and complex backgrounds, binarization can hardly give satisfactory results. Detection based methods bypass the binarization by adopting multi-scale sliding window strategy to get candidate characters from the original image directly. For example, the methods in @cite_4 @cite_2 @cite_24 directly extract features from the original image and use various classifiers to decide whether a character exist in the center of a sliding window. @cite_25 employ a part-based tree-structured model and a sliding window classification to localize the characters in the window. Detection based methods overcome the difficulty of character segmentation and have shown good performance. | {
"cite_N": [
"@cite_4",
"@cite_32",
"@cite_24",
"@cite_2",
"@cite_25",
"@cite_20"
],
"mid": [
"1998042868",
"",
"1981283549",
"2049951199",
"2069472161",
"2122221966"
],
"abstract": [
"This paper focuses on the problem of word detection and recognition in natural images. The problem is significantly more challenging than reading text in scanned documents, and has only recently gained attention from the computer vision community. Sub-components of the problem, such as text detection and cropped image word recognition, have been studied in isolation [7, 4, 20]. However, what is unclear is how these recent approaches contribute to solving the end-to-end problem of word recognition. We fill this gap by constructing and evaluating two systems. The first, representing the de facto state-of-the-art, is a two stage pipeline consisting of text detection followed by a leading OCR engine. The second is a system rooted in generic object recognition, an extension of our previous work in [20]. We show that the latter approach achieves superior performance. While scene text recognition has generally been treated with highly domain-specific methods, our results demonstrate the suitability of applying generic computer vision methods. Adopting this approach opens the door for real world scene text recognition to benefit from the rapid advances that have been taking place in object recognition.",
"",
"The problem of recognizing text in images taken in the wild has gained significant attention from the computer vision community in recent years. Contrary to recognition of printed documents, recognizing scene text is a challenging problem. We focus on the problem of recognizing text extracted from natural scene images and the web. Significant attempts have been made to address this problem in the recent past. However, many of these works benefit from the availability of strong context, which naturally limits their applicability. In this work we present a framework that uses a higher order prior computed from an English dictionary to recognize a word, which may or may not be a part of the dictionary. We show experimental results on publicly available datasets. Furthermore, we introduce a large challenging word dataset with five thousand words to evaluate various steps of our method exhaustively. The main contributions of this work are: (1) We present a framework, which incorporates higher order statistical language models to recognize words in an unconstrained manner (i.e. we overcome the need for restricted word lists, and instead use an English dictionary to compute the priors). (2) We achieve significant improvement (more than 20 ) in word recognition accuracies without using a restricted word list. (3) We introduce a large word recognition dataset (atleast 5 times larger than other public datasets) with character level annotation and benchmark it.",
"Scene text recognition has gained significant attention from the computer vision community in recent years. Recognizing such text is a challenging problem, even more so than the recognition of scanned documents. In this work, we focus on the problem of recognizing text extracted from street images. We present a framework that exploits both bottom-up and top-down cues. The bottom-up cues are derived from individual character detections from the image. We build a Conditional Random Field model on these detections to jointly model the strength of the detections and the interactions between them. We impose top-down cues obtained from a lexicon-based prior, i.e. language statistics, on the model. The optimal word represented by the text image is obtained by minimizing the energy function corresponding to the random field model. We show significant improvements in accuracies on two challenging public datasets, namely Street View Text (over 15 ) and ICDAR 2003 (nearly 10 ).",
"Scene text recognition has inspired great interests from the computer vision community in recent years. In this paper, we propose a novel scene text recognition method using part-based tree-structured character detection. Different from conventional multi-scale sliding window character detection strategy, which does not make use of the character-specific structure information, we use part-based tree-structure to model each type of character so as to detect and recognize the characters at the same time. While for word recognition, we build a Conditional Random Field model on the potential character locations to incorporate the detection scores, spatial constraints and linguistic knowledge into one framework. The final word recognition result is obtained by minimizing the cost function defined on the random field. Experimental results on a range of challenging public datasets (ICDAR 2003, ICDAR 2011, SVT) demonstrate that the proposed method outperforms state-of-the-art methods significantly both for character detection and word recognition.",
"We describe Photo OCR, a system for text extraction from images. Our particular focus is reliable text extraction from smartphone imagery, with the goal of text recognition as a user input modality similar to speech recognition. Commercially available OCR performs poorly on this task. Recent progress in machine learning has substantially improved isolated character classification, we build on this progress by demonstrating a complete OCR system using these techniques. We also incorporate modern data center-scale distributed language modelling. Our approach is capable of recognizing text in a variety of challenging imaging conditions where traditional OCR systems fail, notably in the presence of substantial blur, low resolution, low contrast, high image noise and other distortions. It also operates with low latency, mean processing time is 600 ms per image. We evaluate our system on public benchmark datasets for text extraction and outperform all previously reported results, more than halving the error rate on multiple benchmarks. The system is currently in use in many applications at Google, and is available as a user input modality in Google Translate for Android."
]
} |
1709.01727 | 2751748110 | Scene text recognition has attracted great interests from the computer vision and pattern recognition community in recent years. State-of-the-art methods use concolutional neural networks (CNNs), recurrent neural networks with long short-term memory (RNN-LSTM) or the combination of them. In this paper, we investigate the intrinsic characteristics of text recognition, and inspired by human cognition mechanisms in reading texts, we propose a scene text recognition method with character models on convolutional feature map. The method simultaneously detects and recognizes characters by sliding the text line image with character models, which are learned end-to-end on text line images labeled with text transcripts. The character classifier outputs on the sliding windows are normalized and decoded with Connectionist Temporal Classification (CTC) based algorithm. Compared to previous methods, our method has a number of appealing properties: (1) It avoids the difficulty of character segmentation which hinders the performance of segmentation-based recognition methods; (2) The model can be trained simply and efficiently because it avoids gradient vanishing exploding in training RNN-LSTM based models; (3) It bases on character models trained free of lexicon, and can recognize unknown words. (4) The recognition process is highly parallel and enables fast recognition. Our experiments on several challenging English and Chinese benchmarks, including the IIIT-5K, SVT, ICDAR03 13 and TRW15 datasets, demonstrate that the proposed method yields superior or comparable performance to state-of-the-art methods while the model size is relatively small. | In holistic recognition methods, @cite_11 use whole word sub-image features to recognize the word by comparing to simple black-and-white font-renderings of lexicon words. The methods in @cite_44 @cite_5 @cite_14 use word embedding, in which the recognition becomes a nearest neighbor classification by creating a joint embedding space for word images and the text. In @cite_7 @cite_28 , develop a powerful convolutional neural network to recognize English text by regarding every English word as a class. Thanks to the strong classification ability of CNN and the availability of large set of training images by synthesis, this method shows impressive performance on several benchmarks. Holistic recognition is confined by a pre-defined lexicon, however, although @cite_47 and @cite_6 propose another CNN based model which can recognize unconstrained words by predicting the character at each position in the output text, it is highly sensitive to the non-character space. | {
"cite_N": [
"@cite_14",
"@cite_7",
"@cite_28",
"@cite_6",
"@cite_44",
"@cite_5",
"@cite_47",
"@cite_11"
],
"mid": [
"",
"1491389626",
"",
"2294053032",
"2091526670",
"",
"1812645736",
"1975453746"
],
"abstract": [
"",
"In this work we present a framework for the recognition of natural scene text. Our framework does not require any human-labelled data, and performs word recognition on the whole image holistically, departing from the character based recognition systems of the past. The deep neural network models at the centre of this framework are trained solely on data produced by a synthetic text generation engine -- synthetic data that is highly realistic and sufficient to replace real data, giving us infinite amounts of training data. This excess of data exposes new possibilities for word recognition models, and here we consider three models, each one \"reading\" words in a different way: via 90k-way dictionary encoding, character sequence encoding, and bag-of-N-grams encoding. In the scenarios of language based and completely unconstrained text recognition we greatly improve upon state-of-the-art performance on standard datasets, using our fast, simple machinery and requiring zero data-acquisition costs.",
"",
"We present recursive recurrent neural networks with attention modeling (R2AM) for lexicon-free optical character recognition in natural scene images. The primary advantages of the proposed method are: (1) use of recursive convolutional neural networks (CNNs), which allow for parametrically efficient and effective image feature extraction, (2) an implicitly learned character-level language model, embodied in a recurrent neural network which avoids the need to use N-grams, and (3) the use of a soft-attention mechanism, allowing the model to selectively exploit image features in a coordinated way, and allowing for end-to-end training within a standard backpropagation framework. We validate our method with state-of-the-art performance on challenging benchmark datasets: Street View Text, IIIT5k, ICDAR and Synth90k.",
"A system and method for comparing a text image and a character string are provided. The method includes embedding a character string into a vectorial space by extracting a set of features from the character string and generating a character string representation based on the extracted features, such as a spatial pyramid bag of characters (SPBOC) representation. A text image is embedded into a vectorial space by extracting a set of features from the text image and generating a text image representation based on the text image extracted features. A compatibility between the text image representation and the character string representation is computed, which includes computing a function of the text image representation and character string representation.",
"",
"We develop a representation suitable for the unconstrained recognition of words in natural images: the general case of no fixed lexicon and unknown length. To this end we propose a convolutional neural network (CNN) based architecture which incorporates a Conditional Random Field (CRF) graphical model, taking the whole word image as a single input. The unaries of the CRF are provided by a CNN that predicts characters at each position of the output, while higher order terms are provided by another CNN that detects the presence of N-grams. We show that this entire model (CRF, character predictor, N-gram predictor) can be jointly optimised by back-propagating the structured output loss, essentially requiring the system to perform multi-task learning, and training uses purely synthetically generated data. The resulting model is a more accurate system on standard real-world text recognition benchmarks than character prediction alone, setting a benchmark for systems that have not been trained on a particular lexicon. In addition, our model achieves state-of-the-art accuracy in lexicon-constrained scenarios, without being specifically modelled for constrained recognition. To test the generalisation of our model, we also perform experiments with random alpha-numeric strings to evaluate the method when no visual language model is applicable.",
"Recognizing text in images taken in the wild is a challenging problem that has received great attention in recent years. Previous methods addressed this problem by first detecting individual characters, and then forming them into words. Such approaches often suffer from weak character detections, due to large intra-class variations, even more so than characters from scanned documents. We take a different view of the problem and present a holistic word recognition framework. In this, we first represent the scene text image and synthetic images generated from lexicon words using gradient-based features. We then recognize the text in the image by matching the scene and synthetic image features with our novel weighted Dynamic Time Warping (wDTW) approach. We perform experimental analysis on challenging public datasets, such as Street View Text and ICDAR 2003. Our proposed method significantly outperforms our earlier work in (CVPR 2012), as well as many other recent works, such as (ECCV 2012), al.(ICPR 2012), (ICCV 2011)."
]
} |
1709.01186 | 2751279010 | Measuring the salience of a word is an essential step in numerous NLP tasks. Heuristic approaches such as tfidf have been used so far to estimate the salience of words. We propose (NWS) scores, unlike heuristics, are learnt from a corpus. Specifically, we learn word salience scores such that, using pre-trained word embeddings as the input, can accurately predict the words that appear in a sentence, given the words that appear in the sentences preceding or succeeding that sentence. Experimental results on sentence similarity prediction show that the learnt word salience scores perform comparably or better than some of the state-of-the-art approaches for representing sentences on benchmark datasets for sentence similarity, while using only a fraction of the training and prediction times required by prior methods. Moreover, our NWS scores positively correlate with psycholinguistic measures such as concreteness, and imageability implying a close connection to the salience as perceived by humans. | Surprisingly, averaging word embeddings to create sentence embeddings has shown comparable performances to sentence embeddings that are learnt using more sophisticated word-order sensitive methods. For example, @cite_6 proposed a method to find the optimal weights for combining word embeddings when creating sentence embeddings using unigram probabilities, by maximising the likelihood of the occurrences of words in a corpus. Siamese CBOW learns word embeddings such that we can accurately compute sentence embeddings by averaging the word embeddings. Although averaging is an order insensitive operator, @cite_0 empirically showed that it can accurately predict the content and word order in sentences. This can be understood intuitively by recalling that words that appear between two words are often different in contexts where those two words are swapped. For example, in the two sentences is a large that lives in Africa'' and Large such as live in Africa'', the words that appear in between and are different, giving rise to different sentence embeddings even when sentence embeddings are computed by averaging the individual word embeddings. Instead of considering all words equally for sentence embedding purposes, attention-based models learn the amount of weight (attention) we must assign to each word in a given context. | {
"cite_N": [
"@cite_0",
"@cite_6"
],
"mid": [
"2515741950",
"2752172973"
],
"abstract": [
"There is a lot of research interest in encoding variable length sentences into fixed length vectors, in a way that preserves the sentence meanings. Two common methods include representations based on averaging word vectors, and representations based on the hidden states of recurrent neural networks such as LSTMs. The sentence vectors are used as features for subsequent machine learning tasks or for pre-training in the context of deep learning. However, not much is known about the properties that are encoded in these sentence representations and about the language information they capture. We propose a framework that facilitates better understanding of the encoded representations. We define prediction tasks around isolated aspects of sentence structure (namely sentence length, word content, and word order), and score representations by the ability to train a classifier to solve each prediction task when using the representation as input. We demonstrate the potential contribution of the approach by analyzing different sentence representation mechanisms. The analysis sheds light on the relative strengths of different sentence embedding methods with respect to these low level prediction tasks, and on the effect of the encoded vector's dimensionality on the resulting representations.",
"The success of neural network methods for computing word embeddings has motivated methods for generating semantic embeddings of longer pieces of text, such as sentences and paragraphs. Surprisingly, (ICLR'16) showed that such complicated methods are outperformed, especially in out-of-domain (transfer learning) settings, by simpler methods involving mild retraining of word embeddings and basic linear regression. The method of requires retraining with a substantial labeled dataset such as Paraphrase Database (, 2013). @PARASPLIT The current paper goes further, showing that the following completely unsupervised sentence embedding is a formidable baseline: Use word embeddings computed using one of the popular methods on unlabeled corpus like Wikipedia, represent the sentence by a weighted average of the word vectors, and then modify them a bit using PCA SVD. This weighting improves performance by about 10 to 30 in textual similarity tasks, and beats sophisticated supervised methods including RNN's and LSTM's. It even improves 's embeddings. This simple method should be used as the baseline to beat in future, especially when labeled training data is scarce or nonexistent. @PARASPLIT The paper also gives a theoretical explanation of the success of the above unsupervised method using a latent variable generative model for sentences, which is a simple extension of the model in (TACL'16) with new \"smoothing\" terms that allow for words occurring out of context, as well as high probabilities for words like and, not in all contexts."
]
} |
1709.01186 | 2751279010 | Measuring the salience of a word is an essential step in numerous NLP tasks. Heuristic approaches such as tfidf have been used so far to estimate the salience of words. We propose (NWS) scores, unlike heuristics, are learnt from a corpus. Specifically, we learn word salience scores such that, using pre-trained word embeddings as the input, can accurately predict the words that appear in a sentence, given the words that appear in the sentences preceding or succeeding that sentence. Experimental results on sentence similarity prediction show that the learnt word salience scores perform comparably or better than some of the state-of-the-art approaches for representing sentences on benchmark datasets for sentence similarity, while using only a fraction of the training and prediction times required by prior methods. Moreover, our NWS scores positively correlate with psycholinguistic measures such as concreteness, and imageability implying a close connection to the salience as perceived by humans. | Our proposed method for learning NWS scores are based on the prior observation that averaging is an effective heuristic for creating sentence embeddings from word embeddings. However, unlike sentence embedding learning methods that do not learn word salience scores , our goal in this paper is to learn word salience scores and not sentence embeddings. We compute sentence embeddings only for the purpose of evaluating the word salience scores we learn. Moreover, our work differs from Siamese CBOW in that we do not learn word embeddings but take pre-trained word embeddings as the input for learning word salience scores. NWS scores we learn in this paper are also different from the salience scores learnt by @cite_6 because they do not constrain their word salience scores such that they can be used to predict the words that occur in adjacent sentences. | {
"cite_N": [
"@cite_6"
],
"mid": [
"2752172973"
],
"abstract": [
"The success of neural network methods for computing word embeddings has motivated methods for generating semantic embeddings of longer pieces of text, such as sentences and paragraphs. Surprisingly, (ICLR'16) showed that such complicated methods are outperformed, especially in out-of-domain (transfer learning) settings, by simpler methods involving mild retraining of word embeddings and basic linear regression. The method of requires retraining with a substantial labeled dataset such as Paraphrase Database (, 2013). @PARASPLIT The current paper goes further, showing that the following completely unsupervised sentence embedding is a formidable baseline: Use word embeddings computed using one of the popular methods on unlabeled corpus like Wikipedia, represent the sentence by a weighted average of the word vectors, and then modify them a bit using PCA SVD. This weighting improves performance by about 10 to 30 in textual similarity tasks, and beats sophisticated supervised methods including RNN's and LSTM's. It even improves 's embeddings. This simple method should be used as the baseline to beat in future, especially when labeled training data is scarce or nonexistent. @PARASPLIT The paper also gives a theoretical explanation of the success of the above unsupervised method using a latent variable generative model for sentences, which is a simple extension of the model in (TACL'16) with new \"smoothing\" terms that allow for words occurring out of context, as well as high probabilities for words like and, not in all contexts."
]
} |
1709.01140 | 2751368278 | The exponentially increasing use of moving platforms for video capture introduces the urgent need to develop the general background subtraction algorithms with the capability to deal with the moving background. In this paper, we propose a multilayer-based framework for online background subtraction for videos captured by moving cameras. Unlike the previous treatments of the problem, the proposed method is not restricted to binary segmentation of background and foreground, but formulates it as a multi-label segmentation problem by modeling multiple foreground objects in different layers when they appear simultaneously in the scene. We assign an independent processing layer to each foreground object, as well as the background, where both motion and appearance models are estimated, and a probability map is inferred using a Bayesian filtering framework. Finally, Multi-label Graph-cut on Markov Random Field is employed to perform pixel-wise labeling. Extensive evaluation results show that the proposed method outperforms state-of-the-art methods on challenging video sequences. | The freely moving camera introduces a movement in the projected background scene, and thus complicates the background subtraction problem. An intuitive idea to tackle such a problem is compensating the camera motion. A few pioneering works resort to estimating a homography @cite_45 that characterizes the geometric transformation of background scene between consecutive frames. Typically RANSAC @cite_18 and its variants MLESAC @cite_31 are employed to achieve robust estimation using many matches of feature points. Jin al @cite_7 model the scene as a set of planer regions where each background pixel is assumed to belong to one of these regions. Homographies are used to rectify each region to its corresponding planer representation in the model. Zamalieva al @cite_41 leverage geometric information to develop multiple transformation models, and choose one that best describes the relation between consecutive frames. | {
"cite_N": [
"@cite_18",
"@cite_7",
"@cite_41",
"@cite_45",
"@cite_31"
],
"mid": [
"2085261163",
"2147251136",
"16705709",
"2033819227",
""
],
"abstract": [
"A new paradigm, Random Sample Consensus (RANSAC), for fitting a model to experimental data is introduced. RANSAC is capable of interpreting smoothing data containing a significant percentage of gross errors, and is thus ideally suited for applications in automated image analysis where interpretation is based on the data provided by error-prone feature detectors. A major portion of this paper describes the application of RANSAC to the Location Determination Problem (LDP): Given an image depicting a set of landmarks with known locations, determine that point in space from which the image was obtained. In response to a RANSAC requirement, new results are derived on the minimum number of landmarks needed to obtain a solution, and algorithms are presented for computing these minimum-landmark solutions in closed form. These results provide the basis for an automatic system that can solve the LDP under difficult viewing",
"This paper proposes a novel multi-layer homography algorithm for background modeling from a free-moving camera. Background is composed of many planes. Different planes satisfy with different homographies which can be found by our algorithm. Each pixel except for the moving pixel definitely belongs to some plane. Rectified by the corresponding homography, each static pixel in the shared view can find its match in the previous frame. Thus, frames can be rectified to a specific viewpoint for background modeling. Experiment shows it is effective. Our approach can be used in motion detection from a free-moving camera.",
"We introduce a new approach to perform background subtraction in moving camera scenarios. Unlike previous treatments of the problem, we do not restrict the camera motion or the scene geometry. The proposed approach relies on Bayesian selection of the transformation that best describes the geometric relation between consecutive frames. Based on the selected transformation, we propagate a set of learned background and foreground appearance models using a single or a series of homography transforms. The propagated models are subjected to MAP-MRF optimization framework that combines motion, appearance, spatial, and temporal cues; the optimization process provides the final background foreground labels. Extensive experimental evaluation with challenging videos shows that the proposed method outperforms the baseline and state-of-the-art methods in most cases.",
"From the Publisher: A basic problem in computer vision is to understand the structure of a real world scene given several images of it. Recent major developments in the theory and practice of scene reconstruction are described in detail in a unified framework. The book covers the geometric principles and how to represent objects algebraically so they can be computed and applied. The authors provide comprehensive background material and explain how to apply the methods and implement the algorithms directly.",
""
]
} |
1709.01140 | 2751368278 | The exponentially increasing use of moving platforms for video capture introduces the urgent need to develop the general background subtraction algorithms with the capability to deal with the moving background. In this paper, we propose a multilayer-based framework for online background subtraction for videos captured by moving cameras. Unlike the previous treatments of the problem, the proposed method is not restricted to binary segmentation of background and foreground, but formulates it as a multi-label segmentation problem by modeling multiple foreground objects in different layers when they appear simultaneously in the scene. We assign an independent processing layer to each foreground object, as well as the background, where both motion and appearance models are estimated, and a probability map is inferred using a Bayesian filtering framework. Finally, Multi-label Graph-cut on Markov Random Field is employed to perform pixel-wise labeling. Extensive evaluation results show that the proposed method outperforms state-of-the-art methods on challenging video sequences. | Traditionally, statistical representations of the background scene have been proposed to estimate spatially extendable background models. Hayman al @cite_36 built a mixture of Gaussian mosaic background model. Ren al @cite_32 used motion compensation to predict the position of each pixel in a background map, and model the uncertainty of that prediction by a spacial Gaussian distribution. The construction of image mosaic associated with a traditional mixture Gaussian background model was also claimed to be effective in @cite_21 @cite_28 . However, the hyper-parameter required by this parametric model restricts its adaptability and application. On the contrary, we employ nonparametric Kernal Density Estimation method @cite_9 to build models of the appearance of foreground and background regions, making our approach more stable and applicable. | {
"cite_N": [
"@cite_36",
"@cite_28",
"@cite_9",
"@cite_21",
"@cite_32"
],
"mid": [
"2171885832",
"1971213697",
"2118143383",
"2125884443",
"1965428565"
],
"abstract": [
"Statistical background modelling and subtraction has proved to be a popular and effective class of algorithms for segmenting independently moving foreground objects out from a static background, without requiring any a priori information of the properties of foreground objects. We present two contributions on this topic, aimed towards robotics where an active head is mounted on a mobile vehicle. In periods when the vehicle's wheels are not driven, camera translation is virtually zero, and background subtraction techniques are applicable. This is also highly relevant to surveillance and video conferencing. The first part presents an efficient probabilistic framework for when the camera pans and tilts. A unified approach is developed for handling various sources of error, including motion blur, subpixel camera motion, mixed pixels at object boundaries, and also uncertainty in background stabilisation caused by noise, unmodelled radial distortion and small translations of the camera. The second contribution regards a Bayesian approach to specifically incorporate uncertainty concerning whether the background has yet been uncovered by moving foreground objects. This is an important requirement during initialisation of a system. We cannot assume that a background model is available in advance since that would involve storing models for each possible position, in every room, of the robot's operating environment.. Instead the background mode must be generated online, very possibly in the presence of moving objects.",
"Abstract A method of robust feature-detection is proposed for visual tracking with a pan-tilt head. Even with good foreground models, the tracking process is liable to be disrupted by strong features in the background. Previous researchers have shown that the disruption can be somewhat suppressed by the use of image-subtraction. Building on this idea, a more powerful statistical model of background intensity is proposed in which a Gaussian mixture distribution is fitted to each of the pixels on a ‘virtual’ image plane. A fitting algorithm of the ‘Expectation-Maximisation’ type proves to be particularly effective here. Practical tests with contour tracking show marked improvement over image subtraction methods. Since the burden of computation is off-line, the online tracking process can run in real-time, at video field-rate.",
"Automatic understanding of events happening at a site is the ultimate goal for many visual surveillance systems. Higher level understanding of events requires that certain lower level computer vision tasks be performed. These may include detection of unusual motion, tracking targets, labeling body parts, and understanding the interactions between people. To achieve many of these tasks, it is necessary to build representations of the appearance of objects in the scene. This paper focuses on two issues related to this problem. First, we construct a statistical representation of the scene background that supports sensitive detection of moving objects in the scene, but is robust to clutter arising out of natural scene variations. Second, we build statistical representations of the foreground regions (moving objects) that support their tracking and support occlusion reasoning. The probability density functions (pdfs) associated with the background and foreground are likely to vary from image to image and will not in general have a known parametric form. We accordingly utilize general nonparametric kernel density estimation techniques for building these statistical representations of the background and the foreground. These techniques estimate the pdf directly from the data without any assumptions about the underlying distributions. Example results from applications are presented.",
"We present a method for modeling a scene that is observed by a moving camera, where only a portion of the scene is visible at any time. This method uses mixture models to represent pixels in a panoramic view, and to construct a \"background image\" that contains only static (non-moving) parts of the scene. The method can be used to reliably detect moving objects in a video sequence, detect patterns of activity over a wide field of view, and remove moving objects from a video or panoramic mosaic. The method also yields improved results in detecting moving objects and in constructing mosaics in the presence of moving objects, when compared with techniques that are not based on scene modeling. We present examples illustrating the results.",
"A new background subtraction method is proposed in this paper for the foreground detection from a non-stationary background. Usually, motion compensation is required when applying background subtraction to a non-stationary background. In practice, it is difficult to realize this to sufficient pixel accuracy. The problem is further complicated when the moving objects to be detected tracked are small, since the pixel error in motion compensating the background will hide the small targets. A spatial distribution of Gaussians model is proposed to deal with moving object detection where the motion compensation is not exact but approximated. The distribution of each background pixel is temporally and spatially modeled. Based on this statistical model, a pixel in the current frame is classified as belonging to the foreground or background. In addition, a new background restoration and adaptation algorithm is developed for the non-stationary background over an extended period of time. Test cases involving a surveillance system to detect small moving objects (human and car) within a highly textured background and a pan-tilt human tracking system are demonstrated successfully."
]
} |
1709.01140 | 2751368278 | The exponentially increasing use of moving platforms for video capture introduces the urgent need to develop the general background subtraction algorithms with the capability to deal with the moving background. In this paper, we propose a multilayer-based framework for online background subtraction for videos captured by moving cameras. Unlike the previous treatments of the problem, the proposed method is not restricted to binary segmentation of background and foreground, but formulates it as a multi-label segmentation problem by modeling multiple foreground objects in different layers when they appear simultaneously in the scene. We assign an independent processing layer to each foreground object, as well as the background, where both motion and appearance models are estimated, and a probability map is inferred using a Bayesian filtering framework. Finally, Multi-label Graph-cut on Markov Random Field is employed to perform pixel-wise labeling. Extensive evaluation results show that the proposed method outperforms state-of-the-art methods on challenging video sequences. | The layered representation, referring to approaches that model the scene as a set of moving layers, has been used for foreground detection @cite_17 @cite_26 , motion segmentation @cite_38 @cite_39 @cite_3 . In @cite_17 , the background was modeled as the union of nonparametric layer-models to facilitate detecting the foreground under static or dynamic background. Kim al @cite_26 proposed a layered background model where a long-term background model is used besides several multiple short-term background models. Wang al @cite_38 used an iterative method to achieve layered-motion segmentation. Torr al @cite_1 modeled the layers as planes in 3D and integrating priors in a Bayesian framework. @cite_15 models spatial continuity while representing each layer as composed of a set of segments. A common theme of these layered models is the assumption that the video is available beforehand @cite_40 . Such an assumption prevents the use of such approaches for processing videos from streaming sources. Some dynamic textures methods @cite_10 @cite_5 @cite_4 also employed the layered model to tackle the complex dynamic background, but with stationary cameras. To the best of our knowledge, the proposed method is the first layered model applied in the moving camera scenarios. | {
"cite_N": [
"@cite_38",
"@cite_26",
"@cite_4",
"@cite_1",
"@cite_3",
"@cite_39",
"@cite_40",
"@cite_5",
"@cite_15",
"@cite_10",
"@cite_17"
],
"mid": [
"2142912032",
"1508368796",
"",
"2096204758",
"",
"",
"2011410589",
"",
"2082869876",
"",
"2002103521"
],
"abstract": [
"We describe a system for representing moving images with sets of overlapping layers. Each layer contains an intensity map that defines the additive values of each pixel, along with an alpha map that serves as a mask indicating the transparency. The layers are ordered in depth and they occlude each other in accord with the rules of compositing. Velocity maps define how the layers are to be warped over time. The layered representation is more flexible than standard image transforms and can capture many important properties of natural image sequences. We describe some methods for decomposing image sequences into layers using motion analysis, and we discuss how the representation may be used for image coding and other applications. >",
"Scene changes such as moved objects, parked vehicles, or opened closed doors need to be carefully handled so that interesting foreground targets can be detected along with the short-term background layers created by those changes. A simple layered modeling technique is embedded into a codebook-based background subtraction algorithm to update a background model. In addition, important issues related to background updating for visual surveillance are discussed. Experimental results on surveillance examples, such as unloaded packages and unattended objects, are presented by showing those objects as short-term background layers.",
"",
"This paper describes a Bayesian approach for modeling 3D scenes as collection of approximately planar layers that are arbitrarily positioned and oriented in the scene. In contrast to much of the previous work on layer-based motion modeling, which computes layered descriptions of 2D image motion, our work leads to a 3D description of the scene. There are two contributions within the paper. The first is to formulate the prior assumptions about the layers and scene within a Bayesian decision making framework which is used to automatically determine the number of layers and the assignment of individual pixels to layers. The second is algorithmic. In order to achieve the optimization, a Bayesian version of RANSAC is developed with which to initialize the segmentation. Then, a generalized expectation maximization method is used to find the MAP solution.",
"",
"",
"Background subtraction is a widely used concept for detection of moving objects in videos. In the last two decades there has been a lot of development in designing algorithms for background subtraction, as well as wide use of these algorithms in various important applications, such as visual surveillance, sports video analysis, motion capture, etc. Various statistical approaches have been proposed to model scene backgrounds. The concept of background subtraction also has been extended to detect objects from videos captured from moving cameras. This book reviews the concept and practice of background subtraction. We discuss several traditional statistical background subtraction models, including the widely used parametric Gaussian mixture models and non-parametric models. We also discuss the issue of shadow suppression, which is essential for human motion analysis applications. This book discusses approaches and tradeoffs for background maintenance. This book also reviews many of the recent developments in background subtraction paradigm. Recent advances in developing algorithms for background subtraction from moving cameras are described, including motion-compensation-based approaches and motion-segmentation-based approaches. For links to the videos to accompany this book, please see sites.google.com a morganclaypool.com backgroundsubtraction Table of Contents: Preface Acknowledgments Figure Credits Object Detection and Segmentation in Videos Background Subtraction from a Stationary Camera Background Subtraction from a Moving Camera Bibliography Author's Biography",
"",
"We present an unsupervised approach for learning a layered representation of a scene from a video for motion segmentation. Our method is applicable to any video containing piecewise parametric motion. The learnt model is a composition of layers, which consist of one or more segments. The shape of each segment is represented using a binary matte and its appearance is given by the rgb value for each point belonging to the matte. Included in the model are the effects of image projection, lighting, and motion blur. Furthermore, spatial continuity is explicitly modeled resulting in contiguous segments. Unlike previous approaches, our method does not use reference frame(s) for initialization. The two main contributions of our method are: (i) A novel algorithm for obtaining the initial estimate of the model by dividing the scene into rigidly moving components using efficient loopy belief propagation; and (ii) Refining the initial estimate using ? β-swap and ?-expansion algorithms, which guarantee a strong local minima. Results are presented on several classes of objects with different types of camera motion, e.g. videos of a human walking shot with static or translating cameras. We compare our method with the state of the art and demonstrate significant improvements.",
"",
"A framework for robust foreground detection that works under difficult conditions such as dynamic background and moderately moving camera is presented in this paper. The proposed method includes two main components: coarse scene representation as the union of pixel layers, and foreground detection in video by propagating these layers using a maximum-likelihood assignment. We first cluster into \"layers\" those pixels that share similar statistics. The entire scene is then modeled as the union of such nonparametric layer-models. An incoming pixel is detected as foreground if it does not adhere to these adaptive models of the background. A principled way of computing thresholds is used to achieve robust detection performance with a prespecified number of false alarms. Correlation between pixels in the spatial vicinity is exploited to deal with camera motion without precise registration or optical flow. The proposed technique adapts to changes in the scene, and allows to automatically convert persistent foreground objects to background and reconvert them to foreground when they become interesting. This simple framework addresses the important problem of robust foreground and unusual region detection, at about 10 frames per second on a standard laptop computer. The presentation of the proposed approach is complemented by results on challenging real data and comparisons with other standard techniques."
]
} |
1709.01353 | 2751825910 | Measuring visual (dis)similarity between two or more instances within a data distribution is a fundamental task in many applications, specially in image retrieval. Theoretically, non-metric distances are able to generate a more complex and accurate similarity model than metric distances, provided that the non-linear data distribution is precisely captured by the similarity model. In this work, we analyze a simple approach for deep learning networks to be used as an approximation of non-metric similarity functions and we study how these models generalize across different image retrieval datasets. | Content-based image retrieval searches for images by considering their visual content. Given a query image, pictures in a collection are ranked according to their visual similarity with respect to the query. Early methods represent the visual content of images by a set of hand-crafted features, such as SIFT @cite_22 . As a single image may contain hundreds of these features, aggregation techniques like bag-of-words (BOW) @cite_52 , Fisher Vectors @cite_36 or VLAD @cite_46 encode local descriptors into a compact vector, thereby improving computational efficiency and scalability. More recently, because of the latest advancements on deep learning, features obtained from convolutional neural networks (CNN) have rapidly become the new state-of-the-art in image retrieval. | {
"cite_N": [
"@cite_36",
"@cite_46",
"@cite_52",
"@cite_22"
],
"mid": [
"2071027807",
"2012592962",
"2131846894",
"2151103935"
],
"abstract": [
"The problem of large-scale image search has been traditionally addressed with the bag-of-visual-words (BOV). In this article, we propose to use as an alternative the Fisher kernel framework. We first show why the Fisher representation is well-suited to the retrieval problem: it describes an image by what makes it different from other images. One drawback of the Fisher vector is that it is high-dimensional and, as opposed to the BOV, it is dense. The resulting memory and computational costs do not make Fisher vectors directly amenable to large-scale retrieval. Therefore, we compress Fisher vectors to reduce their memory footprint and speed-up the retrieval. We compare three binarization approaches: a simple approach devised for this representation and two standard compression techniques. We show on two publicly available datasets that compressed Fisher vectors perform very well using as little as a few hundreds of bits per image, and significantly better than a very recent compressed BOV approach.",
"We address the problem of image search on a very large scale, where three constraints have to be considered jointly: the accuracy of the search, its efficiency, and the memory usage of the representation. We first propose a simple yet efficient way of aggregating local image descriptors into a vector of limited dimension, which can be viewed as a simplification of the Fisher kernel representation. We then show how to jointly optimize the dimension reduction and the indexing algorithm, so that it best preserves the quality of vector comparison. The evaluation shows that our approach significantly outperforms the state of the art: the search accuracy is comparable to the bag-of-features approach for an image representation that fits in 20 bytes. Searching a 10 million image dataset takes about 50ms.",
"We describe an approach to object and scene retrieval which searches for and localizes all the occurrences of a user outlined object in a video. The object is represented by a set of viewpoint invariant region descriptors so that recognition can proceed successfully despite changes in viewpoint, illumination and partial occlusion. The temporal continuity of the video within a shot is used to track the regions in order to reject unstable regions and reduce the effects of noise in the descriptors. The analogy with text retrieval is in the implementation where matches on descriptors are pre-computed (using vector quantization), and inverted file systems and document rankings are used. The result is that retrieved is immediate, returning a ranked list of key frames shots in the manner of Google. The method is illustrated for matching in two full length feature films.",
"This paper presents a method for extracting distinctive invariant features from images that can be used to perform reliable matching between different views of an object or scene. The features are invariant to image scale and rotation, and are shown to provide robust matching across a substantial range of affine distortion, change in 3D viewpoint, addition of noise, and change in illumination. The features are highly distinctive, in the sense that a single feature can be correctly matched with high probability against a large database of features from many images. This paper also describes an approach to using these features for object recognition. The recognition proceeds by matching individual features to a database of features from known objects using a fast nearest-neighbor algorithm, followed by a Hough transform to identify clusters belonging to a single object, and finally performing verification through least-squares solution for consistent pose parameters. This approach to recognition can robustly identify objects among clutter and occlusion while achieving near real-time performance."
]
} |
1709.01353 | 2751825910 | Measuring visual (dis)similarity between two or more instances within a data distribution is a fundamental task in many applications, specially in image retrieval. Theoretically, non-metric distances are able to generate a more complex and accurate similarity model than metric distances, provided that the non-linear data distribution is precisely captured by the similarity model. In this work, we analyze a simple approach for deep learning networks to be used as an approximation of non-metric similarity functions and we study how these models generalize across different image retrieval datasets. | Some of the most popular similarity learning work, such as OASIS @cite_6 and MLR @cite_32 , are based on linear metric learning by optimizing the weights of a linear transformation matrix. For example, @cite_67 proposed a framework for ranking elements in retrieval tasks by solving a linear optimization problem. Although linear methods are easier to optimize and less prone to overfitting, nonlinear algorithms are expected to achieve higher accuracy modeling the possible nonlinearities of data. | {
"cite_N": [
"@cite_67",
"@cite_32",
"@cite_6"
],
"mid": [
"2019863495",
"2158139921",
"1532499126"
],
"abstract": [
"We present a new framework for multimedia content analysis and retrieval which consists of two independent algorithms. First, we propose a new semi-supervised algorithm called ranking with Local Regression and Global Alignment (LRGA) to learn a robust Laplacian matrix for data ranking. In LRGA, for each data point, a local linear regression model is used to predict the ranking scores of its neighboring points. A unified objective function is then proposed to globally align the local models from all the data points so that an optimal ranking score can be assigned to each data point. Second, we propose a semi-supervised long-term Relevance Feedback (RF) algorithm to refine the multimedia data representation. The proposed long-term RF algorithm utilizes both the multimedia data distribution in multimedia feature space and the history RF information provided by users. A trace ratio optimization problem is then formulated and solved by an efficient algorithm. The algorithms have been applied to several content-based multimedia retrieval applications, including cross-media retrieval, image retrieval, and 3D motion pose data retrieval. Comprehensive experiments on four data sets have demonstrated its advantages in precision, robustness, scalability, and computational efficiency.",
"We study metric learning as a problem of information retrieval. We present a general metric learning algorithm, based on the structural SVM framework, to learn a metric such that rankings of data induced by distance from a query can be optimized against various ranking measures, such as AUC, Precision-at-k, MRR, MAP or NDCG. We demonstrate experimental results on standard classification data sets, and a large-scale online dating recommendation problem.",
"Learning a measure of similarity between pairs of objects is an important generic problem in machine learning. It is particularly useful in large scale applications like searching for an image that is similar to a given image or finding videos that are relevant to a given video. In these tasks, users look for objects that are not only visually similar but also semantically related to a given object. Unfortunately, the approaches that exist today for learning such semantic similarity do not scale to large data sets. This is both because typically their CPU and storage requirements grow quadratically with the sample size, and because many methods impose complex positivity constraints on the space of learned similarity functions. The current paper presents OASIS, an Online Algorithm for Scalable Image Similarity learning that learns a bilinear similarity measure over sparse representations. OASIS is an online dual approach using the passive-aggressive family of learning algorithms with a large margin criterion and an efficient hinge loss cost. Our experiments show that OASIS is both fast and accurate at a wide range of scales: for a data set with thousands of images, it achieves better results than existing state-of-the-art methods, while being an order of magnitude faster. For large, web scale, data sets, OASIS can be trained on more than two million images from 150K text queries within 3 days on a single CPU. On this large scale data set, human evaluations showed that 35 of the ten nearest neighbors of a given test image, as found by OASIS, were semantically relevant to that image. This suggests that query independent similarity could be accurately learned even for large scale data sets that could not be handled before."
]
} |
1709.01353 | 2751825910 | Measuring visual (dis)similarity between two or more instances within a data distribution is a fundamental task in many applications, specially in image retrieval. Theoretically, non-metric distances are able to generate a more complex and accurate similarity model than metric distances, provided that the non-linear data distribution is precisely captured by the similarity model. In this work, we analyze a simple approach for deep learning networks to be used as an approximation of non-metric similarity functions and we study how these models generalize across different image retrieval datasets. | Nonlinear similarity learning based on deep learning has been recently applied to many different visual contexts. In low-level image matching, CNNs have been trained to match pairs of patches for stereo matching @cite_12 @cite_27 and optical flow @cite_53 @cite_17 . In high-level image matching, deep learning techniques have been proposed to learn low-dimensional embedding spaces in face verification @cite_18 , retrieval @cite_47 @cite_58 , classification @cite_55 @cite_66 @cite_49 and product search @cite_16 , either by using siamese @cite_18 or triplet @cite_58 architectures. More recently, deep similarity learning has also been applied to fabric image retrieval @cite_38 by using triplets of samples to ensure that similar features are mapped closer than non-similar features. | {
"cite_N": [
"@cite_38",
"@cite_18",
"@cite_53",
"@cite_55",
"@cite_16",
"@cite_27",
"@cite_49",
"@cite_47",
"@cite_58",
"@cite_66",
"@cite_12",
"@cite_17"
],
"mid": [
"2773636887",
"2157364932",
"2951309005",
"2963775347",
"2021354639",
"",
"",
"2095569536",
"2037955148",
"",
"1955055330",
""
],
"abstract": [
"Abstract Fabric image retrieval is beneficial to many applications including clothing searching, online shopping and cloth modeling. Learning pairwise image similarity is of great importance to an image retrieval task. With the resurgence of Convolutional Neural Networks (CNNs), recent works have achieved significant progresses via deep representation learning with metric embedding, which drives similar examples close to each other in a feature space, and dissimilar ones apart from each other. In this paper, we propose a novel embedding method termed focus ranking that can be easily unified into a CNN for jointly learning image representations and metrics in the context of fine-grained fabric image retrieval. Focus ranking aims to rank similar examples higher than all dissimilar ones by penalizing ranking disorders via the minimization of the overall cost attributed to similar samples being ranked below dissimilar ones. At the training stage, training samples are organized into focus ranking units for efficient optimization. We build a large-scale fabric image retrieval dataset (FIRD) with about 25,000 images of 4300 fabrics, and test the proposed model on the FIRD dataset. Experimental results show the superiority of the proposed model over existing metric embedding models.",
"We present a method for training a similarity metric from data. The method can be used for recognition or verification applications where the number of categories is very large and not known during training, and where the number of training samples for a single category is very small. The idea is to learn a function that maps input patterns into a target space such that the L sub 1 norm in the target space approximates the \"semantic\" distance in the input space. The method is applied to a face verification task. The learning process minimizes a discriminative loss function that drives the similarity metric to be small for pairs of faces from the same person, and large for pairs from different persons. The mapping from raw to the target space is a convolutional network whose architecture is designed for robustness to geometric distortions. The system is tested on the Purdue AR face database which has a very high degree of variability in the pose, lighting, expression, position, and artificial occlusions such as dark glasses and obscuring scarves.",
"Convolutional neural networks (CNNs) have recently been very successful in a variety of computer vision tasks, especially on those linked to recognition. Optical flow estimation has not been among the tasks where CNNs were successful. In this paper we construct appropriate CNNs which are capable of solving the optical flow estimation problem as a supervised learning task. We propose and compare two architectures: a generic architecture and another one including a layer that correlates feature vectors at different image locations. Since existing ground truth data sets are not sufficiently large to train a CNN, we generate a synthetic Flying Chairs dataset. We show that networks trained on this unrealistic data still generalize very well to existing datasets such as Sintel and KITTI, achieving competitive accuracy at frame rates of 5 to 10 fps.",
"Deep learning has proven itself as a successful set of models for learning useful semantic representations of data. These, however, are mostly implicitly learned as part of a classification task. In this paper we propose the triplet network model, which aims to learn useful representations by distance comparisons. A similar model was defined by (2014), tailor made for learning a ranking for image information retrieval. Here we demonstrate using various datasets that our model learns a better representation than that of its immediate competitor, the Siamese network. We also discuss future possible usage as a framework for unsupervised learning.",
"Popular sites like Houzz, Pinterest, and LikeThatDecor, have communities of users helping each other answer questions about products in images. In this paper we learn an embedding for visual search in interior design. Our embedding contains two different domains of product images: products cropped from internet scenes, and products in their iconic form. With such a multi-domain embedding, we demonstrate several applications of visual search including identifying products in scenes and finding stylistically similar products. To obtain the embedding, we train a convolutional neural network on pairs of images. We explore several training architectures including re-purposing object classifiers, using siamese networks, and using multitask learning. We evaluate our search quantitatively and qualitatively and demonstrate high quality results for search across multiple visual domains, enabling new applications in interior design.",
"",
"",
"Recent years have witnessed extensive studies on distance metric learning (DML) for improving similarity search in multimedia information retrieval tasks. Despite their successes, most existing DML methods suffer from two critical limitations: (i) they typically attempt to learn a linear distance function on the input feature space, in which the assumption of linearity limits their capacity of measuring the similarity on complex patterns in real-world applications; (ii) they are often designed for learning distance metrics on uni-modal data, which may not effectively handle the similarity measures for multimedia objects with multimodal representations. To address these limitations, in this paper, we propose a novel framework of online multimodal deep similarity learning (OMDSL), which aims to optimally integrate multiple deep neural networks pretrained with stacked denoising autoencoder. In particular, the proposed framework explores a unified two-stage online learning scheme that consists of (i) learning a flexible nonlinear transformation function for each individual modality, and (ii) learning to find the optimal combination of multiple diverse modalities simultaneously in a coherent process. We conduct an extensive set of experiments to evaluate the performance of the proposed algorithms for multimodal image retrieval tasks, in which the encouraging results validate the effectiveness of the proposed technique.",
"",
"",
"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.",
""
]
} |
1709.01353 | 2751825910 | Measuring visual (dis)similarity between two or more instances within a data distribution is a fundamental task in many applications, specially in image retrieval. Theoretically, non-metric distances are able to generate a more complex and accurate similarity model than metric distances, provided that the non-linear data distribution is precisely captured by the similarity model. In this work, we analyze a simple approach for deep learning networks to be used as an approximation of non-metric similarity functions and we study how these models generalize across different image retrieval datasets. | In general, these methods rely on learning a mapping from image pixels to a low dimensional target space to compute the final similarity decision by using a standard metric. They are designed to find the best projection in which a linear distance can be successfully applied. Instead of projecting the visual data into some linear space, that may or may not exist, our approach seeks to learn the non-metric visual similarity score itself. Similarly, @cite_8 and @cite_60 used a CNN to decide whether or not two input images are a match, applied to pedestrian re-identification and patch matching, respectively. In these methods, the networks are trained as a binary classification problem (i.e. same or different pedestrian patch), whereas in an image retrieval ranking problem, a regression score is required. Inspired by the results of @cite_59 , which showed that combining deep features with similarity learning techniques can be very beneficial for the performance of image retrieval systems, we propose to train a deep learning algorithm to learn non-metric similarities for image retrieval and improve results in top of high quality image representation methods. | {
"cite_N": [
"@cite_60",
"@cite_59",
"@cite_8"
],
"mid": [
"1929856797",
"2123229215",
"1982925187"
],
"abstract": [
"Motivated by recent successes on learning feature representations and on learning feature comparison functions, we propose a unified approach to combining both for training a patch matching system. Our system, dubbed Match-Net, consists of a deep convolutional network that extracts features from patches and a network of three fully connected layers that computes a similarity between the extracted features. To ensure experimental repeatability, we train MatchNet on standard datasets and employ an input sampler to augment the training set with synthetic exemplar pairs that reduce overfitting. Once trained, we achieve better computational efficiency during matching by disassembling MatchNet and separately applying the feature computation and similarity networks in two sequential stages. We perform a comprehensive set of experiments on standard datasets to carefully study the contributions of each aspect of MatchNet, with direct comparisons to established methods. Our results confirm that our unified approach improves accuracy over previous state-of-the-art results on patch matching datasets, while reducing the storage requirement for descriptors. We make pre-trained MatchNet publicly available.",
"Learning effective feature representations and similarity measures are crucial to the retrieval performance of a content-based image retrieval (CBIR) system. Despite extensive research efforts for decades, it remains one of the most challenging open problems that considerably hinders the successes of real-world CBIR systems. The key challenge has been attributed to the well-known semantic gap'' issue that exists between low-level image pixels captured by machines and high-level semantic concepts perceived by human. Among various techniques, machine learning has been actively investigated as a possible direction to bridge the semantic gap in the long term. Inspired by recent successes of deep learning techniques for computer vision and other applications, in this paper, we attempt to address an open problem: if deep learning is a hope for bridging the semantic gap in CBIR and how much improvements in CBIR tasks can be achieved by exploring the state-of-the-art deep learning techniques for learning feature representations and similarity measures. Specifically, we investigate a framework of deep learning with application to CBIR tasks with an extensive set of empirical studies by examining a state-of-the-art deep learning method (Convolutional Neural Networks) for CBIR tasks under varied settings. From our empirical studies, we find some encouraging results and summarize some important insights for future research.",
"Person re-identification is to match pedestrian images from disjoint camera views detected by pedestrian detectors. Challenges are presented in the form of complex variations of lightings, poses, viewpoints, blurring effects, image resolutions, camera settings, occlusions and background clutter across camera views. In addition, misalignment introduced by the pedestrian detector will affect most existing person re-identification methods that use manually cropped pedestrian images and assume perfect detection. In this paper, we propose a novel filter pairing neural network (FPNN) to jointly handle misalignment, photometric and geometric transforms, occlusions and background clutter. All the key components are jointly optimized to maximize the strength of each component when cooperating with others. In contrast to existing works that use handcrafted features, our method automatically learns features optimal for the re-identification task from data. The learned filter pairs encode photometric transforms. Its deep architecture makes it possible to model a mixture of complex photometric and geometric transforms. We build the largest benchmark re-id dataset with 13, 164 images of 1, 360 pedestrians. Unlike existing datasets, which only provide manually cropped pedestrian images, our dataset provides automatically detected bounding boxes for evaluation close to practical applications. Our neural network significantly outperforms state-of-the-art methods on this dataset."
]
} |
1709.01353 | 2751825910 | Measuring visual (dis)similarity between two or more instances within a data distribution is a fundamental task in many applications, specially in image retrieval. Theoretically, non-metric distances are able to generate a more complex and accurate similarity model than metric distances, provided that the non-linear data distribution is precisely captured by the similarity model. In this work, we analyze a simple approach for deep learning networks to be used as an approximation of non-metric similarity functions and we study how these models generalize across different image retrieval datasets. | Neural networks have been previously proposed to model relationships between objects in different domains, such as classification in few-shot learning @cite_61 @cite_33 or visual question answering @cite_21 . The main difference between these methods and this work lies in the optimization problem to be solved: whereas @cite_61 @cite_33 @cite_62 aim to learn if a certain relation between a pair of images is occurring by optimizing a classification loss function (e.g. whether a query image is similar to a samples from a known class or not), we introduce a novel loss function specifically designed to solve ranking problems, which improves similarity scores on top of a standard metric by returning a regression value (i.e. how similar the two images are). | {
"cite_N": [
"@cite_61",
"@cite_21",
"@cite_33",
"@cite_62"
],
"mid": [
"2963341924",
"2768228940",
"2963907629",
""
],
"abstract": [
"Learning from a few examples remains a key challenge in machine learning. Despite recent advances in important domains such as vision and language, the standard supervised deep learning paradigm does not offer a satisfactory solution for learning new concepts rapidly from little data. In this work, we employ ideas from metric learning based on deep neural features and from recent advances that augment neural networks with external memories. Our framework learns a network that maps a small labelled support set and an unlabelled example to its label, obviating the need for fine-tuning to adapt to new class types. We then define one-shot learning problems on vision (using Omniglot, ImageNet) and language tasks. Our algorithm improves one-shot accuracy on ImageNet from 87.6 to 93.2 and from 88.0 to 93.8 on Omniglot compared to competing approaches. We also demonstrate the usefulness of the same model on language modeling by introducing a one-shot task on the Penn Treebank.",
"We present a conceptually simple, flexible, and general framework for few-shot learning, where a classifier must learn to recognise new classes given only few examples from each. Our method, called the Relation Network (RN), is trained end-to-end from scratch. During meta-learning, it learns to learn a deep distance metric to compare a small number of images within episodes, each of which is designed to simulate the few-shot setting. Once trained, a RN is able to classify images of new classes by computing relation scores between query images and the few examples of each new class without further updating the network. Besides providing improved performance on few-shot learning, our framework is easily extended to zero-shot learning. Extensive experiments on five benchmarks demonstrate that our simple approach provides a unified and effective approach for both of these two tasks.",
"Relational reasoning is a central component of generally intelligent behavior, but has proven difficult for neural networks to learn. In this paper we describe how to use Relation Networks (RNs) as a simple plug-and-play module to solve problems that fundamentally hinge on relational reasoning. We tested RN-augmented networks on three tasks: visual question answering using a challenging dataset called CLEVR, on which we achieve state-of-the-art, super-human performance; text-based question answering using the bAbI suite of tasks; and complex reasoning about dynamical physical systems. Then, using a curated dataset called Sort-of-CLEVR we show that powerful convolutional networks do not have a general capacity to solve relational questions, but can gain this capacity when augmented with RNs. Thus, by simply augmenting convolutions, LSTMs, and MLPs with RNs, we can remove computational burden from network components that are not well-suited to handle relational reasoning, reduce overall network complexity, and gain a general ability to reason about the relations between entities and their properties.",
""
]
} |
1709.01295 | 2753386433 | The ability to semantically interpret hand-drawn line sketches, although very challenging, can pave way for novel applications in multimedia. We propose SketchParse, the first deep-network architecture for fully automatic parsing of freehand object sketches. SketchParse is configured as a two-level fully convolutional network. The first level contains shared layers common to all object categories. The second level contains a number of expert sub-networks. Each expert specializes in parsing sketches from object categories which contain structurally similar parts. Effectively, the two-level configuration enables our architecture to scale up efficiently as additional categories are added. We introduce a router layer which (i) relays sketch features from shared layers to the correct expert (ii) eliminates the need to manually specify object category during inference. To bypass laborious part-level annotation, we sketchify photos from semantic object-part image datasets and use them for training. Our architecture also incorporates object pose prediction as a novel auxiliary task which boosts overall performance while providing supplementary information regarding the sketch. We demonstrate SketchParse's abilities (i) on two challenging large-scale sketch datasets (ii) in parsing unseen, semantically related object categories (iii) in improving fine-grained sketch-based image retrieval. As a novel application, we also outline how SketchParse's output can be used to generate caption-style descriptions for hand-drawn sketches. | Existing deep-learning approaches for semantic parsing of photos can be categorized into two groups. The first group consists of approaches for scene-level semantic parsing (i.e. output an object label for each pixel in the scene) @cite_44 @cite_6 @cite_43 . The second group of approaches attempt semantic parsing of objects (i.e. output a part label for each object pixel) @cite_1 @cite_25 @cite_3 . Compared to our choice (of a scene parsing net), this latter group of object-parsing approaches seemingly appear better candidates for the base architecture. However, they pose specific difficulties for adoption. For instance, the hypercolumns approach of @cite_1 requires training separate part classifiers for each class. Also, the evaluation is confined to a small number of classes (animals and human beings). The approach of @cite_25 consists of a complex multi-stage hybrid CNN-RNN architecture evaluated on only two animal categories ( horse and cow ). The approach of @cite_3 fuses object part score evidence from scene, object and part level to obtain impressive results for two categories (human, large animals). However, it is not clear how their method can be adopted for our purpose. | {
"cite_N": [
"@cite_1",
"@cite_6",
"@cite_3",
"@cite_44",
"@cite_43",
"@cite_25"
],
"mid": [
"1948751323",
"2952637581",
"2346977708",
"2412782625",
"2949295283",
""
],
"abstract": [
"Recognition algorithms based on convolutional networks (CNNs) typically use the output of the last layer as a feature representation. However, the information in this layer may be too coarse spatially to allow precise localization. On the contrary, earlier layers may be precise in localization but will not capture semantics. To get the best of both worlds, we define the hypercolumn at a pixel as the vector of activations of all CNN units above that pixel. Using hypercolumns as pixel descriptors, we show results on three fine-grained localization tasks: simultaneous detection and segmentation [22], where we improve state-of-the-art from 49.7 mean APr [22] to 60.0, keypoint localization, where we get a 3.3 point boost over [20], and part labeling, where we show a 6.6 point gain over a strong baseline.",
"We propose a novel semantic segmentation algorithm by learning a deconvolution network. We learn the network on top of the convolutional layers adopted from VGG 16-layer net. The deconvolution network is composed of deconvolution and unpooling layers, which identify pixel-wise class labels and predict segmentation masks. We apply the trained network to each proposal in an input image, and construct the final semantic segmentation map by combining the results from all proposals in a simple manner. The proposed algorithm mitigates the limitations of the existing methods based on fully convolutional networks by integrating deep deconvolution network and proposal-wise prediction; our segmentation method typically identifies detailed structures and handles objects in multiple scales naturally. Our network demonstrates outstanding performance in PASCAL VOC 2012 dataset, and we achieve the best accuracy (72.5 ) among the methods trained with no external data through ensemble with the fully convolutional network.",
"Parsing articulated objects, e.g. humans and animals, into semantic parts (e.g. body, head and arms, etc.) from natural images is a challenging and fundamental problem for computer vision. A big difficulty is the large variability of scale and location for objects and their corresponding parts. Even limited mistakes in estimating scale and location will degrade the parsing output and cause errors in boundary details. To tackle these difficulties, we propose a \"Hierarchical Auto-Zoom Net\" (HAZN) for object part parsing which adapts to the local scales of objects and parts. HAZN is a sequence of two \"Auto-Zoom Net\" (AZNs), each employing fully convolutional networks that perform two tasks: (1) predict the locations and scales of object instances (the first AZN) or their parts (the second AZN); (2) estimate the part scores for predicted object instance or part regions. Our model can adaptively \"zoom\" (resize) predicted image regions into their proper scales to refine the parsing. We conduct extensive experiments over the PASCAL part datasets on humans, horses, and cows. For humans, our approach significantly outperforms the state-of-the-arts by 5 mIOU and is especially better at segmenting small instances and small parts. We obtain similar improvements for parsing cows and horses over alternative methods. In summary, our strategy of first zooming into objects and then zooming into parts is very effective. It also enables us to process different regions of the image at different scales adaptively so that, for example, we do not need to waste computational resources scaling the entire image.",
"In this work we address the task of semantic image segmentation with Deep Learning and make three main contributions that are experimentally shown to have substantial practical merit. First , we highlight convolution with upsampled filters, or ‘atrous convolution’, as a powerful tool in dense prediction tasks. Atrous convolution allows us to explicitly control the resolution at which feature responses are computed within Deep Convolutional Neural Networks. It also allows us to effectively enlarge the field of view of filters to incorporate larger context without increasing the number of parameters or the amount of computation. Second , we propose atrous spatial pyramid pooling (ASPP) to robustly segment objects at multiple scales. ASPP probes an incoming convolutional feature layer with filters at multiple sampling rates and effective fields-of-views, thus capturing objects as well as image context at multiple scales. Third , we improve the localization of object boundaries by combining methods from DCNNs and probabilistic graphical models. The commonly deployed combination of max-pooling and downsampling in DCNNs achieves invariance but has a toll on localization accuracy. We overcome this by combining the responses at the final DCNN layer with a fully connected Conditional Random Field (CRF), which is shown both qualitatively and quantitatively to improve localization performance. Our proposed “DeepLab” system sets the new state-of-art at the PASCAL VOC-2012 semantic image segmentation task, reaching 79.7 percent mIOU in the test set, and advances the results on three other datasets: PASCAL-Context, PASCAL-Person-Part, and Cityscapes. All of our code is made publicly available online.",
"Semantic segmentation research has recently witnessed rapid progress, but many leading methods are unable to identify object instances. In this paper, we present Multi-task Network Cascades for instance-aware semantic segmentation. Our model consists of three networks, respectively differentiating instances, estimating masks, and categorizing objects. These networks form a cascaded structure, and are designed to share their convolutional features. We develop an algorithm for the nontrivial end-to-end training of this causal, cascaded structure. Our solution is a clean, single-step training framework and can be generalized to cascades that have more stages. We demonstrate state-of-the-art instance-aware semantic segmentation accuracy on PASCAL VOC. Meanwhile, our method takes only 360ms testing an image using VGG-16, which is two orders of magnitude faster than previous systems for this challenging problem. As a by product, our method also achieves compelling object detection results which surpass the competitive Fast Faster R-CNN systems. The method described in this paper is the foundation of our submissions to the MS COCO 2015 segmentation competition, where we won the 1st place.",
""
]
} |
1709.01295 | 2753386433 | The ability to semantically interpret hand-drawn line sketches, although very challenging, can pave way for novel applications in multimedia. We propose SketchParse, the first deep-network architecture for fully automatic parsing of freehand object sketches. SketchParse is configured as a two-level fully convolutional network. The first level contains shared layers common to all object categories. The second level contains a number of expert sub-networks. Each expert specializes in parsing sketches from object categories which contain structurally similar parts. Effectively, the two-level configuration enables our architecture to scale up efficiently as additional categories are added. We introduce a router layer which (i) relays sketch features from shared layers to the correct expert (ii) eliminates the need to manually specify object category during inference. To bypass laborious part-level annotation, we sketchify photos from semantic object-part image datasets and use them for training. Our architecture also incorporates object pose prediction as a novel auxiliary task which boosts overall performance while providing supplementary information regarding the sketch. We demonstrate SketchParse's abilities (i) on two challenging large-scale sketch datasets (ii) in parsing unseen, semantically related object categories (iii) in improving fine-grained sketch-based image retrieval. As a novel application, we also outline how SketchParse's output can be used to generate caption-style descriptions for hand-drawn sketches. | The initial performance of handcrafted feature-based approaches @cite_50 for sketch recognition has been surpassed in recent times by deep CNN architectures @cite_24 @cite_37 . The sketch router classifier in our architecture is a modified version of 's Sketch-a-Net @cite_24 . While the works mentioned above use sketches, @cite_27 use sketchified photos for training the CNN classifier which outputs class labels. We too use sketchified photos for training. However, the task in our case is parsing and not classification. | {
"cite_N": [
"@cite_24",
"@cite_27",
"@cite_37",
"@cite_50"
],
"mid": [
"1503798456",
"2549858052",
"2125635584",
""
],
"abstract": [
"We propose a multi-scale multi-channel deep neural network framework that, for the first time, yields sketch recognition performance surpassing that of humans. Our superior performance is a result of explicitly embedding the unique characteristics of sketches in our model: (i) a network architecture designed for sketch rather than natural photo statistics, (ii) a multi-channel generalisation that encodes sequential ordering in the sketching process, and (iii) a multi-scale network ensemble with joint Bayesian fusion that accounts for the different levels of abstraction exhibited in free-hand sketches. We show that state-of-the-art deep networks specifically engineered for photos of natural objects fail to perform well on sketch recognition, regardless whether they are trained using photo or sketch. Our network on the other hand not only delivers the best performance on the largest human sketch dataset to date, but also is small in size making efficient training possible using just CPUs.",
"The paper presents a deep Convolutional Neural Network (CNN) framework for free-hand sketch recognition. One of the main challenges in free-hand sketch recognition is to increase the recognition accuracy on sketches drawn by different people. To overcome this problem, we use deep Convolutional Neural Networks (CNNs) that have dominated top results in the field of image recognition. And we use the contours of natural images for training, because sketches drawn by different people may be very different and databases of the sketch images for training are very limited. We propose a CNN training on contours that performs well on sketch recognition over different databases of the sketch images. And we make some adjustments to the contours for training and reach higher recognition accuracy. Experimental results show the effectiveness of the proposed approach.",
"In this paper, we present a system for sketch classification and similarity search. We used deep convolution neural networks (ConvNets), state of the art in the field of image recognition. They enable both classification and medium highlevel features extraction. We make use of ConvNets features as a basis for similarity search using k-Nearest Neighbors (kNN). Evaluation are performed on the TU-Berlin benchmark. Our main contributions are threefold: first, we use ConvNets in contrast to most previous approaches based essentially on hand crafted features. Secondly, we propose a ConvNet that is both more accurate and lighter faster than the two only previous attempts at making use of ConvNets for handsketch recognition. We reached an accuracy of 75.42 . Third, we shown that similarly to their application on natural images, ConvNets allow the extraction of medium-level and high-level features (depending on the depth) which can be used for similarity search.1",
""
]
} |
1709.01295 | 2753386433 | The ability to semantically interpret hand-drawn line sketches, although very challenging, can pave way for novel applications in multimedia. We propose SketchParse, the first deep-network architecture for fully automatic parsing of freehand object sketches. SketchParse is configured as a two-level fully convolutional network. The first level contains shared layers common to all object categories. The second level contains a number of expert sub-networks. Each expert specializes in parsing sketches from object categories which contain structurally similar parts. Effectively, the two-level configuration enables our architecture to scale up efficiently as additional categories are added. We introduce a router layer which (i) relays sketch features from shared layers to the correct expert (ii) eliminates the need to manually specify object category during inference. To bypass laborious part-level annotation, we sketchify photos from semantic object-part image datasets and use them for training. Our architecture also incorporates object pose prediction as a novel auxiliary task which boosts overall performance while providing supplementary information regarding the sketch. We demonstrate SketchParse's abilities (i) on two challenging large-scale sketch datasets (ii) in parsing unseen, semantically related object categories (iii) in improving fine-grained sketch-based image retrieval. As a novel application, we also outline how SketchParse's output can be used to generate caption-style descriptions for hand-drawn sketches. | Our idea of having an initial coarse-category net which routes the input to finer-category experts can be found in some recent works as well @cite_32 @cite_0 , albeit for object classification. In these works, coarse-category net is intimately tied to the main task (viz. classification). In our case, the coarse-category CNN classifier serves a secondary role, helping to route the output of a parallel, shared sub-network to the finer-category parsing experts. Also, unlike above works, the task of our secondary net (classification) is different from the task of experts (segmentation). | {
"cite_N": [
"@cite_0",
"@cite_32"
],
"mid": [
"1937922215",
"2336829997"
],
"abstract": [
"In image classification, visual separability between different object categories is highly uneven, and some categories are more difficult to distinguish than others. Such difficult categories demand more dedicated classifiers. However, existing deep convolutional neural networks (CNN) are trained as flat N-way classifiers, and few efforts have been made to leverage the hierarchical structure of categories. In this paper, we introduce hierarchical deep CNNs (HD-CNNs) by embedding deep CNNs into a category hierarchy. An HD-CNN separates easy classes using a coarse category classifier while distinguishing difficult classes using fine category classifiers. During HD-CNN training, component-wise pretraining is followed by global finetuning with a multinomial logistic loss regularized by a coarse category consistency term. In addition, conditional executions of fine category classifiers and layer parameter compression make HD-CNNs scalable for large-scale visual recognition. We achieve state-of-the-art results on both CIFAR100 and large-scale ImageNet 1000-class benchmark datasets. In our experiments, we build up three different HD-CNNs and they lower the top-1 error of the standard CNNs by 2.65 , 3.1 and 1.1 , respectively.",
"We present a tree-structured network architecture for large-scale image classification. The trunk of the network contains convolutional layers optimized over all classes. At a given depth, the trunk splits into separate branches, each dedicated to discriminate a different subset of classes. Each branch acts as an expert classifying a set of categories that are difficult to tell apart, while the trunk provides common knowledge to all experts in the form of shared features. The training of our “network of experts” is completely end-to-end: the partition of categories into disjoint subsets is learned simultaneously with the parameters of the network trunk and the experts are trained jointly by minimizing a single learning objective over all classes. The proposed structure can be built from any existing convolutional neural network (CNN). We demonstrate its generality by adapting 4 popular CNNs for image categorization into the form of networks of experts. Our experiments on CIFAR100 and ImageNet show that in every case our method yields a substantial improvement in accuracy over the base CNN, and gives the best result achieved so far on CIFAR100. Finally, the improvement in accuracy comes at little additional cost: compared to the base network, the training time is only moderately increased and the number of parameters is comparable or in some cases even lower."
]
} |
1709.01295 | 2753386433 | The ability to semantically interpret hand-drawn line sketches, although very challenging, can pave way for novel applications in multimedia. We propose SketchParse, the first deep-network architecture for fully automatic parsing of freehand object sketches. SketchParse is configured as a two-level fully convolutional network. The first level contains shared layers common to all object categories. The second level contains a number of expert sub-networks. Each expert specializes in parsing sketches from object categories which contain structurally similar parts. Effectively, the two-level configuration enables our architecture to scale up efficiently as additional categories are added. We introduce a router layer which (i) relays sketch features from shared layers to the correct expert (ii) eliminates the need to manually specify object category during inference. To bypass laborious part-level annotation, we sketchify photos from semantic object-part image datasets and use them for training. Our architecture also incorporates object pose prediction as a novel auxiliary task which boosts overall performance while providing supplementary information regarding the sketch. We demonstrate SketchParse's abilities (i) on two challenging large-scale sketch datasets (ii) in parsing unseen, semantically related object categories (iii) in improving fine-grained sketch-based image retrieval. As a novel application, we also outline how SketchParse's output can be used to generate caption-style descriptions for hand-drawn sketches. | Our approach can be viewed as belonging to the category of domain adaptation techniques @cite_21 @cite_49 @cite_48 . These techniques have proven to be successful for various problems, including image parsing @cite_10 @cite_35 @cite_51 . However, unlike most approaches wherein image modality does not change, our domain-adaptation scenario is characterized by extreme modality-level variation between source (image) and target (freehand sketch). This drastically reduces the quantity and quality of data available for transfer learning, making our task more challenging. | {
"cite_N": [
"@cite_35",
"@cite_48",
"@cite_21",
"@cite_49",
"@cite_51",
"@cite_10"
],
"mid": [
"2257483379",
"2107250100",
"1722318740",
"1982696459",
"2094286293",
"2431874326"
],
"abstract": [
"We propose a novel weakly-supervised semantic segmentation algorithm based on Deep Convolutional Neural Network (DCNN). Contrary to existing weakly-supervised approaches, our algorithm exploits auxiliary segmentation annotations available for different categories to guide segmentations on images with only image-level class labels. To make segmentation knowledge transferrable across categories, we design a decoupled encoder-decoder architecture with attention model. In this architecture, the model generates spatial highlights of each category presented in images using an attention model, and subsequently performs binary segmentation for each highlighted region using decoder. Combining attention model, the decoder trained with segmentation annotations in different categories boosts accuracy of weakly-supervised semantic segmentation. The proposed algorithm demonstrates substantially improved performance compared to the state-of-theart weakly-supervised techniques in PASCAL VOC 2012 dataset when our model is trained with the annotations in 60 exclusive categories in Microsoft COCO dataset.",
"Most current image categorization methods require large collections of manually annotated training examples to learn accurate visual recognition models. The time-consuming human labeling effort effectively limits these approaches to recognition problems involving a small number of different object classes. In order to address this shortcoming, in recent years several authors have proposed to learn object classifiers from weakly-labeled Internet images, such as photos retrieved by keyword-based image search engines. While this strategy eliminates the need for human supervision, the recognition accuracies of these methods are considerably lower than those obtained with fully-supervised approaches, because of the noisy nature of the labels associated to Web data. In this paper we investigate and compare methods that learn image classifiers by combining very few manually annotated examples (e.g., 1-10 images per class) and a large number of weakly-labeled Web photos retrieved using keyword-based image search. We cast this as a domain adaptation problem: given a few strongly-labeled examples in a target domain (the manually annotated examples) and many source domain examples (the weakly-labeled Web photos), learn classifiers yielding small generalization error on the target domain. Our experiments demonstrate that, for the same number of strongly-labeled examples, our domain adaptation approach produces significant recognition rate improvements over the best published results (e.g., 65 better when using 5 labeled training examples per class) and that our classifiers are one order of magnitude faster to learn and to evaluate than the best competing method, despite our use of large weakly-labeled data sets.",
"Domain adaptation is an important emerging topic in computer vision. In this paper, we present one of the first studies of domain shift in the context of object recognition. We introduce a method that adapts object models acquired in a particular visual domain to new imaging conditions by learning a transformation that minimizes the effect of domain-induced changes in the feature distribution. The transformation is learned in a supervised manner and can be applied to categories for which there are no labeled examples in the new domain. While we focus our evaluation on object recognition tasks, the transform-based adaptation technique we develop is general and could be applied to nonimage data. Another contribution is a new multi-domain object database, freely available for download. We experimentally demonstrate the ability of our method to improve recognition on categories with few or no target domain labels and moderate to large changes in the imaging conditions.",
"In pattern recognition and computer vision, one is often faced with scenarios where the training data used to learn a model have different distribution from the data on which the model is applied. Regardless of the cause, any distributional change that occurs after learning a classifier can degrade its performance at test time. Domain adaptation tries to mitigate this degradation. In this article, we provide a survey of domain adaptation methods for visual recognition. We discuss the merits and drawbacks of existing domain adaptation approaches and identify promising avenues for research in this rapidly evolving field.",
"The variation between images obtained with different scanners or different imaging protocols presents a major challenge in automatic segmentation of biomedical images. This variation especially hampers the application of otherwise successful supervised-learning techniques which, in order to perform well, often require a large amount of labeled training data that is exactly representative of the target data. We therefore propose to use transfer learning for image segmentation. Transfer-learning techniques can cope with differences in distributions between training and target data, and therefore may improve performance over supervised learning for segmentation across scanners and scan protocols. We present four transfer classifiers that can train a classification scheme with only a small amount of representative training data, in addition to a larger amount of other training data with slightly different characteristics. The performance of the four transfer classifiers was compared to that of standard supervised classification on two magnetic resonance imaging brain-segmentation tasks with multi-site data: white matter, gray matter, and cerebrospinal fluid segmentation; and white-matter- MS-lesion segmentation. The experiments showed that when there is only a small amount of representative training data available, transfer learning can greatly outperform common supervised-learning approaches, minimizing classification errors by up to 60 .",
"Vision-based semantic segmentation in urban scenarios is a key functionality for autonomous driving. Recent revolutionary results of deep convolutional neural networks (DCNNs) foreshadow the advent of reliable classifiers to perform such visual tasks. However, DCNNs require learning of many parameters from raw images, thus, having a sufficient amount of diverse images with class annotations is needed. These annotations are obtained via cumbersome, human labour which is particularly challenging for semantic segmentation since pixel-level annotations are required. In this paper, we propose to use a virtual world to automatically generate realistic synthetic images with pixel-level annotations. Then, we address the question of how useful such data can be for semantic segmentation – in particular, when using a DCNN paradigm. In order to answer this question we have generated a synthetic collection of diverse urban images, named SYNTHIA, with automatically generated class annotations. We use SYNTHIA in combination with publicly available real-world urban images with manually provided annotations. Then, we conduct experiments with DCNNs that show how the inclusion of SYNTHIA in the training stage significantly improves performance on the semantic segmentation task."
]
} |
1709.01295 | 2753386433 | The ability to semantically interpret hand-drawn line sketches, although very challenging, can pave way for novel applications in multimedia. We propose SketchParse, the first deep-network architecture for fully automatic parsing of freehand object sketches. SketchParse is configured as a two-level fully convolutional network. The first level contains shared layers common to all object categories. The second level contains a number of expert sub-networks. Each expert specializes in parsing sketches from object categories which contain structurally similar parts. Effectively, the two-level configuration enables our architecture to scale up efficiently as additional categories are added. We introduce a router layer which (i) relays sketch features from shared layers to the correct expert (ii) eliminates the need to manually specify object category during inference. To bypass laborious part-level annotation, we sketchify photos from semantic object-part image datasets and use them for training. Our architecture also incorporates object pose prediction as a novel auxiliary task which boosts overall performance while providing supplementary information regarding the sketch. We demonstrate SketchParse's abilities (i) on two challenging large-scale sketch datasets (ii) in parsing unseen, semantically related object categories (iii) in improving fine-grained sketch-based image retrieval. As a novel application, we also outline how SketchParse's output can be used to generate caption-style descriptions for hand-drawn sketches. | The effectiveness of addressing auxiliary tasks in tandem with the main task has been shown for several challenging problems in vision @cite_41 @cite_2 @cite_15 @cite_16 . In particular, object classification @cite_40 , detection @cite_43 , geometric context @cite_33 , saliency @cite_26 and adversarial loss @cite_11 have been utilized as auxiliary tasks in deep network-based approaches for semantic parsing. The auxiliary task we employ -- object viewpoint estimation -- has been used in a multi-task setting but for object classification @cite_14 @cite_22 . For instance, @cite_18 project the output of semantic scene segmentation onto a depth map and use a category-aware 3D model approach to enable 3-D pose-based grasping in robots. Zhao and Itti @cite_14 utilize 3-D pose information from toy models of @math categories as a target auxiliary task to improve object classification. @cite_22 also introduce pose-estimation as an auxiliary task along with classification. To the best of our knowledge, we are the first ones to design a custom pose estimation architecture to assist semantic parsing. | {
"cite_N": [
"@cite_18",
"@cite_26",
"@cite_14",
"@cite_33",
"@cite_22",
"@cite_41",
"@cite_43",
"@cite_40",
"@cite_2",
"@cite_15",
"@cite_16",
"@cite_11"
],
"mid": [
"2593617942",
"2147347517",
"",
"2029731618",
"2433982775",
"2963377935",
"2949295283",
"2963691928",
"1907729166",
"",
"2139191027",
"2554423077"
],
"abstract": [
"Recent robotic manipulation competitions have highlighted that sophisticated robots still struggle to achieve fast and reliable perception of task-relevant objects in complex, realistic scenarios. To improve these systems' perceptive speed and robustness, we present SegICP, a novel integrated solution to object recognition and pose estimation. SegICP couples convolutional neural networks and multi-hypothesis point cloud registration to achieve both robust pixel-wise semantic segmentation as well as accurate and real-time 6-DOF pose estimation for relevant objects. Our architecture achieves 1 cm position error and < 5° angle error in real time without an initial seed. We evaluate and benchmark SegICP against an annotated dataset generated by motion capture.",
"A key problem in salient object detection is how to effectively model the semantic properties of salient objects in a data-driven manner. In this paper, we propose a multi-task deep saliency model based on a fully convolutional neural network with global input (whole raw images) and global output (whole saliency maps). In principle, the proposed saliency model takes a data-driven strategy for encoding the underlying saliency prior information, and then sets up a multi-task learning scheme for exploring the intrinsic correlations between saliency detection and semantic image segmentation. Through collaborative feature learning from such two correlated tasks, the shared fully convolutional layers produce effective features for object perception. Moreover, it is capable of capturing the semantic information on salient objects across different levels using the fully convolutional layers, which investigate the feature-sharing properties of salient object detection with a great reduction of feature redundancy. Finally, we present a graph Laplacian regularized nonlinear regression model for saliency refinement. Experimental results demonstrate the effectiveness of our approach in comparison with the state-of-the-art approaches.",
"",
"We address the task of learning a semantic segmentation from weakly supervised data. Our aim is to devise a system that predicts an object label for each pixel by making use of only image level labels during training – the information whether a certain object is present or not in the image. Such coarse tagging of images is faster and easier to obtain as opposed to the tedious task of pixelwise labeling required in state of the art systems. We cast this task naturally as a multiple instance learning (MIL) problem. We use Semantic Texton Forest (STF) as the basic framework and extend it for the MIL setting. We make use of multitask learning (MTL) to regularize our solution. Here, an external task of geometric context estimation is used to improve on the task of semantic segmentation. We report experimental results on the MSRC21 and the very challenging VOC2007 datasets. On MSRC21 dataset we are able, by using 276 weakly labeled images, to achieve the performance of a supervised STF trained on pixelwise labeled training set of 56 images, which is a significant reduction in supervision needed.",
"In the Object Recognition task, there exists a dichotomy between the categorization of objects and estimating object pose, where the former necessitates a view-invariant representation, while the latter requires a representation capable of capturing pose information over different categories of objects. With the rise of deep architectures, the prime focus has been on object category recognition. Deep learning methods have achieved wide success in this task. In contrast, object pose estimation using these approaches has received relatively less attention. In this work, we study how Convolutional Neural Networks (CNN) architectures can be adapted to the task of simultaneous object recognition and pose estimation. We investigate and analyze the layers of various CNN models and extensively compare between them with the goal of discovering how the layers of distributed representations within CNNs represent object pose information and how this contradicts with object category representations. We extensively experiment on two recent large and challenging multi-view datasets and we achieve better than the state-of-the-art.",
"We present an algorithm for simultaneous face detection, landmarks localization, pose estimation and gender recognition using deep convolutional neural networks (CNN). The proposed method called, HyperFace, fuses the intermediate layers of a deep CNN using a separate CNN followed by a multi-task learning algorithm that operates on the fused features. It exploits the synergy among the tasks which boosts up their individual performances. Additionally, we propose two variants of HyperFace: (1) HyperFace-ResNet that builds on the ResNet-101 model and achieves significant improvement in performance, and (2) Fast-HyperFace that uses a high recall fast face detector for generating region proposals to improve the speed of the algorithm. Extensive experiments show that the proposed models are able to capture both global and local information in faces and performs significantly better than many competitive algorithms for each of these four tasks.",
"Semantic segmentation research has recently witnessed rapid progress, but many leading methods are unable to identify object instances. In this paper, we present Multi-task Network Cascades for instance-aware semantic segmentation. Our model consists of three networks, respectively differentiating instances, estimating masks, and categorizing objects. These networks form a cascaded structure, and are designed to share their convolutional features. We develop an algorithm for the nontrivial end-to-end training of this causal, cascaded structure. Our solution is a clean, single-step training framework and can be generalized to cascades that have more stages. We demonstrate state-of-the-art instance-aware semantic segmentation accuracy on PASCAL VOC. Meanwhile, our method takes only 360ms testing an image using VGG-16, which is two orders of magnitude faster than previous systems for this challenging problem. As a by product, our method also achieves compelling object detection results which surpass the competitive Fast Faster R-CNN systems. The method described in this paper is the foundation of our submissions to the MS COCO 2015 segmentation competition, where we won the 1st place.",
"",
"This paper proposes a joint multi-task learning algorithm to better predict attributes in images using deep convolutional neural networks (CNN). We consider learning binary semantic attributes through a multi-task CNN model, where each CNN will predict one binary attribute. The multi-task learning allows CNN models to simultaneously share visual knowledge among different attribute categories. Each CNN will generate attribute-specific feature representations, and then we apply multi-task learning on the features to predict their attributes. In our multi-task framework, we propose a method to decompose the overall model’s parameters into a latent task matrix and combination matrix. Furthermore, under-sampled classifiers can leverage shared statistics from other classifiers to improve their performance. Natural grouping of attributes is applied such that attributes in the same group are encouraged to share more knowledge. Meanwhile, attributes in different groups will generally compete with each other, and consequently share less knowledge. We show the effectiveness of our method on two popular attribute datasets.",
"",
"This paper presents an approach to view-invariant action recognition, where human poses and motions exhibit large variations across different camera viewpoints. When each viewpoint of a given set of action classes is specified as a learning task then multitask learning appears suitable for achieving view invariance in recognition. We extend the standard multitask learning to allow identifying: (1) latent groupings of action views (i.e., tasks), and (2) discriminative action parts, along with joint learning of all tasks. This is because it seems reasonable to expect that certain distinct views are more correlated than some others, and thus identifying correlated views could improve recognition. Also, part-based modeling is expected to improve robustness against self-occlusion when actors are imaged from different views. Results on the benchmark datasets show that we outperform standard multitask learning by 21.9 , and the state-of-the-art alternatives by 4.5-6 .",
"Adversarial training has been shown to produce state of the art results for generative image modeling. In this paper we propose an adversarial training approach to train semantic segmentation models. We train a convolutional semantic segmentation network along with an adversarial network that discriminates segmentation maps coming either from the ground truth or from the segmentation network. The motivation for our approach is that it can detect and correct higher-order inconsistencies between ground truth segmentation maps and the ones produced by the segmentation net. Our experiments show that our adversarial training approach leads to improved accuracy on the Stanford Background and PASCAL VOC 2012 datasets."
]
} |
1709.01041 | 2752514874 | Deep Neural Networks trained on large datasets can be easily transferred to new domains with far fewer labeled examples by a process called fine-tuning. This has the advantage that representations learned in the large source domain can be exploited on smaller target domains. However, networks designed to be optimal for the source task are often prohibitively large for the target task. In this work we address the compression of networks after domain transfer. We focus on compression algorithms based on low-rank matrix decomposition. Existing methods base compression solely on learned network weights and ignore the statistics of network activations. We show that domain transfer leads to large shifts in network activations and that it is desirable to take this into account when compressing. We demonstrate that considering activation statistics when compressing weights leads to a rank-constrained regression problem with a closed-form solution. Because our method takes into account the target domain, it can more optimally remove the redundancy in the weights. Experiments show that our Domain Adaptive Low Rank (DALR) method significantly outperforms existing low-rank compression techniques. With our approach, the fc6 layer of VGG19 can be compressed more than 4x more than using truncated SVD alone -- with only a minor or no loss in accuracy. When applied to domain-transferred networks it allows for compression down to only 5-20 of the original number of parameters with only a minor drop in performance. | Network pruning. A straight forward way to reduce the memory footprint of a neural network is by removing unimportant parameters. This process can be conducted while training @cite_5 @cite_3 @cite_18 @cite_32 , or by analyzing the influence of each parameter once the network has been trained @cite_23 . For instance, in @cite_25 , the authors use tensor low rank constraints to (iteratively) reduce the number of parameters of the fully connected layer. | {
"cite_N": [
"@cite_18",
"@cite_32",
"@cite_3",
"@cite_23",
"@cite_5",
"@cite_25"
],
"mid": [
"2156150815",
"2114766824",
"",
"1935978687",
"2951134251",
"2520760693"
],
"abstract": [
"The use of information from all second-order derivatives of the error function to perform network pruning (i.e., removing unimportant weights from a trained network) in order to improve generalization, simplify networks, reduce hardware or storage requirements, increase the speed of further training, and, in some cases, enable rule extraction, is investigated. The method, Optimal Brain Surgeon (OBS), is significantly better than magnitude-based methods and Optimal Brain Damage, which often remove the wrong weights. OBS, permits pruning of more weights than other methods (for the same error on the training set), and thus yields better generalization on test data. Crucial to OBS is a recursion relation for calculating the inverse Hessian matrix H sup -1 from training data and structural information of the set. OBS deletes the correct weights from a trained XOR network in every case. >",
"We have used information-theoretic ideas to derive a class of practical and nearly optimal schemes for adapting the size of a neural network. By removing unimportant weights from a network, several improvements can be expected: better generalization, fewer training examples required, and improved speed of learning and or classification. The basic idea is to use second-derivative information to make a tradeoff between network complexity and training set error. Experiments confirm the usefulness of the methods on a real-world application.",
"",
"Deep neural networks have achieved remarkable performance in both image classification and object detection problems, at the cost of a large number of parameters and computational complexity. In this work, we show how to reduce the redundancy in these parameters using a sparse decomposition. Maximum sparsity is obtained by exploiting both inter-channel and intra-channel redundancy, with a fine-tuning step that minimize the recognition loss caused by maximizing sparsity. This procedure zeros out more than 90 of parameters, with a drop of accuracy that is less than 1 on the ILSVRC2012 dataset. We also propose an efficient sparse matrix multiplication algorithm on CPU for Sparse Convolutional Neural Networks (SCNN) models. Our CPU implementation demonstrates much higher efficiency than the off-the-shelf sparse matrix libraries, with a significant speedup realized over the original dense network. In addition, we apply the SCNN model to the object detection problem, in conjunction with a cascade model and sparse fully connected layers, to achieve significant speedups.",
"Nowadays, the number of layers and of neurons in each layer of a deep network are typically set manually. While very deep and wide networks have proven effective in general, they come at a high memory and computation cost, thus making them impractical for constrained platforms. These networks, however, are known to have many redundant parameters, and could thus, in principle, be replaced by more compact architectures. In this paper, we introduce an approach to automatically determining the number of neurons in each layer of a deep network during learning. To this end, we propose to make use of structured sparsity during learning. More precisely, we use a group sparsity regularizer on the parameters of the network, where each group is defined to act on a single neuron. Starting from an overcomplete network, we show that our approach can reduce the number of parameters by up to 80 while retaining or even improving the network accuracy.",
"To attain a favorable performance on large-scale datasets, convolutional neural networks (CNNs) are usually designed to have very high capacity involving millions of parameters. In this work, we aim at optimizing the number of neurons in a network, thus the number of parameters. We show that, by incorporating sparse constraints into the objective function, it is possible to decimate the number of neurons during the training stage. As a result, the number of parameters and the memory footprint of the neural network are also reduced, which is also desirable at the test time. We evaluated our method on several well-known CNN structures including AlexNet, and VGG over different datasets including ImageNet. Extensive experimental results demonstrate that our method leads to compact networks. Taking first fully connected layer as an example, our compact CNN contains only (30 , ) of the original neurons without any degradation of the top-1 classification accuracy."
]
} |
1709.01041 | 2752514874 | Deep Neural Networks trained on large datasets can be easily transferred to new domains with far fewer labeled examples by a process called fine-tuning. This has the advantage that representations learned in the large source domain can be exploited on smaller target domains. However, networks designed to be optimal for the source task are often prohibitively large for the target task. In this work we address the compression of networks after domain transfer. We focus on compression algorithms based on low-rank matrix decomposition. Existing methods base compression solely on learned network weights and ignore the statistics of network activations. We show that domain transfer leads to large shifts in network activations and that it is desirable to take this into account when compressing. We demonstrate that considering activation statistics when compressing weights leads to a rank-constrained regression problem with a closed-form solution. Because our method takes into account the target domain, it can more optimally remove the redundancy in the weights. Experiments show that our Domain Adaptive Low Rank (DALR) method significantly outperforms existing low-rank compression techniques. With our approach, the fc6 layer of VGG19 can be compressed more than 4x more than using truncated SVD alone -- with only a minor or no loss in accuracy. When applied to domain-transferred networks it allows for compression down to only 5-20 of the original number of parameters with only a minor drop in performance. | Computationally efficient layer representations. Several approaches have addressed the problem of reducing computational resources by modifying the internal representation of each layer taking into account the inherent redundancy of its parameters. Common approaches exploit linear structures within convolutional layers and approach each convolutional kernel using low-rank kernels @cite_34 @cite_29 @cite_28 . The main idea relies on the fact that performing a convolution with the original kernels is equivalent to convolving with a set of base filters followed by a linear combination of their output. @cite_12 , the authors propose two network layers that are based on dictionary learning to perform sparse representation learning, directly within the network. In general, these approaches show significant reduction in the computational needs with a minimum drop of performance. | {
"cite_N": [
"@cite_28",
"@cite_29",
"@cite_34",
"@cite_12"
],
"mid": [
"2422848620",
"1996901117",
"2167215970",
"2585683604"
],
"abstract": [
"Deep learning and convolutional neural networks (ConvNets) have been successfully applied to most relevant tasks in the computer vision community. However, these networks are computationally demanding and not suitable for embedded devices where memory and time consumption are relevant. In this paper, we propose DecomposeMe, a simple but effective technique to learn features using 1D convolutions. The proposed architecture enables both simplicity and filter sharing leading to increased learning capacity. A comprehensive set of large-scale experiments on ImageNet and Places2 demonstrates the ability of our method to improve performance while significantly reducing the number of parameters required. Notably, on Places2, we obtain an improvement in relative top-1 classification accuracy of 7.7 with an architecture that requires 92 fewer parameters compared to VGG-B. The proposed network is also demonstrated to generalize to other tasks by converting existing networks.",
"The focus of this paper is speeding up the application 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 [15], showing a possible 2.5× speedup with no loss in accuracy, and 4.5× speedup with less than 1 drop in accuracy, still achieving state-of-the-art on standard benchmarks.",
"We present techniques for speeding up the test-time evaluation of large convolutional networks, designed for object recognition tasks. These models deliver impressive accuracy, but each image evaluation requires millions of floating point operations, making their deployment on smartphones and Internet-scale clusters problematic. The computation is dominated by the convolution operations in the lower layers of the model. We exploit the redundancy present within the convolutional filters to derive approximations that significantly reduce the required computation. Using large state-of-the-art models, we demonstrate speedups of convolutional layers on both CPU and GPU by a factor of 2 x, while keeping the accuracy within 1 of the original model.",
"Whereas CNNs have demonstrated immense progress in many vision problems, they suffer from a dependence on monumental amounts of labeled training data. On the other hand, dictionary learning does not scale to the size of problems that CNNs can handle, despite being very effective at low-level vision tasks such as denoising and inpainting. Recently, interest has grown in adapting dictionary learning methods for supervised tasks such as classification and inverse problems. We propose two new network layers that are based on dictionary learning: a sparse factorization layer and a convolutional sparse factorization layer, analogous to fully-connected and convolutional layers, respectively. Using our derivations, these layers can be dropped in to existing CNNs, trained together in an end-to-end fashion with back-propagation, and leverage semisupervision in ways classical CNNs cannot. We experimentally compare networks with these two new layers against a baseline CNN. Our results demonstrate that networks with either of the sparse factorization layers are able to outperform classical CNNs when supervised data are few. They also show performance improvements in certain tasks when compared to the CNN with no sparse factorization layers with the same exact number of parameters."
]
} |
1709.01041 | 2752514874 | Deep Neural Networks trained on large datasets can be easily transferred to new domains with far fewer labeled examples by a process called fine-tuning. This has the advantage that representations learned in the large source domain can be exploited on smaller target domains. However, networks designed to be optimal for the source task are often prohibitively large for the target task. In this work we address the compression of networks after domain transfer. We focus on compression algorithms based on low-rank matrix decomposition. Existing methods base compression solely on learned network weights and ignore the statistics of network activations. We show that domain transfer leads to large shifts in network activations and that it is desirable to take this into account when compressing. We demonstrate that considering activation statistics when compressing weights leads to a rank-constrained regression problem with a closed-form solution. Because our method takes into account the target domain, it can more optimally remove the redundancy in the weights. Experiments show that our Domain Adaptive Low Rank (DALR) method significantly outperforms existing low-rank compression techniques. With our approach, the fc6 layer of VGG19 can be compressed more than 4x more than using truncated SVD alone -- with only a minor or no loss in accuracy. When applied to domain-transferred networks it allows for compression down to only 5-20 of the original number of parameters with only a minor drop in performance. | Parameter quantization. Previous works mentioned above on efficient representations focus on modifying the representation of a complete layer. Parameter quantization is slightly different as it aims at finding efficient representations of each parameter individually (ie, representing each parameter with fewer bits). A common practice to minimize the memory footprint of the network and reduce the computational cost during inference consists of training using @math bit to represent the parameters while performing inference more efficiently using 16-bits without significantly affecting the performance. More aggressive quantization processes have been also analyzed in @cite_41 where the authors propose an approach to directly thresholding values at 0 resulting in a decrease of the top-1 performance on ImageNet by less tan 10 @cite_42 the authors propose an approach to combine pruning, quantization and coding to reduce the computational complexity of the network. The authors first analyze the relevance of each parameter pruning the irrelevant ones and then, after fine-tuning the pruned network, the remaining parameters are quantized. Results show a significant reduction in the number of parameters (up to 35x) without affecting the performance of the network. | {
"cite_N": [
"@cite_41",
"@cite_42"
],
"mid": [
"1724438581",
"2119144962"
],
"abstract": [
"Deep convolutional neural networks (CNN) has become the most promising method for object recognition, repeatedly demonstrating record breaking results for image classification and object detection in recent years. However, a very deep CNN generally involves many layers with millions of parameters, making the storage of the network model to be extremely large. This prohibits the usage of deep CNNs on resource limited hardware, especially cell phones or other embedded devices. In this paper, we tackle this model storage issue by investigating information theoretical vector quantization methods for compressing the parameters of CNNs. In particular, we have found in terms of compressing the most storage demanding dense connected layers, vector quantization methods have a clear gain over existing matrix factorization methods. Simply applying k-means clustering to the weights or conducting product quantization can lead to a very good balance between model size and recognition accuracy. For the 1000-category classification task in the ImageNet challenge, we are able to achieve 16-24 times compression of the network with only 1 loss of classification accuracy using the state-of-the-art CNN.",
"Neural networks are both computationally intensive and memory intensive, making them difficult to deploy on embedded systems with limited hardware resources. To address this limitation, we introduce \"deep compression\", a three stage pipeline: pruning, trained quantization and Huffman coding, that work together to reduce the storage requirement of neural networks by 35x to 49x without affecting their accuracy. Our method first prunes the network by learning only the important connections. Next, we quantize the weights to enforce weight sharing, finally, we apply Huffman coding. After the first two steps we retrain the network to fine tune the remaining connections and the quantized centroids. Pruning, reduces the number of connections by 9x to 13x; Quantization then reduces the number of bits that represent each connection from 32 to 5. On the ImageNet dataset, our method reduced the storage required by AlexNet by 35x, from 240MB to 6.9MB, without loss of accuracy. Our method reduced the size of VGG-16 by 49x from 552MB to 11.3MB, again with no loss of accuracy. This allows fitting the model into on-chip SRAM cache rather than off-chip DRAM memory. Our compression method also facilitates the use of complex neural networks in mobile applications where application size and download bandwidth are constrained. Benchmarked on CPU, GPU and mobile GPU, compressed network has 3x to 4x layerwise speedup and 3x to 7x better energy efficiency."
]
} |
1709.01041 | 2752514874 | Deep Neural Networks trained on large datasets can be easily transferred to new domains with far fewer labeled examples by a process called fine-tuning. This has the advantage that representations learned in the large source domain can be exploited on smaller target domains. However, networks designed to be optimal for the source task are often prohibitively large for the target task. In this work we address the compression of networks after domain transfer. We focus on compression algorithms based on low-rank matrix decomposition. Existing methods base compression solely on learned network weights and ignore the statistics of network activations. We show that domain transfer leads to large shifts in network activations and that it is desirable to take this into account when compressing. We demonstrate that considering activation statistics when compressing weights leads to a rank-constrained regression problem with a closed-form solution. Because our method takes into account the target domain, it can more optimally remove the redundancy in the weights. Experiments show that our Domain Adaptive Low Rank (DALR) method significantly outperforms existing low-rank compression techniques. With our approach, the fc6 layer of VGG19 can be compressed more than 4x more than using truncated SVD alone -- with only a minor or no loss in accuracy. When applied to domain-transferred networks it allows for compression down to only 5-20 of the original number of parameters with only a minor drop in performance. | Network distillation. These approaches aim at mimicking a complicated model using a simpler one. The key idea consists of training an ensemble of large networks and then use their combined output to train a simpler model @cite_21 . Several approaches have built on this idea to train the network based on the soft output of the ensemble @cite_0 , or to train the network mimicking the behavior not only of the last layer but also of intermediate ones @cite_10 . | {
"cite_N": [
"@cite_0",
"@cite_21",
"@cite_10"
],
"mid": [
"1821462560",
"",
"1690739335"
],
"abstract": [
"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.",
"",
"While depth tends to improve network performances, it also makes gradient-based training more difficult since deeper networks tend to be more non-linear. The recently proposed knowledge distillation approach is aimed at obtaining small and fast-to-execute models, and it has shown that a student network could imitate the soft output of a larger teacher network or ensemble of networks. In this paper, we extend this idea to allow the training of a student that is deeper and thinner than the teacher, using not only the outputs but also the intermediate representations learned by the teacher as hints to improve the training process and final performance of the student. Because the student intermediate hidden layer will generally be smaller than the teacher's intermediate hidden layer, additional parameters are introduced to map the student hidden layer to the prediction of the teacher hidden layer. This allows one to train deeper students that can generalize better or run faster, a trade-off that is controlled by the chosen student capacity. For example, on CIFAR-10, a deep student network with almost 10.4 times less parameters outperforms a larger, state-of-the-art teacher network."
]
} |
1709.00928 | 2751719873 | Mobile applications are being used every day by more than half of the world's population to perform a great variety of tasks. With the increasingly widespread usage of these applications, the need arises for efficient techniques to test them. Many frameworks allow automating the process of application testing, however existing frameworks mainly rely on the application developer for providing testing scripts for each developed application, thus preventing reuse of these tests for similar applications. In this paper, we present a novel approach for the automation of testing Android applications by leveraging machine learning techniques and reusing popular test scenarios. We discuss and demonstrate the potential benefits of our approach in an empirical study where we show that our developed testing tool, based on the proposed approach, outperforms standard methods in realistic settings. | Producing a tool that will allow testing any arbitrary mobile application automatically is an extremely challenging problem, perhaps nearly as difficult as the underlying general software testing automation task. Throughout the years, there has been an extensive study in the field of testing automation of . However, a recent study by @cite_8 has discovered that many of the mobile application bugs are unique and tend to be different from the ones presented in traditional desktop applications, mainly due to the inherent difference in architecture and development methodologies. Thus, traditional approaches for desktop applications testing cannot be naturally translated into mobile applications. | {
"cite_N": [
"@cite_8"
],
"mid": [
"2004921952"
],
"abstract": [
"Users increasingly rely on mobile applications for computational needs. Google Android is a popular mobile platform, hence the reliability of Android applications is becoming increasingly important. Many Android correctness issues, however, fall outside the scope of traditional verification techniques, as they are due to the novelty of the platform and its GUI-oriented application construction paradigm. In this paper we present an approach for automating the testing process for Android applications, with a focus on GUI bugs. We first conduct a bug mining study to understand the nature and frequency of bugs affecting Android applications; our study finds that GUI bugs are quite numerous. Next, we present techniques for detecting GUI bugs by automatic generation of test cases, feeding the application random events, instrumenting the VM, producing log trace files and analyzing them post-run. We show how these techniques helped to re-discover existing bugs and find new bugs, and how they could be used to prevent certain bug categories. We believe our study and techniques have the potential to help developers increase the quality of Android applications."
]
} |
1709.00928 | 2751719873 | Mobile applications are being used every day by more than half of the world's population to perform a great variety of tasks. With the increasingly widespread usage of these applications, the need arises for efficient techniques to test them. Many frameworks allow automating the process of application testing, however existing frameworks mainly rely on the application developer for providing testing scripts for each developed application, thus preventing reuse of these tests for similar applications. In this paper, we present a novel approach for the automation of testing Android applications by leveraging machine learning techniques and reusing popular test scenarios. We discuss and demonstrate the potential benefits of our approach in an empirical study where we show that our developed testing tool, based on the proposed approach, outperforms standard methods in realistic settings. | The need for efficient mobile application testing methods has yielded many testing automation frameworks, including tools like Appium http: appium.io , Selendroid http: selendroid.io and Robotium https: github.com RobotiumTech robotium , to name a few (a recent survey is available at @cite_20 ). These frameworks allow a developer to write testing scripts in her programming language of choice, and later run these scripts over and over again to check the application in different user behavior scenarios. The main limitation of such tools is that these manually constructed scripts are coded for a designated application in mind, therefore the developer is bound to invest ample time in reusing these tests for new applications and to accommodate for changes in the functionality of existing applications. | {
"cite_N": [
"@cite_20"
],
"mid": [
"2016422503"
],
"abstract": [
"To cope with frequent upgrades of mobile devices and technologies, engineers need a reusable and cost-effective environment for testing mobile applications and an elastic infrastructure to support large-scale test automation."
]
} |
1709.00928 | 2751719873 | Mobile applications are being used every day by more than half of the world's population to perform a great variety of tasks. With the increasingly widespread usage of these applications, the need arises for efficient techniques to test them. Many frameworks allow automating the process of application testing, however existing frameworks mainly rely on the application developer for providing testing scripts for each developed application, thus preventing reuse of these tests for similar applications. In this paper, we present a novel approach for the automation of testing Android applications by leveraging machine learning techniques and reusing popular test scenarios. We discuss and demonstrate the potential benefits of our approach in an empirical study where we show that our developed testing tool, based on the proposed approach, outperforms standard methods in realistic settings. | In order to reduce the need for writing redundant testing scripts for mobile applications, which share many common characteristics, a great amount of research efforts have lately focused on the development of automated testing techniques and algorithms to allow for testing applications automatically. In this paper, as with most of recent papers in the field, we focus on the Android platform @cite_3 . The choice to use the Android platform is mainly due to the fact that it is the most common mobile operation system on the market to date and due to its open-source nature that allows the academic community to get full access to the applications and the platform source code. Moreover, the large variety of Android models and versions on the market make the test automation task significantly important. | {
"cite_N": [
"@cite_3"
],
"mid": [
"2950387995"
],
"abstract": [
"Mobile applications, often simply called \"apps\", are increasingly widespread, and we use them daily to perform a number of activities. Like all software, apps must be adequately tested to gain confidence that they behave correctly. Therefore, in recent years, researchers and practitioners alike have begun to investigate ways to automate apps testing. In particular, because of Android's open source nature and its large share of the market, a great deal of research has been performed on input generation techniques for apps that run on the Android operating systems. At this point in time, there are in fact a number of such techniques in the literature, which differ in the way they generate inputs, the strategy they use to explore the behavior of the app under test, and the specific heuristics they use. To better understand the strengths and weaknesses of these existing approaches, and get general insight on ways they could be made more effective, in this paper we perform a thorough comparison of the main existing test input generation tools for Android. In our comparison, we evaluate the effectiveness of these tools, and their corresponding techniques, according to four metrics: code coverage, ability to detect faults, ability to work on multiple platforms, and ease of use. Our results provide a clear picture of the state of the art in input generation for Android apps and identify future research directions that, if suitably investigated, could lead to more effective and efficient testing tools for Android."
]
} |
1709.00928 | 2751719873 | Mobile applications are being used every day by more than half of the world's population to perform a great variety of tasks. With the increasingly widespread usage of these applications, the need arises for efficient techniques to test them. Many frameworks allow automating the process of application testing, however existing frameworks mainly rely on the application developer for providing testing scripts for each developed application, thus preventing reuse of these tests for similar applications. In this paper, we present a novel approach for the automation of testing Android applications by leveraging machine learning techniques and reusing popular test scenarios. We discuss and demonstrate the potential benefits of our approach in an empirical study where we show that our developed testing tool, based on the proposed approach, outperforms standard methods in realistic settings. | All of the above approaches have been successfully deployed in different experiments to have shown to produce a good coverage of the applications' state-space. Nevertheless, they all share a common prominent limitation: these approaches aim to find only bugs and defects in the application, meaning real-time application crashes which are caused by uncaught exceptions thrown in the code @cite_3 . However, many of the bugs presented in today's applications are related to the program logic (e.g., a login screen that can be by-passed without entering valid username and password, an email-composing screen that allows sending emails to an invalid email address, etc). @cite_8 present an empirical study of common Android bugs. The authors find that logical bugs are about @math times more prevalent than technical bugs. | {
"cite_N": [
"@cite_3",
"@cite_8"
],
"mid": [
"2950387995",
"2004921952"
],
"abstract": [
"Mobile applications, often simply called \"apps\", are increasingly widespread, and we use them daily to perform a number of activities. Like all software, apps must be adequately tested to gain confidence that they behave correctly. Therefore, in recent years, researchers and practitioners alike have begun to investigate ways to automate apps testing. In particular, because of Android's open source nature and its large share of the market, a great deal of research has been performed on input generation techniques for apps that run on the Android operating systems. At this point in time, there are in fact a number of such techniques in the literature, which differ in the way they generate inputs, the strategy they use to explore the behavior of the app under test, and the specific heuristics they use. To better understand the strengths and weaknesses of these existing approaches, and get general insight on ways they could be made more effective, in this paper we perform a thorough comparison of the main existing test input generation tools for Android. In our comparison, we evaluate the effectiveness of these tools, and their corresponding techniques, according to four metrics: code coverage, ability to detect faults, ability to work on multiple platforms, and ease of use. Our results provide a clear picture of the state of the art in input generation for Android apps and identify future research directions that, if suitably investigated, could lead to more effective and efficient testing tools for Android.",
"Users increasingly rely on mobile applications for computational needs. Google Android is a popular mobile platform, hence the reliability of Android applications is becoming increasingly important. Many Android correctness issues, however, fall outside the scope of traditional verification techniques, as they are due to the novelty of the platform and its GUI-oriented application construction paradigm. In this paper we present an approach for automating the testing process for Android applications, with a focus on GUI bugs. We first conduct a bug mining study to understand the nature and frequency of bugs affecting Android applications; our study finds that GUI bugs are quite numerous. Next, we present techniques for detecting GUI bugs by automatic generation of test cases, feeding the application random events, instrumenting the VM, producing log trace files and analyzing them post-run. We show how these techniques helped to re-discover existing bugs and find new bugs, and how they could be used to prevent certain bug categories. We believe our study and techniques have the potential to help developers increase the quality of Android applications."
]
} |
1709.00928 | 2751719873 | Mobile applications are being used every day by more than half of the world's population to perform a great variety of tasks. With the increasingly widespread usage of these applications, the need arises for efficient techniques to test them. Many frameworks allow automating the process of application testing, however existing frameworks mainly rely on the application developer for providing testing scripts for each developed application, thus preventing reuse of these tests for similar applications. In this paper, we present a novel approach for the automation of testing Android applications by leveraging machine learning techniques and reusing popular test scenarios. We discuss and demonstrate the potential benefits of our approach in an empirical study where we show that our developed testing tool, based on the proposed approach, outperforms standard methods in realistic settings. | Another significant limitation of the above three approaches is that, in general, they will not be able to reach screens of the application that need a specific input in order to advance. Specifically, the above approaches will probably get stuck when reaching a login screen, being unable to cover a (potentially) significant part of the application. @cite_3 points out this problem as a future research direction, saying that allowing tools to explore the application in presence of login forms and similar complex inputs, which may be hard to generate randomly or by means of systematic techniques, will help explore new behaviors. In this work, we suggest to overcome this limitation by using machine learning techniques that enable our new approach to test new, previously unexplored, logical conditions in the screen using our pre-defined set of expected behaviors, as we will describe later on. Moreover, using very limited inputs supplied by the programmer, previously impassable screens can now be passed by feeding these inputs at the right time. Thus, using our approach, one can test applications more comprehensively, finding new logical bugs and reaching new states in the application. | {
"cite_N": [
"@cite_3"
],
"mid": [
"2950387995"
],
"abstract": [
"Mobile applications, often simply called \"apps\", are increasingly widespread, and we use them daily to perform a number of activities. Like all software, apps must be adequately tested to gain confidence that they behave correctly. Therefore, in recent years, researchers and practitioners alike have begun to investigate ways to automate apps testing. In particular, because of Android's open source nature and its large share of the market, a great deal of research has been performed on input generation techniques for apps that run on the Android operating systems. At this point in time, there are in fact a number of such techniques in the literature, which differ in the way they generate inputs, the strategy they use to explore the behavior of the app under test, and the specific heuristics they use. To better understand the strengths and weaknesses of these existing approaches, and get general insight on ways they could be made more effective, in this paper we perform a thorough comparison of the main existing test input generation tools for Android. In our comparison, we evaluate the effectiveness of these tools, and their corresponding techniques, according to four metrics: code coverage, ability to detect faults, ability to work on multiple platforms, and ease of use. Our results provide a clear picture of the state of the art in input generation for Android apps and identify future research directions that, if suitably investigated, could lead to more effective and efficient testing tools for Android."
]
} |
1709.00930 | 2751625733 | Exiting deep-learning based dense stereo matching methods often rely on ground-truth disparity maps as the training signals, which are however not always available in many situations. In this paper, we design a simple convolutional neural network architecture that is able to learn to compute dense disparity maps directly from the stereo inputs. Training is performed in an end-to-end fashion without the need of ground-truth disparity maps. The idea is to use image warping error (instead of disparity-map residuals) as the loss function to drive the learning process, aiming to find a depth-map that minimizes the warping error. While this is a simple concept well-known in stereo matching, to make it work in a deep-learning framework, many non-trivial challenges must be overcome, and in this work we provide effective solutions. Our network is self-adaptive to different unseen imageries as well as to different camera settings. Experiments on KITTI and Middlebury stereo benchmark datasets show that our method outperforms many state-of-the-art stereo matching methods with a margin, and at the same time significantly faster. | Estimating a dense depth disparity map from a stereo image pair is a long lasting problem that has been studied for decades. Interested readers are referred to @cite_10 , @cite_19 and @cite_5 for overviews. In this section, we provide a brief discussion on related works. | {
"cite_N": [
"@cite_19",
"@cite_5",
"@cite_10"
],
"mid": [
"2208592848",
"2609532991",
""
],
"abstract": [
"This paper presents a literature survey on existing disparity map algorithms. It focuses on four main stages of processing as proposed by Scharstein and Szeliski in a taxonomy and evaluation of dense two-frame stereo correspondence algorithms performed in 2002. To assist future researchers in developing their own stereo matching algorithms, a summary of the existing algorithms developed for every stage of processing is also provided. The survey also notes the implementation of previous software-based and hardware-based algorithms. Generally, the main processing module for a software-based implementation uses only a central processing unit. By contrast, a hardware-based implementation requires one or more additional processors for its processing module, such as graphical processing unit or a field programmable gate array. This literature survey also presents a method of qualitative measurement that is widely used by researchers in the area of stereo vision disparity mappings.",
"Recent years have witnessed amazing progress in AI related fields such as computer vision, machine learning and autonomous vehicles. As with any rapidly growing field, however, it becomes increasingly difficult to stay up-to-date or enter the field as a beginner. While several topic specific survey papers have been written, to date no general survey on problems, datasets and methods in computer vision for autonomous vehicles exists. This paper attempts to narrow this gap by providing a state-of-the-art survey on this topic. Our survey includes both the historically most relevant literature as well as the current state-of-the-art on several specific topics, including recognition, reconstruction, motion estimation, tracking, scene understanding and end-to-end learning. Towards this goal, we first provide a taxonomy to classify each approach and then analyze the performance of the state-of-the-art on several challenging benchmarking datasets including KITTI, ISPRS, MOT and Cityscapes. Besides, we discuss open problems and current research challenges. To ease accessibility and accommodate missing references, we will also provide an interactive platform which allows to navigate topics and methods, and provides additional information and project links for each paper.",
""
]
} |
1709.00893 | 2740899359 | Aspect-level sentiment classification aims at identifying the sentiment polarity of specific target in its context. Previous approaches have realized the importance of targets in sentiment classification and developed various methods with the goal of precisely modeling their contexts via generating target-specific representations. However, these studies always ignore the separate modeling of targets. In this paper, we argue that both targets and contexts deserve special treatment and need to be learned their own representations via interactive learning. Then, we propose the interactive attention networks (IAN) to interactively learn attentions in the contexts and targets, and generate the representations for targets and contexts separately. With this design, the IAN model can well represent a target and its collocative context, which is helpful to sentiment classification. Experimental results on SemEval 2014 Datasets demonstrate the effectiveness of our model. | Recently, kinds of neural networks methods, such as Recursive Neural networks @cite_22 @cite_13 @cite_19 , Recursive Neural Tensor Networks @cite_20 , Recurrent Neural networks @cite_1 @cite_7 , Tree-LSTMs @cite_5 and Hierarchical LSTMs @cite_3 , have achieved a promising result on sentiment analysis. However, the neural network based approaches just make use of the contexts without consideration of targets which also make great contributions to judging the sentiment polarity of target. | {
"cite_N": [
"@cite_22",
"@cite_7",
"@cite_1",
"@cite_3",
"@cite_19",
"@cite_5",
"@cite_13",
"@cite_20"
],
"mid": [
"71795751",
"",
"",
"2518806256",
"2252128625",
"2104246439",
"",
"2251939518"
],
"abstract": [
"We introduce a novel machine learning framework based on recursive autoencoders for sentence-level prediction of sentiment label distributions. Our method learns vector space representations for multi-word phrases. In sentiment prediction tasks these representations outperform other state-of-the-art approaches on commonly used datasets, such as movie reviews, without using any pre-defined sentiment lexica or polarity shifting rules. We also evaluate the model's ability to predict sentiment distributions on a new dataset based on confessions from the experience project. The dataset consists of personal user stories annotated with multiple labels which, when aggregated, form a multinomial distribution that captures emotional reactions. Our algorithm can more accurately predict distributions over such labels compared to several competitive baselines.",
"",
"",
"Opinion mining from customer reviews has become pervasive in recent years. Sentences in reviews, however, are usually classified independently, even though they form part of a review's argumentative structure. Intuitively, sentences in a review build and elaborate upon each other; knowledge of the review structure and sentential context should thus inform the classification of each sentence. We demonstrate this hypothesis for the task of aspect-based sentiment analysis by modeling the interdependencies of sentences in a review with a hierarchical bidirectional LSTM. We show that the hierarchical model outperforms two non-hierarchical baselines, obtains results competitive with the state-of-the-art, and outperforms the state-of-the-art on five multilingual, multi-domain datasets without any hand-engineered features or external resources.",
"Recursive neural network is one of the most successful deep learning models for natural language processing due to the compositional nature of text. The model recursively composes the vector of a parent phrase from those of child words or phrases, with a key component named composition function. Although a variety of composition functions have been proposed, the syntactic information has not been fully encoded in the composition process. We propose two models, Tag Guided RNN (TGRNN for short) which chooses a composition function according to the part-ofspeech tag of a phrase, and Tag Embedded RNN RNTN (TE-RNN RNTN for short) which learns tag embeddings and then combines tag and word embeddings together. In the fine-grained sentiment classification, experiment results show the proposed models obtain remarkable improvement: TG-RNN TE-RNN obtain remarkable improvement over baselines, TE-RNTN obtains the second best result among all the top performing models, and all the proposed models have much less parameters complexity than their counterparts.",
"Because of their superior ability to preserve sequence information over time, Long Short-Term Memory (LSTM) networks, a type of recurrent neural network with a more complex computational unit, have obtained strong results on a variety of sequence modeling tasks. The only underlying LSTM structure that has been explored so far is a linear chain. However, natural language exhibits syntactic properties that would naturally combine words to phrases. We introduce the Tree-LSTM, a generalization of LSTMs to tree-structured network topologies. Tree-LSTMs outperform all existing systems and strong LSTM baselines on two tasks: predicting the semantic relatedness of two sentences (SemEval 2014, Task 1) and sentiment classification (Stanford Sentiment Treebank).",
"",
"Semantic word spaces have been very useful but cannot express the meaning of longer phrases in a principled way. Further progress towards understanding compositionality in tasks such as sentiment detection requires richer supervised training and evaluation resources and more powerful models of composition. To remedy this, we introduce a Sentiment Treebank. It includes fine grained sentiment labels for 215,154 phrases in the parse trees of 11,855 sentences and presents new challenges for sentiment compositionality. To address them, we introduce the Recursive Neural Tensor Network. When trained on the new treebank, this model outperforms all previous methods on several metrics. It pushes the state of the art in single sentence positive negative classification from 80 up to 85.4 . The accuracy of predicting fine-grained sentiment labels for all phrases reaches 80.7 , an improvement of 9.7 over bag of features baselines. Lastly, it is the only model that can accurately capture the effects of negation and its scope at various tree levels for both positive and negative phrases."
]
} |
1709.00867 | 2752795045 | A source traffic model for machine-to-machine communications is presented in this paper. We consider a model in which devices operate in a regular mode until they are triggered into an alarm mode by an alarm event. The positions of devices and events are modeled by means of Poisson point processes, where the generated traffic by a given device depends on its position and event positions. We first consider the case where devices and events are static and devices generate traffic according to a Bernoulli process, where we derive the total rate from the devices at the base station. We then extend the model by defining a two-state Markov chain for each device, which allows for devices to stay in alarm mode for a geometrically distributed holding time. The temporal characteristics of this model are analyzed via the autocovariance function, where the effect of event density and mean holding time are shown. | Models for M2M traffic can broadly be categorized into two groups, source models and aggregated models @cite_6 . In a source model, each source has its own set of tunable parameters which control its individual traffic characteristics. In an aggregated traffic model all sources have a common set of parameters. A source model thus allows for more fine-grained control; however this comes at a higher cost in computational complexity because of a higher number of parameters. | {
"cite_N": [
"@cite_6"
],
"mid": [
"2963436339"
],
"abstract": [
"The Internet of Things (IoT) is large scale by nature, which is manifested by the massive number of connected devices as well as their vast spatial existence. Cellular networks, which provide ubiquitous, reliable, and efficient wireless access, will play fundamental rule in delivering the first-mile access for the data tsunami to be generated by the IoT. However, cellular networks may have scalability problems to provide uplink connectivity to massive numbers of connected things. To characterize the scalability of cellular uplink in the context of IoT networks, this paper develops a traffic-aware spatiotemporal mathematical model for IoT devices supported by cellular uplink connectivity. The developed model is based on stochastic geometry and queueing theory to account for the traffic requirement per IoT device, the different transmission strategies, and the mutual interference between the IoT devices. To this end, the developed model is utilized to characterize the extent to which cellular networks can accommodate IoT traffic as well as to assess and compare three different transmission strategies that incorporate a combination of transmission persistency, backoff, and power-ramping. The analysis and the results clearly illustrate the scalability problem imposed by IoT on cellular network and offer insights into effective scenarios for each transmission strategy."
]
} |
1709.00653 | 2751486157 | One key challenge in talent search is to translate complex criteria of a hiring position into a search query, while it is relatively easy for a searcher to list examples of suitable candidates for a given position. To improve search e ciency, we propose the next generation of talent search at LinkedIn, also referred to as Search By Ideal Candidates. In this system, a searcher provides one or several ideal candidates as the input to hire for a given position. The system then generates a query based on the ideal candidates and uses it to retrieve and rank results. Shifting from the traditional Query-By-Keyword to this new Query-By-Example system poses a number of challenges: How to generate a query that best describes the candidates? When moving to a completely di erent paradigm, how does one leverage previous product logs to learn ranking models and or evaluate the new system with no existing usage logs? Finally, given the di erent nature between the two search paradigms, the ranking features typically used for Query-By-Keyword systems might not be optimal for Query- By-Example. This paper describes our approach to solving these challenges. We present experimental results con rming the e ectiveness of the proposed solution, particularly on query building and search ranking tasks. As of writing this paper, the new system has been available to all LinkedIn members. | Query-By-Example ( QBE ) in image retrieval (also referred as content-based image retrieval), especially in dermatology studies, originates from the seminal work in @cite_13 @cite_27 in 1980s and 1990s. QBE in image retrieval systems typically utilizes the example images containing objects of interest to generate visual features which include color and texture for detecting and identifying the particular objects easily. More recently, there has been a focus on Query-By-Example in text domain @cite_17 @cite_9 . Text based QBE systems extract key phrases from unstructured query documents by using techniques like tf-idf and also combine them with semantic matching, such as, LSI and LDA. | {
"cite_N": [
"@cite_9",
"@cite_27",
"@cite_13",
"@cite_17"
],
"mid": [
"",
"2101498401",
"2066771327",
"2111593814"
],
"abstract": [
"",
"Content-based image retrieval (CBIR) has become one of the most active research areas in the past few years. Many visual feature representations have been explored and many systems built. While these research efforts establish the basis of CBIR, the usefulness of the proposed approaches is limited. Specifically, these efforts have relatively ignored two distinct characteristics of CBIR systems: (1) the gap between high-level concepts and low-level features, and (2) the subjectivity of human perception of visual content. This paper proposes a relevance feedback based interactive retrieval approach, which effectively takes into account the above two characteristics in CBIR. During the retrieval process, the user's high-level query and perception subjectivity are captured by dynamically updated weights based on the user's feedback. The experimental results over more than 70000 images show that the proposed approach greatly reduces the user's effort of composing a query, and captures the user's information need more precisely.",
"We present an approach for picture indexing and abstraction. Picture indexing facilitates information retrieval from a pictorial database consisting of picture objects and picture relations. To construct picture indexes, abstraction operations to perform picture object clustering and classification are formulated. To substantiate the abstraction operations, we also formalize syntactic abstraction rules and semantic abstraction rules. We then illustrate by examples how to apply these abstraction operations to obtain various picture indexes, and how to construct icons to facilitate accessing of pictorial data.",
"We are experiencing an unprecedented increase of content contributed by users in forums such as blogs, social networking sites and microblogging services. Such abundance of content complements content on web sites and traditional media forums such as news papers, news and financial streams, and so on. Given such plethora of information there is a pressing need to cross reference information across textual services. For example, commonly we read a news item and we wonder if there are any blogs reporting related content or vice versa. In this paper, we present techniques to automate the process of cross referencing online information content. We introduce methodologies to extract phrases from a given \"query document\" to be used as queries to search interfaces with the goal to retrieve content related to the query document. In particular, we consider two techniques to extract and score key phrases. We also consider techniques to complement extracted phrases with information present in external sources such as Wikipedia and introduce an algorithm called RelevanceRank for this purpose. We discuss both these techniques in detail and provide an experimental study utilizing a large number of human judges from Amazons's Mechanical Turk service. Detailed experiments demonstrate the effectiveness and efficiency of the proposed techniques for the task of automating retrieval of documents related to a query document."
]
} |
1709.00653 | 2751486157 | One key challenge in talent search is to translate complex criteria of a hiring position into a search query, while it is relatively easy for a searcher to list examples of suitable candidates for a given position. To improve search e ciency, we propose the next generation of talent search at LinkedIn, also referred to as Search By Ideal Candidates. In this system, a searcher provides one or several ideal candidates as the input to hire for a given position. The system then generates a query based on the ideal candidates and uses it to retrieve and rank results. Shifting from the traditional Query-By-Keyword to this new Query-By-Example system poses a number of challenges: How to generate a query that best describes the candidates? When moving to a completely di erent paradigm, how does one leverage previous product logs to learn ranking models and or evaluate the new system with no existing usage logs? Finally, given the di erent nature between the two search paradigms, the ranking features typically used for Query-By-Keyword systems might not be optimal for Query- By-Example. This paper describes our approach to solving these challenges. We present experimental results con rming the e ectiveness of the proposed solution, particularly on query building and search ranking tasks. As of writing this paper, the new system has been available to all LinkedIn members. | Since the seminal work of @cite_25 and Balabanovi ' c et al @cite_4 , item-based collaborative filtering has been widely used to identify similarity between recommended items. Later efforts include, but are not limited to @cite_5 @cite_20 @cite_2 @cite_8 . @cite_2 provide probabilistic frameworks which make the collaborative filtering more robust when dealing with sparse data. @cite_5 combine the content-based approach (Section ) and collaborative filtering together so that the whole approach could handle the cold-start problem. In @cite_8 , Koren combines collaborative filtering with latent factors to utilize both the neighborhood information and latent factors. In our own previous work @cite_14 , we use such a technique to define career-path similarity for similar people recommendation system on LinkedIn. For a comprehensive coverage of the topic, please refer to the survey @cite_10 . | {
"cite_N": [
"@cite_14",
"@cite_4",
"@cite_8",
"@cite_2",
"@cite_5",
"@cite_10",
"@cite_25",
"@cite_20"
],
"mid": [
"2048587746",
"2043403353",
"1994389483",
"2117311203",
"2158515176",
"2100235918",
"2042281163",
"2159094788"
],
"abstract": [
"For decades large corporations as well as labor placement services have maintained extensive yet static resume databanks. Online professional networks like LinkedIn have taken these resume databanks to a dynamic, constantly updated and massive scale professional profile dataset spanning career records from hundreds of industries, millions of companies and hundreds of millions of people worldwide. Using this professional profile dataset, this paper attempts to model profiles of individuals as a sequence of positions held by them as a time-series of nodes, each of which represents one particular position or job experience in the individual's career trajectory. These career trajectory models can be employed in various utility applications including career trajectory planning for students in schools & universities using knowledge inferred from real world career outcomes. They can also be employed for decoding sequences to uncover paths leading to certain professional milestones from a user's current professional status. We deploy the proposed technique to ascertain professional similarity between two individuals by developing a similarity measure SimCareers (Similar Career Paths). The measure employs sequence alignment between two career trajectories to quantify professional similarity between career paths. To the best of our knowledge, SimCareers is the first framework to model professional similarity between two people taking account their career trajectory information. We posit, that using the temporal and structural features of a career trajectory for modeling profile similarity is a far more superior approach than using similarity measures on semi-structured attribute representation of a profile for this application. We validate our hypothesis by extensive quantitative evaluations on a gold dataset of similar profiles generated from recruiting activity logs from actual recruiters using LinkedIn. In addition, we show significant improvements in engagement by running an A B test on a real-world application called Similar Profiles on LinkedIn, world's largest online professional network.",
"",
"Recommender systems provide users with personalized suggestions for products or services. These systems often rely on Collaborating Filtering (CF), where past transactions are analyzed in order to establish connections between users and products. The two more successful approaches to CF are latent factor models, which directly profile both users and products, and neighborhood models, which analyze similarities between products or users. In this work we introduce some innovations to both approaches. The factor and neighborhood models can now be smoothly merged, thereby building a more accurate combined model. Further accuracy improvements are achieved by extending the models to exploit both explicit and implicit feedback by the users. The methods are tested on the Netflix data. Results are better than those previously published on that dataset. In addition, we suggest a new evaluation metric, which highlights the differences among methods, based on their performance at a top-K recommendation task.",
"Memory-based methods for collaborative filtering predict new ratings by averaging (weighted) ratings between, respectively, pairs of similar users or items. In practice, a large number of ratings from similar users or similar items are not available, due to the sparsity inherent to rating data. Consequently, prediction quality can be poor. This paper re-formulates the memory-based collaborative filtering problem in a generative probabilistic framework, treating individual user-item ratings as predictors of missing ratings. The final rating is estimated by fusing predictions from three sources: predictions based on ratings of the same item by other users, predictions based on different item ratings made by the same user, and, third, ratings predicted based on data from other but similar users rating other but similar items. Existing user-based and item-based approaches correspond to the two simple cases of our framework. The complete model is however more robust to data sparsity, because the different types of ratings are used in concert, while additional ratings from similar users towards similar items are employed as a background model to smooth the predictions. Experiments demonstrate that the proposed methods are indeed more robust against data sparsity and give better recommendations.",
"We have developed a method for recommending items that combines content and collaborative data under a single probabilistic framework. We benchmark our algorithm against a naive Bayes classifier on the cold-start problem, where we wish to recommend items that no one in the community has yet rated. We systematically explore three testing methodologies using a publicly available data set, and explain how these methods apply to specific real-world applications. We advocate heuristic recommenders when benchmarking to give competent baseline performance. We introduce a new performance metric, the CROC curve, and demonstrate empirically that the various components of our testing strategy combine to obtain deeper understanding of the performance characteristics of recommender systems. Though the emphasis of our testing is on cold-start recommending, our methods for recommending and evaluation are general.",
"As one of the most successful approaches to building recommender systems, collaborative filtering (CF) uses the known preferences of a group of users to make recommendations or predictions of the unknown preferences for other users. In this paper, we first introduce CF tasks and their main challenges, such as data sparsity, scalability, synonymy, gray sheep, shilling attacks, privacy protection, etc., and their possible solutions. We then present three main categories of CF techniques: memory-based, modelbased, and hybrid CF algorithms (that combine CF with other recommendation techniques), with examples for representative algorithms of each category, and analysis of their predictive performance and their ability to address the challenges. From basic techniques to the state-of-the-art, we attempt to present a comprehensive survey for CF techniques, which can be served as a roadmap for research and practice in this area.",
"Recommender systems apply knowledge discovery techniques to the problem of making personalized recommendations for information, products or services during a live interaction. These systems, especially the k-nearest neighbor collaborative ltering based ones, are achieving widespread success on the Web. The tremendous growth in the amount of available information and the number of visitors to Web sites in recent years poses some key challenges for recommender systems. These are: producing high quality recommendations, performing many recommendations per second for millions of users and items and achieving high coverage in the face of data sparsity. In traditional collaborative ltering systems the amount of work increases with the number of participants in the system. New recommender system technologies are needed that can quickly produce high quality recommendations, even for very large-scale problems. To address these issues we have explored item-based collaborative ltering techniques. Item-based techniques rst analyze the user-item matrix to identify relationships between di erent items, and then use these relationships to indirectly compute recommendations for users. In this paper we analyze di erent item-based recommendation generation algorithms. We look into di erent techniques for computing item-item similarities (e.g., item-item correlation vs. cosine similarities between item vectors) and di erent techniques for obtaining recommendations from them (e.g., weighted sum vs. regression model). Finally, we experimentally evaluate our results and compare them to the basic k-nearest neighbor approach. Our experiments suggest that item-based algorithms provide dramatically better performance than user-based algorithms, while at the same time providing better quality than the best available userbased algorithms.",
"Recommendation algorithms are best known for their use on e-commerce Web sites, where they use input about a customer's interests to generate a list of recommended items. Many applications use only the items that customers purchase and explicitly rate to represent their interests, but they can also use other attributes, including items viewed, demographic data, subject interests, and favorite artists. At Amazon.com, we use recommendation algorithms to personalize the online store for each customer. The store radically changes based on customer interests, showing programming titles to a software engineer and baby toys to a new mother. There are three common approaches to solving the recommendation problem: traditional collaborative filtering, cluster models, and search-based methods. Here, we compare these methods with our algorithm, which we call item-to-item collaborative filtering. Unlike traditional collaborative filtering, our algorithm's online computation scales independently of the number of customers and number of items in the product catalog. Our algorithm produces recommendations in real-time, scales to massive data sets, and generates high quality recommendations."
]
} |
1709.00653 | 2751486157 | One key challenge in talent search is to translate complex criteria of a hiring position into a search query, while it is relatively easy for a searcher to list examples of suitable candidates for a given position. To improve search e ciency, we propose the next generation of talent search at LinkedIn, also referred to as Search By Ideal Candidates. In this system, a searcher provides one or several ideal candidates as the input to hire for a given position. The system then generates a query based on the ideal candidates and uses it to retrieve and rank results. Shifting from the traditional Query-By-Keyword to this new Query-By-Example system poses a number of challenges: How to generate a query that best describes the candidates? When moving to a completely di erent paradigm, how does one leverage previous product logs to learn ranking models and or evaluate the new system with no existing usage logs? Finally, given the di erent nature between the two search paradigms, the ranking features typically used for Query-By-Keyword systems might not be optimal for Query- By-Example. This paper describes our approach to solving these challenges. We present experimental results con rming the e ectiveness of the proposed solution, particularly on query building and search ranking tasks. As of writing this paper, the new system has been available to all LinkedIn members. | Relevance feedback is a popular approach used in text retrieval. It leverages searcher's feedback on an initial set of retrieval results to refine the original query. The query refinement could be in the form of re-weighting the query terms or automatically expanding the query with new terms. Rocchio @cite_6 is widely considered to be the first formalization of relevance feedback technique, developed on the vector space model. He proposes query refinement based on the difference between the average vector of the relevant documents and the average vector of the non-relevant documents. A subsequent approach uses probabilistic models that estimate the probability a document is relevant to the information need @cite_23 @cite_1 , where the probabilistic models are inferred from the feedback documents. | {
"cite_N": [
"@cite_1",
"@cite_23",
"@cite_6"
],
"mid": [
"2169213601",
"2043909051",
"2164547069"
],
"abstract": [
"We explore the relation between classical probabilistic models of information retrieval and the emerging language modeling approaches. It has long been recognized that the primary obstacle to effective performance of classical models is the need to estimate a relevance model : probabilities of words in the relevant class. We propose a novel technique for estimating these probabilities using the query alone. We demonstrate that our technique can produce highly accurate relevance models, addressing important notions of synonymy and polysemy. Our experiments show relevance models outperforming baseline language modeling systems on TREC retrieval and TDT tracking tasks. The main contribution of this work is an effective formal method for estimating a relevance model with no training data.",
"This paper examines statistical techniques for exploiting relevance information to weight search terms. These techniques are presented as a natural extension of weighting methods using information about the distribution of index terms in documents in general. A series of relevance weighting functions is derived and is justified by theoretical considerations. In particular, it is shown that specific weighted search methods are implied by a general probabilistic theory of retrieval. Different applications of relevance weighting are illustrated by experimental results for test collections.",
"1332840 Primer compositions DOW CORNINGCORP 6 Oct 1971 [30 Dec 1970] 46462 71 Heading C3T [Also in Divisions B2 and C4] A primer composition comprises 1 pbw of tetra ethoxy or propoxy silane or poly ethyl or propyl silicate or any mixture thereof, 0A75-2A5 pbw of bis(acetylacetonyl) diisopropyl titanate, 0A75- 5 pbw of a compound CF 3 CH 2 CH 2 Si[OSi(CH 3 ) 2 - X] 3 wherein each X is H or -CH 2 CH 2 Si- (OOCCH 3 ) 3 , at least one being the latter, and 1-20 pbw of a ketone, hydrocarbon or halohydrocarbon solvent boiling not above 150‹ C. In the examples 1 pbw each of bis(acetylacetonyl)diisopropyl titanate, polyethyl silicate and are dissolved in 10 pbw of acetone or in 9 pbw of light naphtha and 1 of methylisobutylketone. The solutions are used to prime Ti panels, to which a Pt-catalysed room-temperature vulcanizable poly-trifluoropropylmethyl siloxanebased rubber is then applied."
]
} |
1709.00740 | 2753681123 | Distances on symbolic musical sequences are needed for a variety of applications, from music retrieval to automatic music generation. These musical sequences belong to a given corpus (or style) and it is obvious that a good distance on musical sequences should take this information into account; being able to define a distance ex nihilo which could be applicable to all music styles seems implausible. A distance could also be invariant under some transformations, such as transpositions, so that it can be used as a distance between musical motives rather than musical sequences. However, to our knowledge, none of the approaches to devise musical distances seem to address these issues. This paper introduces a method to build transposition-invariant distances on symbolic musical sequences which are learned from data. It is a hybrid distance which combines learned feature representations of musical sequences with a handcrafted rank distance. This distance depends less on the musical encoding of the data than previous methods and gives perceptually good results. We demonstrate its efficiency on the dataset of chorale melodies by J.S. Bach. | An approach in the context of shape matching which shares the same motivations as ours can be found in @cite_21 . Contrary to images, two shapes are considered to be identical if we can obtain one by applying a group transformation on the other one. These group transformations can be as above the group of displacements (translations and rotations), but can also be, in the context of shapes, the affine group (translations, rotations and rescalings). A distance between shapes is then constructed by introducing a distance on the integral invariants of a shape. These handcrafted quantities are invariant with respect to the group of transformations acting on shape contours and thus assert that the constructed distance is well-defined on shapes. The main difference with the approaches in image is that the feature representations (here the integral invariants) is not learned from data but constructed by hand. | {
"cite_N": [
"@cite_21"
],
"mid": [
"2143357187"
],
"abstract": [
"For shapes represented as closed planar contours, we introduce a class of functionals which are invariant with respect to the Euclidean group and which are obtained by performing integral operations. While such integral invariants enjoy some of the desirable properties of their differential counterparts, such as locality of computation (which allows matching under occlusions) and uniqueness of representation (asymptotically), they do not exhibit the noise sensitivity associated with differential quantities and, therefore, do not require presmoothing of the input shape. Our formulation allows the analysis of shapes at multiple scales. Based on integral invariants, we define a notion of distance between shapes. The proposed distance measure can be computed efficiently and allows warping the shape boundaries onto each other; its computation results in optimal point correspondence as an intermediate step. Numerical results on shape matching demonstrate that this framework can match shapes despite the deformation of subparts, missing parts and noise. As a quantitative analysis, we report matching scores for shape retrieval from a database"
]
} |
1709.00588 | 2752284099 | Batched sparse (BATS) code is a promising technology for reliable data transmission in multi-hop wireless networks. As a BATS code consists of an outer code and an inner code that typically is a random linear network code, one main research topic for BATS codes is to design an inner code with good performance in transmission efficiency and complexity. In this paper, this issue is addressed with a focus on the problem of minimizing the total number of packets transmitted by the source and intermediate nodes. Subsequently, the problem is formulated as a mixed integer nonlinear programming (MINLP) problem that is NP-hard in general. By exploiting the properties of inner codes and the incomplete beta function, we construct a nonlinear programming (NLP) problem that gives a valid upper bound on the best performance that can be achieved by any feasible solutions. Moreover, both centralized and decentralized real-time optimization strategies are developed. In particular, the decentralized approach is performed independently by each node to find a feasible solution in linear time with the use of look-up tables. Numerical results show that the gap in performance between our proposed approaches and the upper bound is very small, which demonstrates that all feasible solutions developed in the paper are near-optimal with a guaranteed performance bound. | Ng @cite_22 studied the performance of finite-length BATS codes with respect to BP decoding. In @cite_15 , @cite_13 and @cite_2 , the authors proposed a BATS-based network protocol and evaluated the performance over lossy channels. In particular, Yang @cite_15 designed a simple packet interleaving scheme to combat against the bursty losses. In the mean time, Huang @cite_13 proposed a FUN framework, where an inner-encoding algorithm was designed to mix the packets belonging to two intersecting flows. Zhang @cite_2 further extended their previous work @cite_13 and indicated both theoretically and practically that their algorithms performed better than the exiting approaches in TDMA multi-hop networks. In @cite_23 , the authors proposed a distributed two-phase cooperative broadcasting protocol, which uses BATS codes in the first phase to help the peer-to-peer (P2P) communications. | {
"cite_N": [
"@cite_22",
"@cite_23",
"@cite_2",
"@cite_15",
"@cite_13"
],
"mid": [
"2949775276",
"2962724635",
"",
"2059684490",
"2006529249"
],
"abstract": [
"BATS codes were proposed for communication through networks with packet loss. A BATS code consists of an outer code and an inner code. The outer code is a matrix generation of a fountain code, which works with the inner code that comprises random linear coding at the intermediate network nodes. In this paper, the performance of finite-length BATS codes is analyzed with respect to both belief propagation (BP) decoding and inactivation decoding. Our results enable us to evaluate efficiently the finite-length performance in terms of the number of batches used for decoding ranging from 1 to a given maximum number, and provide new insights on the decoding performance. Specifically, for a fixed number of input symbols and a range of the number of batches used for decoding, we obtain recursive formulae to calculate respectively the stopping time distribution of BP decoding and the inactivation probability in inactivation decoding. We also find that both the failure probability of BP decoding and the expected number of inactivations in inactivation decoding can be expressed in a power-sum form where the number of batches appears only as the exponent. This power-sum expression reveals clearly how the decoding failure probability and the expected number of inactivation decrease with the number of batches. When the number of batches used for decoding follows a Poisson distribution, we further derive recursive formulae with potentially lower computational complexity for both decoding algorithms. For the BP decoder that consumes batches one by one, three formulae are provided to characterize the expected number of consumed batches until all the input symbols are decoded.",
"In this paper, we consider the wireless broadcasting scenario with a source node sending some common information to a group of closely located users, where each link is subject to certain packet erasures. To ensure reliable information reception by all users, the conventional approach generally requires repeated transmission by the source until all the users are able to decode the information, which is inefficient in many practical scenarios. In this paper, by exploiting the close proximity among the users, we propose a novel two-phase wireless broadcasting protocol with user cooperations based on an efficient batched network code, known as batched sparse (BATS) code. In the first phase, the information packets are encoded into batches with BATS encoder and sequentially broadcasted by the source node until certain terminating criterion is met. In the second phase, the users cooperate with each other by exchanging the network-coded information via peer-to-peer (P2P) communications based on their respective received packets. A fully distributed and light-weight scheduling algorithm is proposed to improve the efficiency of the P2P communication in the second phase. The performance of the proposed two-phase protocol is analyzed and the channel rank distribution at the instance of decoding is derived, based on which the optimal BATS code is designed. Simulation results show that the proposed protocol outperforms most existing cooperative packet exchange schemes, especially when the inter-user links are not reliable. Lastly, the performance of the proposed scheme is further verified via testbed experiments.",
"",
"BATS code is a low-complexity random linear network coding scheme that can achieve asymptotic bandwidth optimality for many types of networks with packet loss. In this paper, we propose a BATS code based network protocol and evaluate the performance by real-device experiments. Our results demonstrate significant ready-to-implement gain of network coding over forwarding in multi-hop network transmission with packet loss. We also propose an improved protocol to handle the practical issues observed in the experiments.",
"To address the problem of information spreading over lossy communication channels, this paper proposes a joint FoUntain coding and Network coding (FUN) approach. Different from the Transmission Control Protocol (TCP), our FUN approach is a mechanism of Forward Error Correction (FEC), which does not use retransmission for recovery of lost packets. The novelty of our FUN approach lies in combining the best features of fountain coding, intra-session network coding, and cross-next-hop network coding. As such, our FUN approach is capable of achieving unprecedented high throughput over lossy channels. Experimental results demonstrate that our FUN approach achieves higher throughput than the existing schemes for multihop wireless networks."
]
} |
1709.00663 | 2753880617 | Zero shot learning in Image Classification refers to the setting where images from some novel classes are absent in the training data but other information such as natural language descriptions or attribute vectors of the classes are available. This setting is important in the real world since one may not be able to obtain images of all the possible classes at training. While previous approaches have tried to model the relationship between the class attribute space and the image space via some kind of a transfer function in order to model the image space correspondingly to an unseen class, we take a different approach and try to generate the samples from the given attributes, using a conditional variational autoencoder, and use the generated samples for classification of the unseen classes. By extensive testing on four benchmark datasets, we show that our model outperforms the state of the art, particularly in the more realistic generalized setting, where the training classes can also appear at the test time along with the novel classes. | In another popular approach to ZSL, the seen class attributes are treated as the basis vectors @cite_30 which map images(visual features) into the semantic embedding space via the convex combination of the class label embedding vectors weighed by the predictive probabilities for the different training class labels. SYNC @cite_42 tries to align the semantic space to the image space by learning manifold embedding of graphs composed of object classes. @cite_34 present a sparse coding framework based on an unsupervised domain adaptation for ZSL. | {
"cite_N": [
"@cite_30",
"@cite_42",
"@cite_34"
],
"mid": [
"2950700180",
"",
"2209594346"
],
"abstract": [
"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.",
"",
"Zero-shot learning (ZSL) can be considered as a special case of transfer learning where the source and target domains have different tasks label spaces and the target domain is unlabelled, providing little guidance for the knowledge transfer. A ZSL method typically assumes that the two domains share a common semantic representation space, where a visual feature vector extracted from an image video can be projected embedded using a projection function. Existing approaches learn the projection function from the source domain and apply it without adaptation to the target domain. They are thus based on naive knowledge transfer and the learned projections are prone to the domain shift problem. In this paper a novel ZSL method is proposed based on unsupervised domain adaptation. Specifically, we formulate a novel regularised sparse coding framework which uses the target domain class labels' projections in the semantic space to regularise the learned target domain projection thus effectively overcoming the projection domain shift problem. Extensive experiments on four object and action recognition benchmark datasets show that the proposed ZSL method significantly outperforms the state-of-the-arts."
]
} |
1709.00663 | 2753880617 | Zero shot learning in Image Classification refers to the setting where images from some novel classes are absent in the training data but other information such as natural language descriptions or attribute vectors of the classes are available. This setting is important in the real world since one may not be able to obtain images of all the possible classes at training. While previous approaches have tried to model the relationship between the class attribute space and the image space via some kind of a transfer function in order to model the image space correspondingly to an unseen class, we take a different approach and try to generate the samples from the given attributes, using a conditional variational autoencoder, and use the generated samples for classification of the unseen classes. By extensive testing on four benchmark datasets, we show that our model outperforms the state of the art, particularly in the more realistic generalized setting, where the training classes can also appear at the test time along with the novel classes. | A key component of zero-shot learning is the semantic embedding of the class labels. Previous work in ZSL has used human labeled visual attributes @cite_38 to help detect unseen object categories for the datasets where such human labeled attributes are available. This work also uses the attribute vectors as the semantic class embeddings. Distributed representations of the class name such as @cite_32 have also been used as the semantic embedding. In this work we use word2vec for datasets where attribute vectors are not available. | {
"cite_N": [
"@cite_38",
"@cite_32"
],
"mid": [
"2098411764",
"2950133940"
],
"abstract": [
"We propose to shift the goal of recognition from naming to describing. Doing so allows us not only to name familiar objects, but also: to report unusual aspects of a familiar object (“spotty dog”, not just “dog”); to say something about unfamiliar objects (“hairy and four-legged”, not just “unknown”); and to learn how to recognize new objects with few or no visual examples. Rather than focusing on identity assignment, we make inferring attributes the core problem of recognition. These attributes can be semantic (“spotty”) or discriminative (“dogs have it but sheep do not”). Learning attributes presents a major new challenge: generalization across object categories, not just across instances within a category. In this paper, we also introduce a novel feature selection method for learning attributes that generalize well across categories. We support our claims by thorough evaluation that provides insights into the limitations of the standard recognition paradigm of naming and demonstrates the new abilities provided by our attribute-based framework.",
"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."
]
} |
1709.00663 | 2753880617 | Zero shot learning in Image Classification refers to the setting where images from some novel classes are absent in the training data but other information such as natural language descriptions or attribute vectors of the classes are available. This setting is important in the real world since one may not be able to obtain images of all the possible classes at training. While previous approaches have tried to model the relationship between the class attribute space and the image space via some kind of a transfer function in order to model the image space correspondingly to an unseen class, we take a different approach and try to generate the samples from the given attributes, using a conditional variational autoencoder, and use the generated samples for classification of the unseen classes. By extensive testing on four benchmark datasets, we show that our model outperforms the state of the art, particularly in the more realistic generalized setting, where the training classes can also appear at the test time along with the novel classes. | Prior work on conditional image generation based on textual descriptions was successful in generating synthetic images that appear natural @cite_1 . The authors also generate real looking images in the zero shot setting. This serves as the primary motivation behind our approach. However, we work in the feature space instead of the image space. | {
"cite_N": [
"@cite_1"
],
"mid": [
"2949999304"
],
"abstract": [
"Automatic synthesis of realistic images from text would be interesting and useful, but current AI systems are still far from this goal. However, in recent years generic and powerful recurrent neural network architectures have been developed to learn discriminative text feature representations. Meanwhile, deep convolutional generative adversarial networks (GANs) have begun to generate highly compelling images of specific categories, such as faces, album covers, and room interiors. In this work, we develop a novel deep architecture and GAN formulation to effectively bridge these advances in text and image model- ing, translating visual concepts from characters to pixels. We demonstrate the capability of our model to generate plausible images of birds and flowers from detailed text descriptions."
]
} |
1709.00672 | 2751669761 | In recent years, deep discriminative models have achieved extraordinary performance on supervised learning tasks, significantly outperforming their generative counterparts. However, their success relies on the presence of a large amount of labeled data. How can one use the same discriminative models for learning useful features in the absence of labels? We address this question in this paper, by jointly modeling the distribution of data and latent features in a manner that explicitly assigns zero probability to unobserved data. Rather than maximizing the marginal probability of observed data, we maximize the joint probability of the data and the latent features using a two step EM-like procedure. To prevent the model from overfitting to our initial selection of latent features, we use adversarial regularization. Depending on the task, we allow the latent features to be one-hot or real-valued vectors and define a suitable prior on the features. For instance, one-hot features correspond to class labels and are directly used for the unsupervised and semi-supervised classification task, whereas real-valued feature vectors are fed as input to simple classifiers for auxiliary supervised discrimination tasks. The proposed model, which we dub discriminative encoder (or DisCoder), is flexible in the type of latent features that it can capture. The proposed model achieves state-of-the-art performance on several challenging tasks. | Self-supervised learning proceeds by generating labels from the data using information present in the structure of the data. There has been a growing interest in such methods, since no extra effort is need for labeling these samples. For instance, Dosovitskiy @cite_35 train a model for training a convolutional network that assigns each image to its own class. In particular, each image is used as a seed to generate a class of images by applying various transformations. Hence, the output layer of the CNN grows linearly with the number of images in the dataset, severely limiting the scalability of the model. | {
"cite_N": [
"@cite_35"
],
"mid": [
"2148349024"
],
"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)."
]
} |
1709.00672 | 2751669761 | In recent years, deep discriminative models have achieved extraordinary performance on supervised learning tasks, significantly outperforming their generative counterparts. However, their success relies on the presence of a large amount of labeled data. How can one use the same discriminative models for learning useful features in the absence of labels? We address this question in this paper, by jointly modeling the distribution of data and latent features in a manner that explicitly assigns zero probability to unobserved data. Rather than maximizing the marginal probability of observed data, we maximize the joint probability of the data and the latent features using a two step EM-like procedure. To prevent the model from overfitting to our initial selection of latent features, we use adversarial regularization. Depending on the task, we allow the latent features to be one-hot or real-valued vectors and define a suitable prior on the features. For instance, one-hot features correspond to class labels and are directly used for the unsupervised and semi-supervised classification task, whereas real-valued feature vectors are fed as input to simple classifiers for auxiliary supervised discrimination tasks. The proposed model, which we dub discriminative encoder (or DisCoder), is flexible in the type of latent features that it can capture. The proposed model achieves state-of-the-art performance on several challenging tasks. | Another set of methods force the model to learn the relative position of various patches in the image with respect to each other. For instance, jigsaw networks @cite_8 permute the patches of the image, and train the network to predict the correct permutation. Similarly, context prediction networks @cite_19 are trained to predict the correct relationship relationship between two patches. To successfully complete this task, the models are forced to learn representations that capture the global structure of the image. Models that rely on videos as training data, attempt to learn features by exploiting the motion information present in videos. In particular, the approach in @cite_19 , identifies two patches that correspond to the same trajectory in a video, and minimizes the distance between their representations. This forces the model to learn feature that are invariant to orientation of the object in the video. | {
"cite_N": [
"@cite_19",
"@cite_8"
],
"mid": [
"2950187998",
"2321533354"
],
"abstract": [
"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 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)."
]
} |
1709.00575 | 2753027563 | The ubiquity of metaphor in our everyday communication makes it an important problem for natural language understanding. Yet, the majority of metaphor processing systems to date rely on hand-engineered features and there is still no consensus in the field as to which features are optimal for this task. In this paper, we present the first deep learning architecture designed to capture metaphorical composition. Our results demonstrate that it outperforms the existing approaches in the metaphor identification task. | presented a metaphor identification method that uses representations constructed from human property norms @cite_13 . They first learn a mapping from the skip-gram embedding vector space to the property norm space using linear regression, which allows them to generate property norm representations for unseen words. The authors then train an SVM classifier to detect metaphors using these representations as input. have shown that the cognitively-driven property norms outperform standard skip-gram representations in this task. | {
"cite_N": [
"@cite_13"
],
"mid": [
"2251625530"
],
"abstract": [
"Multimodal semantic models attempt to ground distributional semantics through the integration of visual or perceptual information. Feature norms provide useful insight into human concept acquisition, but cannot be used to ground large-scale semantics because they are expensive to produce. We present an automatic method for predicting feature norms for new concepts by learning a mapping from a text-based distributional semantic space to a space built using feature norms. Our experimental results show that we are able to generalise feature-based concept representations, which opens up the possibility of developing large-scale semantic models grounded in a proxy for human perceptual data."
]
} |
1709.00513 | 2796573347 | There is an increasing interest on accelerating neural networks for real-time applications. We study the student-teacher strategy, in which a small and fast student network is trained with the auxiliary information learned from a large and accurate teacher network. We propose to use conditional adversarial networks to learn the loss function to transfer knowledge from teacher to student. The proposed method is particularly effective for relatively small student networks. Moreover, experimental results show the effect of network size when the modern networks are used as student. We empirically study the trade-off between inference time and classification accuracy, and provide suggestions on choosing a proper student network. | has been extensively studied over recent years since @cite_17 . GAN trains two neural networks, the generator and the discriminator, in an adversarial learning process that alternatively updates the two networks. We apply GAN in the conditional setting @cite_1 @cite_40 @cite_29 @cite_34 , where the generator is conditioned on input images. Unlike previous works that focused on image generation, we aim at learning a loss function for knowledge distillation, which requires quite different architectural choices for our generator and discriminator. | {
"cite_N": [
"@cite_29",
"@cite_1",
"@cite_40",
"@cite_34",
"@cite_17"
],
"mid": [
"2405756170",
"2125389028",
"2552465644",
"",
"2099471712"
],
"abstract": [
"Automatic synthesis of realistic images from text would be interesting and useful, but current AI systems are still far from this goal. However, in recent years generic and powerful recurrent neural network architectures have been developed to learn discriminative text feature representations. Meanwhile, deep convolutional generative adversarial networks (GANs) have begun to generate highly compelling images of specific categories, such as faces, album covers, and room interiors. In this work, we develop a novel deep architecture and GAN formulation to effectively bridge these advances in text and image modeling, translating visual concepts from characters to pixels. We demonstrate the capability of our model to generate plausible images of birds and flowers from detailed text descriptions.",
"Generative Adversarial Nets [8] were recently introduced as a novel way to train generative models. In this work we introduce the conditional version of generative adversarial nets, which can be constructed by simply feeding the data, y, we wish to condition on to both the generator and discriminator. We show that this model can generate MNIST digits conditioned on class labels. We also illustrate how this model could be used to learn a multi-modal model, and provide preliminary examples of an application to image tagging in which we demonstrate how this approach can generate descriptive tags which are not part of training labels.",
"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.",
"",
"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."
]
} |
1709.00537 | 2751369593 | We propose a communicationally and computationally efficient algorithm for high-dimensional distributed sparse learning. At each iteration, local machines compute the gradient on local data and the master machine solves one shifted @math regularized minimization problem. The communication cost is reduced from constant times of the dimension number for the state-of-the-art algorithm to constant times of the sparsity number via Two-way Truncation procedure. Theoretically, we prove that the estimation error of the proposed algorithm decreases exponentially and matches that of the centralized method under mild assumptions. Extensive experiments on both simulated data and real data verify that the proposed algorithm is efficient and has performance comparable with the centralized method on solving high-dimensional sparse learning problems. | There is much previous work on distributed optimizations such as ( @cite_4 ; @cite_6 ; @cite_12 ; Shamir and Srebro @cite_8 ; Arjevani and Shamir @cite_9 ; @cite_18 ; Zhang and Xiao @cite_16 ). Initially, most distributed algorithms used averaging estimators formed by local machines ( @cite_4 ; @cite_12 ). Then Zhang and Xiao @cite_16 , @cite_3 and @cite_15 proposed more communication-efficient distributed optimization algorithms. More recently, using ideas of the approximate Newton-type method, @cite_1 and @cite_11 further improved the computational efficiency of this type of method. | {
"cite_N": [
"@cite_18",
"@cite_4",
"@cite_8",
"@cite_9",
"@cite_1",
"@cite_6",
"@cite_3",
"@cite_15",
"@cite_16",
"@cite_12",
"@cite_11"
],
"mid": [
"2235185944",
"",
"2078831169",
"",
"2395943772",
"2130062883",
"2963992805",
"1839685963",
"",
"",
"2395344476"
],
"abstract": [
"We study distributed optimization algorithms for minimizing the average of convex functions. The applications include empirical risk minimization problems in statistical machine learning where the datasets are large and have to be stored on different machines. We design a distributed stochastic variance reduced gradient algorithm that, under certain conditions on the condition number, simultaneously achieves the optimal parallel runtime, amount of communication and rounds of communication among all distributed first-order methods up to constant factors. Our method and its accelerated extension also outperform existing distributed algorithms in terms of the rounds of communication as long as the condition number is not too large compared to the size of data in each machine. We also prove a lower bound for the number of rounds of communication for a broad class of distributed first-order methods including the proposed algorithms in this paper. We show that our accelerated distributed stochastic variance reduced gradient algorithm achieves this lower bound so that it uses the fewest rounds of communication among all distributed first-order algorithms.",
"",
"We consider the problem of distributed stochastic optimization, where each of several machines has access to samples from the same source distribution, and the goal is to jointly optimize the expected objective w.r.t. the source distribution, minimizing: (1) overall runtime; (2) communication costs; (3) number of samples used. We study this problem systematically, highlighting fundamental limitations, and differences versus distributed consensus problems where each machine has a different, independent, objective. We show how the best known guarantees are obtained by an accelerated mini-batched SGD approach, and contrast the runtime and sample costs of the approach with those of other distributed optimization algorithms.",
"",
"We present the Communication-efficient Surrogate Likelihood (CSL) framework for solving distributed statistical learning problems. CSL provides a communication-efficient surrogate to the global likelihood that can be used for low-dimensional estimation, high-dimensional regularized estimation and Bayesian inference. For low-dimensional estimation, CSL provably improves upon the averaging schemes and facilitates the construction of confidence intervals. For high-dimensional regularized estimation, CSL leads to a minimax optimal estimator with minimal communication cost. For Bayesian inference, CSL can be used to form a communication-efficient quasi-posterior distribution that converges to the true posterior. This quasi-posterior procedure significantly improves the computational efficiency of MCMC algorithms even in a non-distributed setting. The methods are illustrated through empirical studies.",
"Online prediction methods are typically presented as serial algorithms running on a single processor. However, in the age of web-scale prediction problems, it is increasingly common to encounter situations where a single processor cannot keep up with the high rate at which inputs arrive. In this work, we present the distributed mini-batch algorithm, a method of converting many serial gradient-based online prediction algorithms into distributed algorithms. We prove a regret bound for this method that is asymptotically optimal for smooth convex loss functions and stochastic inputs. Moreover, our analysis explicitly takes into account communication latencies between nodes in the distributed environment. We show how our method can be used to solve the closely-related distributed stochastic optimization problem, achieving an asymptotically linear speed-up over multiple processors. Finally, we demonstrate the merits of our approach on a web-scale online prediction problem.",
"We present a novel Newton-type method for distributed optimization, which is particularly well suited for stochastic optimization and learning problems. For quadratic objectives, the method enjoys a linear rate of convergence which provably improves with the data size, requiring an essentially constant number of iterations under reasonable assumptions. We provide theoretical and empirical evidence of the advantages of our method compared to other approaches, such as one-shot parameter averaging and ADMM.",
"We devise a one-shot approach to distributed sparse regression in the high-dimensional setting. The key idea is to average \"debiased\" or \"desparsified\" lasso estimators. We show the approach converges at the same rate as the lasso as long as the dataset is not split across too many machines. We also extend the approach to generalized linear models.",
"",
"",
"We propose a novel, efficient approach for distributed sparse learning in high-dimensions, where observations are randomly partitioned across machines. Computationally, at each round our method only requires the master machine to solve a shifted ell_1 regularized M-estimation problem, and other workers to compute the gradient. In respect of communication, the proposed approach provably matches the estimation error bound of centralized methods within constant rounds of communications (ignoring logarithmic factors). We conduct extensive experiments on both simulated and real world datasets, and demonstrate encouraging performances on high-dimensional regression and classification tasks."
]
} |
1709.00443 | 2752186468 | Non-frontal lip views contain useful information which can be used to enhance the performance of frontal view lipreading. However, the vast majority of recent lipreading works, including the deep learning approaches which significantly outperform traditional approaches, have focused on frontal mouth images. As a consequence, research on joint learning of visual features and speech classification from multiple views is limited. In this work, we present an end-to-end multi-view lipreading system based on Bidirectional Long-Short Memory (BLSTM) networks. To the best of our knowledge, this is the first model which simultaneously learns to extract features directly from the pixels and performs visual speech classification from multiple views and also achieves state-of-the-art performance. The model consists of multiple identical streams, one for each view, which extract features directly from different poses of mouth images. The temporal dynamics in each stream view are modelled by a BLSTM and the fusion of multiple streams views takes place via another BLSTM. An absolute average improvement of 3 and 3.8 over the frontal view performance is reported on the OuluVS2 database when the best two (frontal and profile) and three views (frontal, profile, 45) are combined, respectively. The best three-view model results in a 10.5 absolute improvement over the current multi-view state-of-the-art performance on OuluVS2, without using external databases for training, achieving a maximum classification accuracy of 96.9 . | Pose-invariant lipreading works belong to the second group. The main goal of such works is to reduce the impact of different poses as it is known that the performance decreases in mismatched train test conditions, i.e., the classifier is trained and tested on different poses. There are two main approaches in the literature for pose-invariant lipreading. The first one trains classifiers using data from all available views in order to build a generic classifier @cite_26 . The second approach applies a mapping to transform features from non-frontal views to the frontal view. @cite_11 apply a linear mapping to transform profile view features to frontal view features. This approach has been extended to mapping other views like 30 , 45 and 60 to the frontal view @cite_12 or to the 30 view @cite_5 . This approach makes it possible to collect a large amount of data on the optimal view and train a robust model. However, as the number of features to be generated by the linear mapping increases the performance is degraded @cite_11 . | {
"cite_N": [
"@cite_5",
"@cite_26",
"@cite_12",
"@cite_11"
],
"mid": [
"2078711706",
"",
"1590203237",
"196088823"
],
"abstract": [
"Computer lip-reading systems are usually designed to work using a full-frontal view of the face. However, many human experts tend to prefer to lip-read using an angled view. In this paper we consider issues related to the best viewing angle for an automated lip-reading system. In particular, we seek answers to the following questions: 1) Do computers lip-read better using a frontal or a non-frontal view of the face? 2) What is the best viewing angle for a computer lip-reading system? 3) How can a computer lip-reading system be made to work independently of viewing angle? We investigate these issues using a purpose built audio-visual dataset that contains simultaneous recordings of a speaker reciting continuous speech at five angles. We find that the system performs best on a non-frontal view, perhaps because lip gestures, such as lip-protrusion and lip-rounding, are more pronounced when viewing from an angle. We also describe a simple linear mapping that allows us to map any view of the face to the view that we find to be optimal. Hence we present a view-independent lip-reading system.",
"",
"In this paper we study the adaptation of visual and audio-visual speech recognition systems to non-ideal visual conditions. We focus on the effects of a changing pose of the speaker relative to the camera, a problem encountered in natural situations. To that purpose, we introduce a pose normalization technique and perform speech recognition from multiple views by generating virtual frontal views from non-frontal images. The proposed method is inspired by pose-invariant face recognition studies and relies on linear regression to find an approximate mapping between images from different poses. Lipreading experiments quantify the loss of performance related to pose changes and the proposed pose normalization techniques, while audio-visual results analyse how an audio-visual system should account for non-frontal poses in terms of the weight assigned to the visual modality in the audio-visual classifier.",
"In recent work, we have concentrated on the problem of lipreading from non-frontal views (poses). In particular, we have focused on the use of profile views, and proposed two approaches for lipreading on basis of visual features extracted from such views: (a) Direct statistical modeling of the features, namely use of view-dependent statistical models; and (b) Normalization of such features by their projection onto the space'' of frontal-view visual features, which allows employing one set of statistical models for all available views. The latter approach has been considered for two only poses (frontal and profile views), and for visual features of a specific dimensionality. In this paper, we further extend this work, by investigating its applicability to the case where data from three views are available (frontal, left- and right-profile). In addition, we examine the effect of visual feature dimensionality on the pose-normalization approach. Our experiments demonstrate that results generalize well to three views, but also that feature dimensionality is crucial to the effectiveness of the approach. In particular, feature dimensionality larger than 30 is detrimental to multi-pose visual speech recognition performance."
]
} |
1709.00443 | 2752186468 | Non-frontal lip views contain useful information which can be used to enhance the performance of frontal view lipreading. However, the vast majority of recent lipreading works, including the deep learning approaches which significantly outperform traditional approaches, have focused on frontal mouth images. As a consequence, research on joint learning of visual features and speech classification from multiple views is limited. In this work, we present an end-to-end multi-view lipreading system based on Bidirectional Long-Short Memory (BLSTM) networks. To the best of our knowledge, this is the first model which simultaneously learns to extract features directly from the pixels and performs visual speech classification from multiple views and also achieves state-of-the-art performance. The model consists of multiple identical streams, one for each view, which extract features directly from different poses of mouth images. The temporal dynamics in each stream view are modelled by a BLSTM and the fusion of multiple streams views takes place via another BLSTM. An absolute average improvement of 3 and 3.8 over the frontal view performance is reported on the OuluVS2 database when the best two (frontal and profile) and three views (frontal, profile, 45) are combined, respectively. The best three-view model results in a 10.5 absolute improvement over the current multi-view state-of-the-art performance on OuluVS2, without using external databases for training, achieving a maximum classification accuracy of 96.9 . | To the best of our knowledge, there are only three works which have attempted multi-view lipreading, i.e., using multiple views of the lips simultaneously. Lucey and Potamianos @cite_24 concatenated discrete cosine transform features extracted from frontal and profile views and fed them to a Hidden Markov Model (HMM). The performance of the 2-view system outperformed the frontal view model. @cite_0 experimented with different CNN architectures which takes as input multiple lip views. The CNNs were combined with LSTMs and trained end-to-end on the OuluVS2 database. They just tested the combination of all 5 views which turned out to be worse than the frontal and profile views, possibly due to lack of enough training data. Finally, @cite_1 used principal componenent analysis (PCA) networks together with LSTMs and HMMs in order to combine multiple views from the OuluVS2 database. The only combination which outperformed the frontal view is frontal + 30 . Similarly to @cite_0 , the combination of all views led to worse performance than most individual views. | {
"cite_N": [
"@cite_24",
"@cite_1",
"@cite_0"
],
"mid": [
"",
"2602044573",
"2605246626"
],
"abstract": [
"",
"Automatic visual speech recognition is an interesting problem in pattern recognition especially when audio data is noisy or not readily available. It is also a very challenging task mainly because of the lower amount of information in the visual articulations compared to the audible utterance. In this work, principle component analysis is applied to the image patches — extracted from the video data — to learn the weights of a two-stage convolutional network. Block histograms are then extracted as the unsupervised learning features. These features are employed to learn a recurrent neural network with a set of long short-term memory cells to obtain spatiotemporal features. Finally, the obtained features are used in a tandem GMM-HMM system for speech recognition. Our results show that the proposed method has outperformed the baseline techniques applied to the OuluVS2 audiovisual database for phrase recognition with the frontal view cross-validation and testing sentence correctness reaching 79 and 73 , respectively, as compared to the baseline of 74 on cross-validation.",
"It is well known that automatic lip-reading (ALR), also known as visual speech recognition (VSR), enhances the performance of speech recognition in a noisy environment and also has applications itself. However, ALR is a challenging task due to various lip shapes and ambiguity of visemes (the basic unit of visual speech information). In this paper, we tackle ALR as a classification task using end-to-end neural network based on convolutional neural network and long short-term memory architecture. We conduct single, cross, and multi-view experiments in speaker independent setting with various network configuration to integrate the multi-view data. We achieve 77.9 , 83.8 , and 78.6 classification accuracies in average on single, cross, and multi-view respectively. This result is better than the best score (76 ) of preliminary single-view results given by ACCV 2016 workshop on multi-view lip-reading audio-visual challenges. It also shows that additional view information helps to improve the performance of ALR with neural network architecture."
]
} |
1709.00779 | 2753797269 | Cell search is the process for a user to detect its neighboring base stations (BSs) and make a cell selection decision. Due to the importance of beamforming gain in millimeter wave (mmWave) and massive MIMO cellular networks, the directional cell search delay performance is investigated. A cellular network with fixed BS and user locations is considered, so that strong temporal correlations exist for the SINR experienced at each BS and user. For Poisson cellular networks with Rayleigh fading channels, a closed-form expression for the spatially averaged mean cell search delay of all users is derived. This mean cell search delay for a noise-limited network (e.g., mmWave network) is proved to be infinite whenever the non-line-of-sight (NLOS) path loss exponent is larger than 2. For interference-limited networks, a phase transition for the mean cell search delay is shown to exist in terms of the number of BS antennas beams @math : the mean cell search delay is infinite when @math is smaller than a threshold and finite otherwise. Beam-sweeping is also demonstrated to be effective in decreasing the cell search delay, especially for the cell edge users. | Due its analytical tractability for cellular networks @cite_23 @cite_1 @cite_20 , stochastic geometry is a natural candidate for analyzing the directional cell search delay in such fixed cellular networks. In particular, stochastic geometry has already been widely used to investigate fixed Poisson network performance through the local delay metric @cite_4 @cite_23 @cite_17 @cite_2 @cite_13 , which characterizes the number of time slots needed for the SINR to exceed a certain SINR level. @cite_4 @cite_23 , the local delay for fixed ad hoc networks was found to be infinite under several standard scenarios such as Rayleigh fading with constant noise. A new phase transition was identified for the interference-limited case in terms of the mean local delay: the latter is finite when certain parameters are above a threshold, and infinite otherwise. The local delay for noise-limited and interference-limited fixed Poisson networks was also investigated in @cite_17 @cite_2 @cite_13 , where it is shown that power control is an efficient method to ensure a finite mean local delay. These previous works mainly focused on omni-directional communications. | {
"cite_N": [
"@cite_4",
"@cite_1",
"@cite_23",
"@cite_2",
"@cite_13",
"@cite_20",
"@cite_17"
],
"mid": [
"2168164497",
"2150166076",
"2006166592",
"2047372980",
"2591767098",
"2145873277",
"1986417832"
],
"abstract": [
"We study a slotted version of the Aloha Medium Access (MAC) protocol in a Mobile Ad-hoc Network (MANET). Our model features transmitters randomly located in the Euclidean plane, according to a Poisson point process and a set of receivers representing the next-hop from every transmitter. We concentrate on the so-called outage scenario, where a successful transmission requires a Signal-to-Interference-and-Noise (SINR) larger than some threshold. We analyze the local delays in such a network, namely the number of times slots required for nodes to transmit a packet to their prescribed next-hop receivers. The analysis depends very much on the receiver scenario and on the variability of the fading. In most cases, each node has finite-mean geometric random delay and thus a positive next hop throughput. However, the spatial (or large population) averaging of these individual finite mean-delays leads to infinite values in several practical cases, including the Rayleigh fading and positive thermal noise case. In some cases it exhibits an interesting phase transition phenomenon where the spatial average is finite when certain model parameters (receiver distance, thermal noise, Aloha medium access probability) are below a threshold and infinite above. To the best of our knowledge, this phenomenon, which we propose to call the wireless contention phase transition, has not been discussed in the literature. We comment on the relationships between the above facts and the heavy tails found in the so-called \"RESTART\" algorithm. We argue that the spatial average of the mean local delays is infinite primarily because of the outage logic, where one transmits full packets at time slots when the receiver is covered at the required SINR and where one wastes all the other time slots. This results in the \"RESTART\" mechanism, which in turn explains why we have infinite spatial average. Adaptive coding offers another nice way of breaking the outage RESTART logic. We show examples where the average delays are finite in the adaptive coding case, whereas they are infinite in the outage case.",
"Cellular networks are usually modeled by placing the base stations on a grid, with mobile users either randomly scattered or placed deterministically. These models have been used extensively but suffer from being both highly idealized and not very tractable, so complex system-level simulations are used to evaluate coverage outage probability and rate. More tractable models have long been desirable. We develop new general models for the multi-cell signal-to-interference-plus-noise ratio (SINR) using stochastic geometry. Under very general assumptions, the resulting expressions for the downlink SINR CCDF (equivalent to the coverage probability) involve quickly computable integrals, and in some practical special cases can be simplified to common integrals (e.g., the Q-function) or even to simple closed-form expressions. We also derive the mean rate, and then the coverage gain (and mean rate loss) from static frequency reuse. We compare our coverage predictions to the grid model and an actual base station deployment, and observe that the proposed model is pessimistic (a lower bound on coverage) whereas the grid model is optimistic, and that both are about equally accurate. In addition to being more tractable, the proposed model may better capture the increasingly opportunistic and dense placement of base stations in future networks.",
"This volume bears on wireless network modeling and performance analysis. The aim is to show how stochastic geometry can be used in a more or less systematic way to analyze the phenomena that arise in this context. It first focuses on medium access control mechanisms used in ad hoc networks and in cellular networks. It then discusses the use of stochastic geometry for the quantitative analysis of routing algorithms in mobile ad hoc networks. The appendix also contains a concise summary of wireless communication principles and of the network architectures considered in the two volumes.",
"The delay till success (DTS) is the mean number of transmissions needed, averaged over the fading, until a single packet is successfully received (decoded) over a wireless link. This paper shows that under a mean and a peak power constraint, random power control can significantly reduce the DTS. We derive the optimal power control policies that minimize the DTS at one link of given length. For most commonly used fading distributions, these optimal power control policies are random on-off policies, whose parameters depend on the fading statistics and the link distance. We present two applications of this result in the context of noise-limited wireless networks: minimizing the local delay (mean delay for successful nearest-neighbor communication) and minimizing the local anycast delay (mean delay for a transmission to any node).",
"",
"Wireless networks are fundamentally limited by the intensity of the received signals and by their interference. Since both of these quantities depend on the spatial location of the nodes, mathematical techniques have been developed in the last decade to provide communication-theoretic results accounting for the networks geometrical configuration. Often, the location of the nodes in the network can be modeled as random, following for example a Poisson point process. In this case, different techniques based on stochastic geometry and the theory of random geometric graphs -including point process theory, percolation theory, and probabilistic combinatorics-have led to results on the connectivity, the capacity, the outage probability, and other fundamental limits of wireless networks. This tutorial article surveys some of these techniques, discusses their application to model wireless networks, and presents some of the main results that have appeared in the literature. It also serves as an introduction to the field for the other papers in this special issue.",
"Communication between two neighboring nodes is a very basic operation in wireless networks. Yet very little research has focused on the local delay in networks with randomly placed nodes, defined as the mean time it takes a node to connect to its nearest neighbor. We study this problem for Poisson networks, first considering interference only, then noise only, and finally and briefly, interference plus noise. In the noiseless case, we analyze four different types of nearest-neighbor communication and compare the extreme cases of high mobility, where a new Poisson process is drawn in each time slot, and no mobility, where only a single realization exists and nodes stay put forever. It turns out that the local delay behaves rather differently in the two cases. We also provide the low- and high-rate asymptotic behavior of the minimum achievable delay in each case. In the cases with noise, power control is essential to keep the delay finite, and randomized power control can drastically reduce the required (mean) power for finite local delay."
]
} |
1709.00572 | 2939633001 | In recent years, there have been numerous developments towards solving multimodal tasks, aiming to learn a stronger representation than through a single modality. Certain aspects of the data can be particularly useful in this case - for example, correlations in the space or time domain across modalities - but should be wisely exploited in order to benefit from their full predictive potential. We propose two deep learning architectures with multimodal cross-connections that allow for dataflow between several feature extractors (XFlow). Our models derive more interpretable features and achieve better performances than models which do not exchange representations, usefully exploiting correlations between audio and visual data, which have a different dimensionality and are nontrivially exchangeable. Our work improves on existing multimodal deep learning algorithms in two essential ways: (1) it presents a novel method for performing cross-modality (before features are learned from individual modalities) and (2) extends the previously proposed cross-connections which only transfer information between streams that process compatible data. Illustrating some of the representations learned by the connections, we analyse their contribution to the increase in discrimination ability and reveal their compatibility with a lip-reading network intermediate representation. We provide the research community with Digits, a new dataset consisting of three data types extracted from videos of people saying the digits 0-9. Results show that both cross-modal architectures outperform their baselines (by up to 11.5 ) when evaluated on the AVletters, CUAVE and Digits datasets, achieving state-of-the-art results. | Applications of multimodal learning are plentiful---for example, @cite_21 find a joint embedding of cooking recipes and images, regularizing the network via a high-level classification objective. This ensures that both recipe and image embeddings utilise the high-level discriminative weights in a similar way, adding a discrimination-based alignment degree to the process. @cite_18 use unsupervised adversarial learning to tackle image segmentation in the medical domain. They transfer a convolutional neural network from the source (MRI) to the target (CT) images domain, using a domain adaptation module (DAM) that maps the target input data to the source feature space. Along with a domain critic module (CNN discriminator), the DAM is placed in a minimax two-player game used to train the framework. @cite_16 develop a teacher network for facial emotion recognition and use it for training a student to learn speech embeddings in the absence of labels. This is achieved by transferring expression annotations from the visual domain (faces) to the speech domain (voices) through cross-modal distillation. | {
"cite_N": [
"@cite_18",
"@cite_21",
"@cite_16"
],
"mid": [
"2798785261",
"2737041163",
"2886300652"
],
"abstract": [
"Convolutional networks (ConvNets) have achieved great successes in various challenging vision tasks. However, the performance of ConvNets would degrade when encountering the domain shift. The domain adaptation is more significant while challenging in the field of biomedical image analysis, where cross-modality data have largely different distributions. Given that annotating the medical data is especially expensive, the supervised transfer learning approaches are not quite optimal. In this paper, we propose an unsupervised domain adaptation framework with adversarial learning for cross-modality biomedical image segmentations. Specifically, our model is based on a dilated fully convolutional network for pixel-wise prediction. Moreover, we build a plug-and-play domain adaptation module (DAM) to map the target input to features which are aligned with source domain feature space. A domain critic module (DCM) is set up for discriminating the feature space of both domains. We optimize the DAM and DCM via an adversarial loss without using any target domain label. Our proposed method is validated by adapting a ConvNet trained with MRI images to unpaired CT data for cardiac structures segmentations, and achieved very promising results.",
"In this paper, we introduce Recipe1M, a new large-scale, structured corpus of over 1m cooking recipes and 800k food images. As the largest publicly available collection of recipe data, Recipe1M affords the ability to train high-capacity models on aligned, multi-modal data. Using these data, we train a neural network to find a joint embedding of recipes and images that yields impressive results on an image-recipe retrieval task. Additionally, we demonstrate that regularization via the addition of a high-level classification objective both improves retrieval performance to rival that of humans and enables semantic vector arithmetic. We postulate that these embeddings will provide a basis for further exploration of the Recipe1M dataset and food and cooking in general. Code, data and models are publicly available",
"Obtaining large, human labelled speech datasets to train models for emotion recognition is a notoriously challenging task, hindered by annotation cost and label ambiguity. In this work, we consider the task of learning embeddings for speech classification without access to any form of labelled audio. We base our approach on a simple hypothesis: that the emotional content of speech correlates with the facial expression of the speaker. By exploiting this relationship, we show that annotations of expression can be transferred from the visual domain (faces) to the speech domain (voices) through cross-modal distillation. We make the following contributions: (i) we develop a strong teacher network for facial emotion recognition that achieves the state of the art on a standard benchmark; (ii) we use the teacher to train a student, tabula rasa, to learn representations (embeddings) for speech emotion recognition without access to labelled audio data; and (iii) we show that the speech emotion embedding can be used for speech emotion recognition on external benchmark datasets. Code, models and data are available."
]
} |
1709.00572 | 2939633001 | In recent years, there have been numerous developments towards solving multimodal tasks, aiming to learn a stronger representation than through a single modality. Certain aspects of the data can be particularly useful in this case - for example, correlations in the space or time domain across modalities - but should be wisely exploited in order to benefit from their full predictive potential. We propose two deep learning architectures with multimodal cross-connections that allow for dataflow between several feature extractors (XFlow). Our models derive more interpretable features and achieve better performances than models which do not exchange representations, usefully exploiting correlations between audio and visual data, which have a different dimensionality and are nontrivially exchangeable. Our work improves on existing multimodal deep learning algorithms in two essential ways: (1) it presents a novel method for performing cross-modality (before features are learned from individual modalities) and (2) extends the previously proposed cross-connections which only transfer information between streams that process compatible data. Illustrating some of the representations learned by the connections, we analyse their contribution to the increase in discrimination ability and reveal their compatibility with a lip-reading network intermediate representation. We provide the research community with Digits, a new dataset consisting of three data types extracted from videos of people saying the digits 0-9. Results show that both cross-modal architectures outperform their baselines (by up to 11.5 ) when evaluated on the AVletters, CUAVE and Digits datasets, achieving state-of-the-art results. | Finally, @cite_25 tackle the audiovisual task presented here by developing a cross-modal approach that fuses input modalities with temporal structure. A maximum correlation loss term is used to facilitate cross-modal learning, while an attention model adjusts the contribution of modalities towards the joint representation. Once more, we note the lack of involved information exchange in earlier stages of the learning process, concluding the section by iterating that our model obtains superior performance through cross-modal feature exchanges that are concurrent with the feature extraction. | {
"cite_N": [
"@cite_25"
],
"mid": [
"2613677041"
],
"abstract": [
"In recent years, Deep Learning has been successfully applied to multimodal learning problems, with the aim of learning useful joint representations in data fusion applications. When the available modalities consist of time series data such as video, audio and sensor signals, it becomes imperative to consider their temporal structure during the fusion process. In this paper, we propose the Correlational Recurrent Neural Network (CorrRNN), a novel temporal fusion model for fusing multiple input modalities that are inherently temporal in nature. Key features of our proposed model include: (i) simultaneous learning of the joint representation and temporal dependencies between modalities, (ii) use of multiple loss terms in the objective function, including a maximum correlation loss term to enhance learning of cross-modal information, and (iii) the use of an attention model to dynamically adjust the contribution of different input modalities to the joint representation. We validate our model via experimentation on two different tasks: video-and sensor-based activity classification, and audio-visual speech recognition. We empirically analyze the contributions of different components of the proposed CorrRNN model, and demonstrate its robustness, effectiveness and state-of-the-art performance on multiple datasets."
]
} |
1709.00372 | 2949974237 | Timeslices are often used to draw and visualize dynamic graphs. While timeslices are a natural way to think about dynamic graphs, they are routinely imposed on continuous data. Often, it is unclear how many timeslices to select: too few timeslices can miss temporal features such as causality or even graph structure while too many timeslices slows the drawing computation. We present a model for dynamic graphs which is not based on timeslices, and a dynamic graph drawing algorithm, DynNoSlice, to draw graphs in this model. In our evaluation, we demonstrate the advantages of this approach over timeslicing on continuous data sets. | Dynamic graphs visualization is a well established field of research, as shown in the recent survey by Beck @cite_27 . If we were to place our approach in this model, it would be an offline time-to-time mapping'' without timeslices, as it works with full knowledge of the input data. Related approaches also include the superimposed time-to-space mapping'', as this approach inherently defines a space-time cube. | {
"cite_N": [
"@cite_27"
],
"mid": [
"2255903209"
],
"abstract": [
"Dynamic graph visualization focuses on the challenge of representing the evolution of relationships between en- tities in readable, scalable, and effective diagrams. This work surveys the growing number of approaches in this discipline. We derive a hierarchical taxonomy of techniques by systematically categorizing and tagging publica- tions. While static graph visualizations are often divided into node-link and matrix representations, we identify the representation of time as the major distinguishing feature for dynamic graph visualizations: either graphs are represented as animated diagrams or as static charts based on a timeline. Evaluations of animated approaches focus on dynamic stability for preserving the viewer's mental map or, in general, compare animated diagrams to timeline-based ones. Finally, we identify and discuss challenges for future research."
]
} |
1709.00372 | 2949974237 | Timeslices are often used to draw and visualize dynamic graphs. While timeslices are a natural way to think about dynamic graphs, they are routinely imposed on continuous data. Often, it is unclear how many timeslices to select: too few timeslices can miss temporal features such as causality or even graph structure while too many timeslices slows the drawing computation. We present a model for dynamic graphs which is not based on timeslices, and a dynamic graph drawing algorithm, DynNoSlice, to draw graphs in this model. In our evaluation, we demonstrate the advantages of this approach over timeslicing on continuous data sets. | Offline Drawing. Offline dynamic graph drawing algorithms capture all of the dynamic data beforehand and optimize across it simultaneously. Foresighted layout algorithms @cite_34 @cite_16 are among the first attempts at offline dynamic graph drawing. They create a supergraph as the union of the elements at each timeslice. The supergraph is then used to define the position of nodes and edge bends. Mental map preservation is the primary focus of this approach. | {
"cite_N": [
"@cite_34",
"@cite_16"
],
"mid": [
"1498317890",
"2473032827"
],
"abstract": [
"First we introduce the concept of graph animations as a sequence of evolving graphs and a generic algorithm which computes a Foresighted Layout for dynamically drawing these graphs while preserving the mental map. The algorithm is generic in the sense that it takes a static graph drawing algorithm as a parameter. In other words, trees can be animated with a static tree layouter, graphs with a static Sugiyama-style layouter or a spring embedder, etc. Second we discuss applications of Foresighted Layout in algorithm animation and visualization of navigation behaviour.",
"In this paper we present a generic algorithm for drawing sequences of graphs. This algorithm works for different layout algorithms and related metrics and adjustment strategies. It differs from previous work on dynamic graph drawing in that it considers all graphs in the sequence (offline) instead of just the previous ones (online) when computing the layout for each graph of the sequence. We introduce several general adjustment strategies and give examples of these strategies in the context of force-directed graph layout. Finally some results from our first prototype implementation are discussed."
]
} |
1709.00372 | 2949974237 | Timeslices are often used to draw and visualize dynamic graphs. While timeslices are a natural way to think about dynamic graphs, they are routinely imposed on continuous data. Often, it is unclear how many timeslices to select: too few timeslices can miss temporal features such as causality or even graph structure while too many timeslices slows the drawing computation. We present a model for dynamic graphs which is not based on timeslices, and a dynamic graph drawing algorithm, DynNoSlice, to draw graphs in this model. In our evaluation, we demonstrate the advantages of this approach over timeslicing on continuous data sets. | Erten @cite_37 explored several ways to superimpose different graphs, including the use of different colors for different timeslices of a layered space-time cube. In GraphAEL @cite_1 @cite_24 , the space-time cube was created by connecting the same nodes in consecutive timeslices with inter-timeslice edges and optimizing this structure with a force-directed algorithm. A similar technique is used by Groh @cite_15 and by Itoh @cite_31 for social networks. Brandes and Mader @cite_12 perform a metric-based evaluation of several strategies for drawing discrete dynamic graphs including aggregation, anchoring, and linking. Linking strategies performed best in balancing drawing quality (measured using stress) and mental map preservation (measured using distance traveled). Rauber @cite_3 , identify vectors @math in the cost gradient whose geometrical interpretation would be identical when working in a layered space-time cube. | {
"cite_N": [
"@cite_37",
"@cite_15",
"@cite_1",
"@cite_3",
"@cite_24",
"@cite_31",
"@cite_12"
],
"mid": [
"2003783600",
"",
"1522840240",
"2586761003",
"1564138918",
"1993866997",
"2169499674"
],
"abstract": [
"In this paper we consider the problem of drawing and displaying a series of related graphs, i.e., graphs that share all, or parts of the same vertex set. We designed and implemented three different algorithms for simultaneous graph drawing and three different visualization schemes. The algorithms are based on a modification of the force-directed algorithm that allows us to take into account vertex weights and edge weights in order to achieve mental map preservation while obtaining individually readable drawings. The implementation is in Java and the system can be downloaded at http: simg.cs.arizona.edu .",
"",
"GraphAEL extracts three types of evolving graphs from the Graph Drawing literature and creates 2D and 3D animations of the evolutions. We study citation graphs, topic graphs, and collaboration graphs. We also create difference graphs which capture the nature of change between two given time periods. GraphAEL can be accessed online at http: graphael.cs.arizona.edu.",
"Many interesting processes can be represented as time-dependent datasets. We define a time-dependent dataset as a sequence of datasets captured at particular time steps. In such a sequence, each dataset is composed of observations (high-dimensional real vectors), and each observation has a corresponding observation across time steps. Dimensionality reduction provides a scalable alternative to create visualizations (projections) that enable insight into the structure of such datasets. However, applying dimensionality reduction independently for each dataset in a sequence may introduce unnecessary variability in the resulting sequence of projections, which makes tracking the evolution of the data significantly more challenging. We show that this issue affects t-SNE, a widely used dimensionality reduction technique. In this context, we propose dynamic t-SNE, an adaptation of t-SNE that introduces a controllable trade-off between temporal coherence and projection reliability. Our evaluation in two time-dependent datasets shows that dynamic t-SNE eliminates unnecessary temporal variability and encourages smooth changes between projections.",
"The graphael system implements several traditional force-directed layout methods, as well as several novel layout methods for non-Euclidean geometries, including hyperbolic and spherical. The system can handle large graphs, using multi-scale variations of the force-directed methods. Moreover, graphael can layout and visualize graphs that evolve though time, using static views, animation, and morphing. The implementation includes a powerful interface that allows the user to put together existing algorithms and visualization techniques, and to easily add new ones. The system is written in Java and is available as a downloadable program or as an applet at http: graphael.cs.arizona.edu.",
"Social media such as blogs and microblogs enable users to easily and rapidly publish information on their personal activities and interests. They are considered to provide valuable information from the viewpoints of sociology, linguistics, and marketing. This paper proposes a novel system for analyzing temporal changes in the activities and interests of bloggers through a 3D visualization of phrase dependency structures in sentences. We first extract events that represent bloggers' activities and interests through analyzing the phrase dependencies of sentences in a blog archive. We roughly categorize the retrieved events according to the thematic roles (such as the experiencer, agent, and location) of the noun within the events, and then store them in a dependency database so that we can retrieve events that involve a given topic. Second, we present a 3D visualization framework for exploring temporal changes in events related to a topic. Our framework enables users to find events about a topic that appear within a specific timing, and drill down details of the events. It also enables users to compare events with different timings and or on multiple topics. Moreover, it allows them to observe an overview of temporal changes in sets of events, and long-term changes in the frequency of events to assist users in finding trends. We implement the proposed system on our own five-year blog archive that focused on Japanese, and we report the usefulness of our system by using various examples.",
"In dynamic graph drawing, the input is a sequence of graphs for which a sequence of layouts is to be generated such that the quality of individual layouts is balanced with layout stability over time. Qualitatively different extensions of drawing algorithms for static graphs to the dynamic case have been proposed, but little is known about their relative utility. We report on a quantitative study comparing the three prototypical extensions via their adaptation for the stress-minimization framework. While some findings are more subtle, the linking approach connecting consecutive instances of the same vertex is found to be the overall method of choice."
]
} |
1709.00372 | 2949974237 | Timeslices are often used to draw and visualize dynamic graphs. While timeslices are a natural way to think about dynamic graphs, they are routinely imposed on continuous data. Often, it is unclear how many timeslices to select: too few timeslices can miss temporal features such as causality or even graph structure while too many timeslices slows the drawing computation. We present a model for dynamic graphs which is not based on timeslices, and a dynamic graph drawing algorithm, DynNoSlice, to draw graphs in this model. In our evaluation, we demonstrate the advantages of this approach over timeslicing on continuous data sets. | Time-to-space mapping approaches @cite_27 use one dimension for space and one dimension for time in order to visualize the network @cite_2 @cite_23 @cite_5 @cite_28 . Recent scalable approaches use dimensionality reduction to show time as a curve in the plane 15Bach,15Elzen . By using spatial dimensions to represent time, these visualizations are substantially different from classic dynamic node-link diagrams. | {
"cite_N": [
"@cite_28",
"@cite_27",
"@cite_23",
"@cite_2",
"@cite_5"
],
"mid": [
"2185441041",
"2255903209",
"2106268337",
"2167306162",
"2009664208"
],
"abstract": [
"Event-based egocentric dynamic networks are an important class of networks widely seen in many domains. In this paper, we present a visual analytics approach for these networks by combining data-driven network simplifications with a novel visualization design - EgoNetCloud. In particular, an integrated data processing pipeline is proposed to prune, compress and filter the networks into smaller but salient abstractions. To accommodate the simplified network into the visual design, we introduce a constrained graph layout algorithm on the dynamic network. Through a real-life case study as well as conversations with the domain expert, we demonstrate the effectiveness of the EgoNetCloud design and system in completing analysis tasks on event-based dynamic networks. The user study comparing EgoNetCloud with a working system on academic search confirms the effectiveness and convenience of our visual analytics based approach.",
"Dynamic graph visualization focuses on the challenge of representing the evolution of relationships between en- tities in readable, scalable, and effective diagrams. This work surveys the growing number of approaches in this discipline. We derive a hierarchical taxonomy of techniques by systematically categorizing and tagging publica- tions. While static graph visualizations are often divided into node-link and matrix representations, we identify the representation of time as the major distinguishing feature for dynamic graph visualizations: either graphs are represented as animated diagrams or as static charts based on a timeline. Evaluations of animated approaches focus on dynamic stability for preserving the viewer's mental map or, in general, compare animated diagrams to timeline-based ones. Finally, we identify and discuss challenges for future research.",
"We present a novel dynamic graph visualization technique based on node-link diagrams. The graphs are drawn side-byside from left to right as a sequence of narrow stripes that are placed perpendicular to the horizontal time line. The hierarchically organized vertices of the graphs are arranged on vertical, parallel lines that bound the stripes; directed edges connect these vertices from left to right. To address massive overplotting of edges in huge graphs, we employ a splatting approach that transforms the edges to a pixel-based scalar field. This field represents the edge densities in a scalable way and is depicted by non-linear color mapping. The visualization method is complemented by interaction techniques that support data exploration by aggregation, filtering, brushing, and selective data zooming. Furthermore, we formalize graph patterns so that they can be interactively highlighted on demand. A case study on software releases explores the evolution of call graphs extracted from the JUnit open source software project. In a second application, we demonstrate the scalability of our approach by applying it to a bibliography dataset containing more than 1.5 million paper titles from 60 years of research history producing a vast amount of relations between title words.",
"Networks have remained a challenge for information visualization designers because of the complex issues of node and link layout coupled with the rich set of tasks that users present. This paper offers a strategy based on two principles: (1) layouts are based on user-defined semantic substrates, which are non-overlapping regions in which node placement is based on node attributes, (2) users interactively adjust sliders to control link visibility to limit clutter and thus ensure comprehensibility of source and destination. Scalability is further facilitated by user control of which nodes are visible. We illustrate our semantic substrates approach as implemented in NVSS 1.0 with legal precedent data for up to 1122 court cases in three regions with 7645 legal citations",
"Networks are present in many fields such as finance, sociology, and transportation. Often these networks are dynamic: they have a structural as well as a temporal aspect. In addition to relations occurring over time, node information is frequently present such as hierarchical structure or time-series data. We present a technique that extends the Massive Sequence View ( msv) for the analysis of temporal and structural aspects of dynamic networks. Using features in the data as well as Gestalt principles in the visualization such as closure, proximity, and similarity, we developed node reordering strategies for the msv to make these features stand out that optionally take the hierarchical node structure into account. This enables users to find temporal properties such as trends, counter trends, periodicity, temporal shifts, and anomalies in the network as well as structural properties such as communities and stars. We introduce the circular msv that further reduces visual clutter. In addition, the (circular) msv is extended to also convey time-series data associated with the nodes. This enables users to analyze complex correlations between edge occurrence and node attribute changes. We show the effectiveness of the reordering methods on both synthetic and a rich real-world dynamic network data set."
]
} |
1709.00372 | 2949974237 | Timeslices are often used to draw and visualize dynamic graphs. While timeslices are a natural way to think about dynamic graphs, they are routinely imposed on continuous data. Often, it is unclear how many timeslices to select: too few timeslices can miss temporal features such as causality or even graph structure while too many timeslices slows the drawing computation. We present a model for dynamic graphs which is not based on timeslices, and a dynamic graph drawing algorithm, DynNoSlice, to draw graphs in this model. In our evaluation, we demonstrate the advantages of this approach over timeslicing on continuous data sets. | Online Drawing. Given an incoming stream of data, online drawing algorithms continuously update the current graph drawing to take into account new data. Misue @cite_18 optimize mental map preservation by maintaining the horizontal and vertical order of the nodes through adjustments. Frishman and Tal @cite_20 proposed a different force-directed algorithm which could also be partially executed on the GPU. Gorochowski @cite_9 use node aging to decide how much a node position should be preserved at a given time. Finally, Crnovrsanin @cite_6 adapted the FM 3 @cite_32 multilevel approach to an online setting. | {
"cite_N": [
"@cite_18",
"@cite_9",
"@cite_32",
"@cite_6",
"@cite_20"
],
"mid": [
"1997952588",
"2005011141",
"1504716054",
"2295139528",
"2100112610"
],
"abstract": [
"Abstract Many models in software and information engineering use graph representations; examples are data flow diagrams, state transition diagrams, flow charts, PERT charts, organization charts, Petri nets and entity-relationship diagrams. The usefulness of these graph representations depends on the quality of the layout of the graphs. Automatic graph layout, which can release humans from graph drawing, is now available in several visualization systems. Most automatic layout facilities take a purely combinatorial description of a graph and produce a layout of the graph; these methods are called 'layout creation' methods. For interactive systems, another kind of layout is needed: a facility which can adjust a layout after a change is made by the user or by the application. Although layout adjustment is essential in interactive systems, most existing layout algorithms are designed for layout creation. The use of a layout creation method for layout adjustment may totally rearrange the layout and thus destroy the user's 'mental map' of the diagram; thus a set of layout adjustment methods, separate from layout creation methods, is needed. This paper discusses some layout adjustment methods and the preservation of the 'mental map' of the diagram. First, several models are proposed to make the concept of 'mental map' more precise. Then two kinds of layout adjustments are described. One is an algorithm for rearranging a diagram to avoid overlapping nodes, and the other is a method aimed at changing the focus of interest of the user without destroying the mental map. Next, some experience with visualization systems in which the techniques have been employed is also described.",
"Networks are widely used to describe many natural and technological systems. Understanding how these evolve over time poses a challenge for existing visualization techniques originally developed for fixed network structures. We describe a method of incorporating the concept of aging into evolving networks, where nodes and edges store information related to the amount of local evolutionary change they have experienced. This property is used to generate visualizations that ensure stable substructures maintain relatively fixed spatial positions, allowing them to act as visual markers and providing context for evolutionary change elsewhere. By further supplementing these visualizations with color cues, the resultant animations enable a clearer portrayal of the underlying evolutionary process.",
"Force-directed graph drawing algorithms are widely used for drawing general graphs. However, these methods do not guarantee a sub-quadratic running time in general. We present a new force-directed method that is based on a combination of an efficient multilevel scheme and a strategy for approximating the repulsive forces in the system by rapidly evaluating potential fields. Given a graph G=(V,E), the asymptotic worst case running time of this method is O(|V|log|V|+|E|) with linear memory requirements. In practice, the algorithm generates nice drawings of graphs containing 100000 nodes in less than 5 minutes. Furthermore, it clearly visualizes even the structures of those graphs that turned out to be challenging for some other methods.",
"Graphs provide a visual means for examining relation data and force-directed methods are often used to lay out graphs for viewing. Making sense of a dynamic graph as it evolves over time is challenging, and previous force-directed methods were designed for static graphs. In this paper, we present an incremental version of a multilevel multi-pole layout method with a refinement scheme incorporated, which enables us to visualize online dynamic networks while maintaining a mental map of the graph structure. We demonstrate the effectiveness of our method and compare it to previous methods using several network data sets.",
"This paper presents an algorithm for drawing a sequence of graphs online. The algorithm strives to maintain the global structure of the graph and, thus, the user's mental map while allowing arbitrary modifications between consecutive layouts. The algorithm works online and uses various execution culling methods in order to reduce the layout time and handle large dynamic graphs. Techniques for representing graphs on the GPU allow a speedup by a factor of up to 17 compared to the CPU implementation. The scalability of the algorithm across GPU generations is demonstrated. Applications of the algorithm to the visualization of discussion threads in Internet sites and to the visualization of social networks are provided."
]
} |
1709.00372 | 2949974237 | Timeslices are often used to draw and visualize dynamic graphs. While timeslices are a natural way to think about dynamic graphs, they are routinely imposed on continuous data. Often, it is unclear how many timeslices to select: too few timeslices can miss temporal features such as causality or even graph structure while too many timeslices slows the drawing computation. We present a model for dynamic graphs which is not based on timeslices, and a dynamic graph drawing algorithm, DynNoSlice, to draw graphs in this model. In our evaluation, we demonstrate the advantages of this approach over timeslicing on continuous data sets. | 3D Graph Drawing. Several 2D algorithms have been extended to 3D. Bru and Frick @cite_30 proposed Gem3D , which extends Gem @cite_36 to 3D. Cruz and Twarog @cite_17 extended the simulated annealing approach of Davidson and Harel @cite_0 to 3D. Other algorithms work in both 2D and 3D, such as GRIP by Gajer and Kobourov @cite_21 . Munzner @cite_35 proposed an algorithm to draw directed graphs in a 3D hyperbolic space. Several other algorithms deal with particular constraints, such as orthogonal @cite_25 , or nested @cite_33 drawings. Cordeil 16Cordeil investigate the visualization of graphs in 3D using immersive environments. | {
"cite_N": [
"@cite_30",
"@cite_35",
"@cite_33",
"@cite_36",
"@cite_21",
"@cite_0",
"@cite_25",
"@cite_17"
],
"mid": [
"1867324358",
"1968305696",
"2021014453",
"1503013560",
"2129084377",
"2021053884",
"2239163339",
"2102102459"
],
"abstract": [
"We present a 3-D version of Gem [6], a randomized adaptive layout algorithm for nicely drawing undirected graphs, based on the spring-embedder paradigm [4]. The new version, Gem-3D, contains several improvements besides the adaptation to 3-D geometry.",
"We present the H3 layout technique for drawing large directed graphs as node-link diagrams in 3D hyperbolic space. We can lay out much larger structures than can be handled using traditional techniques for drawing general graphs because we assume a hierarchical nature of the data. We impose a hierarchy on the graph by using domain-specific knowledge to find an appropriate spanning tree. Links which are not part of the spanning tree do not influence the layout but can be selectively drawn by user request. The volume of hyperbolic 3-space increases exponentially, as opposed to the familiar geometric increase of euclidean 3-space. We exploit this exponential amount of room by computing the layout according to the hyperbolic metric. We optimize the cone tree layout algorithm for 3D hyperbolic space by placing children on a hemisphere around the cone mouth instead of on its perimeter. Hyperbolic navigation affords a Focus+Context view of the structure with minimal visual clutter. We have successfully laid out hierarchies of over 20,000 nodes. Our implementation accommodates navigation through graphs too large to be rendered interactively by allowing the user to explicitly prune or expand subtrees.",
"Abstract Most systems for visualizing large information structures use 2D graphics to view networks of nodes and arcs that represent data. To understand large structures it is often necessary to show both small-scale and large-scale structures. This has been called the problem of focus and context. Distortion, rapid zooming, elision and multiple windows are all techniques that have been developed to provide both focus and context within single representations. We review these techniques and argue that 3D visualization has a number of advantages. A system called NestedVision3D (NV3D) will be presented that has been developed to investigate the use of 3D visualization for understanding the structure of large computer programs. NV3D is a system for visualizing large nested graphs using interactive 3D graphics. It has been tested with graphs containing more than 35,000 nodes and 100,000 relationships. We describe NV3D and its design philosophy. Basic navigation is facilitated by a set of 3D widgets, rapid scaling and interactive elision. More experimental features include animations called snakes, which are used to trace dynamic software behavior.",
"",
"This paper describes a system for Graph dRawing with Intelligent Placement, GRIP. The system is designed for drawing large graphs and uses a novel multi-dimensional force-directed method together with fast energy function minimization. The system allows for drawing graphs with tens of thousands of vertices in under a minute on a mid-range PC. To the best of the authors' knowledge GRIP surpasses the fastest previous algorithms. However, speed is not achieved at the expense of quality as the resulting drawings are quite aesthetically pleasing.",
"The paradigm of simulated annealing is applied to the problem of drawing graphs “nicely.” Our algorithm deals with general undirected graphs with straight-line edges, and employs several simple criteria for the aesthetic quality of the result. The algorithm is flexible, in that the relative weights of the criteria can be changed. For graphs of modest size it produces good results, competitive with those produced by other methods, notably, the “spring method” and its variants.",
"There is a large number of effective methodologies and algorithms for the creation of aesthetically pleasing graph drawings in two dimensions. However, representing graphs in three dimensions offers various benefits. The extra dimension gives greater flexibility for placing the vertices and edges of a graph and crossings can be always avoided. On the other hand new challenges arise: current output media have a two-dimensional nature and can only provide a limited resolution and display area. Thus, the resulting drawings become complex and difficult to survey. These disadvantages can be weakened by the use of navigational operations such as rotation, shifting and zooming. These operations enable an effective use of screen space and allow users to resolve ambiguities in large graphs while maintaining their overall mental map. The possibility of changing the viewpoint in 3D will also diminish the relevance of edge crossing in the (two-dimensional) screen representation of the graph.",
"A recent trend in graph drawing is directed to the visualization of graphs in 3D [1, 5, 6]. A promising research direction concerns the extension of proven 2D techniques to 3D. We present a system extending the simulated annealing algorithm of Davidson and Harel [2] for straight-line two-dimensional drawings of general undirected graphs to three dimensions. This system features an advanced 3D user interface that assists the user in choosing and modifying the cost function and the optimization components on-line."
]
} |
1709.00372 | 2949974237 | Timeslices are often used to draw and visualize dynamic graphs. While timeslices are a natural way to think about dynamic graphs, they are routinely imposed on continuous data. Often, it is unclear how many timeslices to select: too few timeslices can miss temporal features such as causality or even graph structure while too many timeslices slows the drawing computation. We present a model for dynamic graphs which is not based on timeslices, and a dynamic graph drawing algorithm, DynNoSlice, to draw graphs in this model. In our evaluation, we demonstrate the advantages of this approach over timeslicing on continuous data sets. | Space-Time Cubes. Sallaberry @cite_10 visualize dynamic graphs by manipulating the space-time cube. Several other information visualization methods perform similar operations in the space-time cube, as discussed in Bach @cite_11 . These approaches assume that the space-time cube is already given and focus on how to accommodate it in 2D, whereas our approach constructs the space-time cube for dynamic graphs using a continuous time dimension. | {
"cite_N": [
"@cite_10",
"@cite_11"
],
"mid": [
"1748521661",
"2393302563"
],
"abstract": [
"In this paper, we present a new approach to exploring dynamic graphs. We have developed a new clustering algorithm for dynamic graphs which finds an ideal clustering for each time-step and links the clusters together. The resulting time-varying clusters are then used to define two visual representations. The first view is an overview that shows how clusters evolve over time and provides an interface to find and select interesting time-steps. The second view consists of a node link diagram of a selected time-step which uses the clustering to efficiently define the layout. By using the time-dependant clustering, we ensure the stability of our visualization and preserve user mental map by minimizing node motion, while simultaneously producing an ideal layout for each time step. Also, as the clustering is computed ahead of time, the second view updates in linear time which allows for interactivity even for graphs with upwards of tens of thousands of nodes.",
"We present the generalized space-time cube, a descriptive model for visualizations of temporal data. Visualizations are described as operations on the cube, which transform the cube's 3D shape into readable 2D visualizations. Operations include extracting subparts of the cube, flattening it across space or time or transforming the cubes geometry and content. We introduce a taxonomy of elementary space-time cube operations and explain how these operations can be combined and parameterized. The generalized space-time cube has two properties: 1 it is purely conceptual without the need to be implemented, and 2 it applies to all datasets that can be represented in two dimensions plus time e.g. geo-spatial, videos, networks, multivariate data. The proper choice of space-time cube operations depends on many factors, for example, density or sparsity of a cube. Hence, we propose a characterization of structures within space-time cubes, which allows us to discuss strengths and limitations of operations. We finally review interactive systems that support multiple operations, allowing a user to customize his view on the data. With this framework, we hope to facilitate the description, criticism and comparison of temporal data visualizations, as well as encourage the exploration of new techniques and systems. This paper is an extension ofBach etal.'s 2014 work."
]
} |
1709.00303 | 2753577568 | Instilling resilience in critical infrastructure (CI) such as dams or power grids is a major challenge for tomorrow's cities and communities. Resilience, here, pertains to a CI's ability to adapt or rapidly recover from disruptive events. In this paper, the problem of optimizing and managing the resilience of CIs is studied. In particular, a comprehensive two-fold framework is proposed to improve CI resilience by considering both the individual CIs and their collective contribution to an entire system of multiple CIs. To this end, a novel analytical resilience index is proposed to measure the effect of each CI's physical components on its probability of failure. In particular, a Markov chain defining each CI's performance state and a Bayesian network modeling the probability of failure are introduced to infer each CI's resilience index. Then, to maximize the resilience of a system of CIs, a novel approach for allocating resources, such as drones or maintenance personnel, is proposed. In particular, a comprehensive resource allocation framework, based on the tools of contract theory, is proposed enabling the system operator to optimally allocate resources, such as, redundant components or monitoring devices to each individual CI based on its economic contribution to the entire system. The optimal solution of the contract-based resilience resource allocation problem is analytically derived using dynamic programming. The proposed framework is then evaluated using a case study pertaining to hydropower dams and their interdependence to the power grid. Simulation results, within the case study, show that the system operator can economically benefit from allocating the resources while dams have a 60 average improvement over their initial resilience indices. | One limitation of these previous studies @cite_15 and @cite_13 @cite_26 @cite_17 @cite_24 , is that individual CIs are abstracted within the system, e.g. as nodes within a generic graph. This provides no information on improving individual CIs resilience as the solutions introduced in these studies @cite_15 and @cite_13 @cite_26 @cite_17 @cite_24 consider the resilience of an entire system of multiple CIs while being agnostic to each individual CI's resilience properties. Indeed, individual CIs and their specific failures are largely abstracted and not considered in enough details. Hence, in such prior art, when resources are allocated within the system, no information is provided on how to effectively allocate them at the level of each CI. | {
"cite_N": [
"@cite_26",
"@cite_24",
"@cite_15",
"@cite_13",
"@cite_17"
],
"mid": [
"2342927499",
"2290758236",
"2025216328",
"2083420849",
"2044739544"
],
"abstract": [
"In this paper, we propose two metrics, i.e., the optimal repair time and the resilience reduction worth, to measure the criticality of the components of a network system from the perspective of their contribution to system resilience. Specifically, the two metrics quantify: 1) the priority with which a failed component should be repaired and re-installed into the network and 2) the potential loss in the optimal system resilience due to a time delay in the recovery of a failed component, respectively. Given the stochastic nature of disruptive events on infrastructure networks, a Monte Carlo-based method is proposed to generate probability distributions of the two metrics for all of the components of the network; then, a stochastic ranking approach based on the Copeland's pairwise aggregation is used to rank components importance. Numerical results are obtained for the IEEE 30-bus test network and a comparison is made with three classical centrality measures.",
"Critical infrastructure protection (CIP) is envisioned to be one of the most challenging security problems in the coming decade. One key challenge in CIP is the ability to allocate resources, either personnel or cyber, to critical infrastructures with different vulnerability and criticality levels. In this work, a contract- theoretic approach is proposed to solve the problem of resource allocation in critical infrastructure with asymmetric information. A control center (CC) is used to design contracts and offer them to infrastructures' owners. A contract can be seen as an agreement between the CC and infrastructures using which the CC allocates resources and gets rewards in return. Contracts are designed in a way to maximize the CC's benefit and motivate each infrastructure to accept a contract and obtain proper resources for its protection. Infrastructures are defined by both vulnerability levels and criticality levels which are unknown to the CC. Therefore, each infrastructure can claim that it is the most vulnerable or critical to gain more resources. A novel mechanism is developed to handle such an asymmetric information while providing the optimal contract that motivates each infrastructure to reveal its actual type. The necessary and sufficient conditions for such resource allocation contracts under asymmetric information are derived. Simulation results show that the proposed contract-theoretic approach maximizes the CC's utility while ensuring that no infrastructure has an incentive to ask for another contract, despite the lack of exact information at the CC.",
"Abstract This paper proposes a new multi-stage framework to analyze infrastructure resilience. For each stage, a series of resilience-based improvement strategies are highlighted and appropriate correlates of resilience identified, to then be combined for establishing an expected annual resilience metric adequate for both single hazards and concurrent multiple hazard types. Taking the power transmission grid in Harris County, Texas, USA, as a case study, this paper compares an original power grid model with several hypothetical resilience-improved models to quantify their effectiveness at different stages of their response evolution to random hazards and hurricane hazards. Results show that the expected annual resilience is mainly compromised by random hazards due to their higher frequency of occurrence relative to hurricane hazards. In addition, under limited resources, recovery sequences play a crucial role in resilience improvement, while under sufficient availability of resources, deploying redundancy, hardening critical components and ensuring rapid recovery are all effective responses regardless of their ordering. The expected annual resilience of the power grid with all three stage improvements increases 0.034 compared to the original grid. Although the improvement is small in absolute magnitude due to the high reliability of real power grids, it can still save millions of dollars per year as assessed by energy experts. This framework can provide insights to design, maintain, and retrofit resilient infrastructure systems in practice.",
"We consider a cyber-physical system consisting of two interacting networks, i.e., a cyber network overlaying a physical network. It is envisioned that these systems are more vulnerable to attacks since node failures in one network may result in (due to the interdependence) failures in the other network, causing a cascade of failures that would potentially lead to the collapse of the entire infrastructure. The robustness of interdependent systems against this sort of catastrophic failure hinges heavily on the allocation of the (interconnecting) links that connect nodes in one network to nodes in the other network. In this paper, we characterize the optimum inter-link allocation strategy against random attacks in the case where the topology of each individual network is unknown. In particular, we analyze the “regular” allocation strategy that allots exactly the same number of bidirectional internetwork links to all nodes in the system. We show, both analytically and experimentally, that this strategy yields better performance (from a network resilience perspective) compared to all possible strategies, including strategies using random allocation, unidirectional interlinks, etc.",
"Network protection against natural and human-caused hazards has become a topical research theme in engineering and social sciences. This paper focuses on the problem of allocating limited retrofit resources over multiple highway bridges to improve the resilience and robustness of the entire transportation system in question. The main modeling challenges in network retrofit problems are to capture the interdependencies among individual transportation facilities and to cope with the extremely high uncertainty in the decision environment. In this paper, we model the network retrofit problem as a two-stage stochastic programming problem that optimizes a mean-risk objective of the system loss. This formulation hedges well against uncertainty, but also imposes computational challenges due to involvement of integer decision variables and increased dimension of the problem. An efficient algorithm is developed, via extending the well-known L-shaped method using generalized benders decomposition, to efficiently handle the binary integer variables in the first stage and the nonlinear recourse in the second stage of the model formulation. The proposed modeling and solution methods are general and can be applied to other network design problems as well."
]
} |
1709.00192 | 2752896014 | Hyperspectral imaging, providing abundant spatial and spectral information simultaneously, has attracted a lot of interest in recent years. Unfortunately, due to the hardware limitations, the hyperspectral image (HSI) is vulnerable to various degradations, such noises (random noise, HSI denoising), blurs (Gaussian and uniform blur, HSI deblurring), and down-sampled (both spectral and spatial downsample, HSI super-resolution). Previous HSI restoration methods are designed for one specific task only. Besides, most of them start from the 1-D vector or 2-D matrix models and cannot fully exploit the structurally spectral-spatial correlation in 3-D HSI. To overcome these limitations, in this work, we propose a unified low-rank tensor recovery model for comprehensive HSI restoration tasks, in which non-local similarity between spectral-spatial cubic and spectral correlation are simultaneously captured by 3-order tensors. Further, to improve the capability and flexibility, we formulate it as a weighted low-rank tensor recovery (WLRTR) model by treating the singular values differently, and study its analytical solution. We also consider the exclusive stripe noise in HSI as the gross error by extending WLRTR to robust principal component analysis (WLRTR-RPCA). Extensive experiments demonstrate the proposed WLRTR models consistently outperform state-of-the-arts in typical low level vision HSI tasks, including denoising, destriping, deblurring and super-resolution. | Image denoising is a test bed for the various technique. Consequently, numerous approaches for HSI denoising have been proposed @cite_18 @cite_21 @cite_76 . The spectral correlation and nonlocal self-similarity are two kinds of intrinsic characteristic underlying a HSI. Most of previous HSI denoising methods focus on the spectral correlation such as the wavelet methods @cite_6 , total variational methods @cite_79 , the low-rank matrix recovery methods @cite_38 @cite_87 @cite_62 @cite_46 , or the nonlocal self-similarity such as BM4D @cite_12 , HOSVD @cite_92 individually. Recently, @cite_94 firstly modeled them simultaneously in tensor format. However, the TDL @cite_94 is quite heuristic and short of a concise formulation, thus lacking of the flexibility to other HSI restoration tasks. Several tensor works @cite_52 @cite_31 @cite_70 followed the research line of @cite_94 , and model the sparsity of the core tensor coefficients in a principled manner. Interested readers could refer to @cite_99 for detailed background of HSI denoising. In this work, we further take the fine-grained intrinsic sparsity of the core tensor coefficients into consideration with the reweighting strategy, so as to better encode the structural correlation. Moreover, our WLRTR model can be well extended to other HSI issues. | {
"cite_N": [
"@cite_99",
"@cite_38",
"@cite_18",
"@cite_62",
"@cite_31",
"@cite_87",
"@cite_70",
"@cite_92",
"@cite_21",
"@cite_52",
"@cite_6",
"@cite_79",
"@cite_94",
"@cite_46",
"@cite_76",
"@cite_12"
],
"mid": [
"2024661139",
"1994040806",
"",
"",
"",
"",
"",
"2154011501",
"2256948621",
"2198155329",
"2160484748",
"2039596145",
"2095906131",
"2492899067",
"2466594406",
"1974438823"
],
"abstract": [
"A framework and an algorithm are presented in order to remove stationary noise from images. This algorithm is called variational stationary noise remover. It can be interpreted both as a restoration method in a Bayesian framework and as a cartoon+texture decomposition method. In numerous denoising applications, the white noise assumption fails. For example, structured patterns such as stripes appear in the images. The model described here addresses these cases. Applications are presented with images acquired using different modalities: scanning electron microscope, FIB-nanotomography, and an emerging fluorescence microscopy technique called selective plane illumination microscopy.",
"Hyperspectral images (HSIs) are often degraded by a mixture of various kinds of noise in the acquisition process, which can include Gaussian noise, impulse noise, dead lines, stripes, and so on. This paper introduces a new HSI restoration method based on low-rank matrix recovery (LRMR), which can simultaneously remove the Gaussian noise, impulse noise, dead lines, and stripes. By lexicographically ordering a patch of the HSI into a 2-D matrix, the low-rank property of the hyperspectral imagery is explored, which suggests that a clean HSI patch can be regarded as a low-rank matrix. We then formulate the HSI restoration problem into an LRMR framework. To further remove the mixed noise, the “Go Decomposition” algorithm is applied to solve the LRMR problem. Several experiments were conducted in both simulated and real data conditions to verify the performance of the proposed LRMR-based HSI restoration method.",
"",
"",
"",
"",
"",
"In this paper, we propose a very simple and elegant patch-based, machine learning technique for image denoising using the higher order singular value decomposition (HOSVD). The technique simply groups together similar patches from a noisy image (with similarity defined by a statistically motivated criterion) into a 3D stack, computes the HOSVD coefficients of this stack, manipulates these coefficients by hard thresholding, and inverts the HOSVD transform to produce the final filtered image. Our technique chooses all required parameters in a principled way, relating them to the noise model. We also discuss our motivation for adopting the HOSVD as an appropriate transform for image denoising. We experimentally demonstrate the excellent performance of the technique on grayscale as well as color images. On color images, our method produces state-of-the-art results, outperforming other color image denoising algorithms at moderately high noise levels. A criterion for optimal patch-size selection and noise variance estimation from the residual images (after denoising) is also presented.",
"The Magnetic Resonance (MR) Imaging technique has important applications in clinical diagnosis and scientific research. However, in practice the MR images are often corrupted by noise. Existing image denoising methods, mostly designed for natural image denoising do not take into account the multiple dimensionality of the 3D MR images, and are thus not suitable for 3D MR images denoising. In this paper, we present a novel noise reduction method for 3D MR images based on low-rank tensor approximation, considering both the non-local spatial self-similarity and the correlation across the slices of the 3D MR images. Specifically, for each exemplar 3D patch, similar 3D patches are first grouped to form a 4th order tensor. As the similar patches contain similar structures, the latent clear MR images can be recovered by a low-rank tensor approximation. To this end, an adaptive higher order singular value thresholding method is proposed. Experimental results on 3D MR images show that the proposed method can provide substantial improvements over the current state-of-the-art image denoising methods in terms of both objective metric and subjective visual quality.",
"Patch-based low-rank models have shown effective in exploiting spatial redundancy of natural images especially for the application of image denoising. However, two-dimensional low-rank model can not fully exploit the spatio-temporal correlation in larger data sets such as multispectral images and 3D MRIs. In this work, we propose a novel low-rank tensor approximation framework with Laplacian Scale Mixture (LSM) modeling for multi-frame image denoising. First, similar 3D patches are grouped to form a tensor of d-order and high-order Singular Value Decomposition (HOSVD) is applied to the grouped tensor. Then the task of multiframe image denoising is formulated as a Maximum A Posterior (MAP) estimation problem with the LSM prior for tensor coefficients. Both unknown sparse coefficients and hidden LSM parameters can be efficiently estimated by the method of alternating optimization. Specifically, we have derived closed-form solutions for both subproblems. Experimental results on spectral and dynamic MRI images show that the proposed algorithm can better preserve the sharpness of important image structures and outperform several existing state-of-the-art multiframe denoising methods (e.g., BM4D and tensor dictionary learning).",
"In this paper, a new noise reduction algorithm is introduced and applied to the problem of denoising hyperspectral imagery. This algorithm resorts to the spectral derivative domain, where the noise level is elevated, and benefits from the dissimilarity of the signal regularity in the spatial and the spectral dimensions of hyperspectral images. The performance of the new algorithm is tested on two different hyperspectral datacubes: an Airborne Visible Infrared Imaging Spectrometer (AVIRIS) datacube that is acquired in a vegetation-dominated site and a simulated AVIRIS datacube that simulates a geological site. The new algorithm provides signal-to-noise-ratio improvement up to 84.44 and 98.35 in the first and the second datacubes, respectively.",
"The amount of noise included in a hyperspectral image limits its application and has a negative impact on hyperspectral image classification, unmixing, target detection, and so on. In hyperspectral images, because the noise intensity in different bands is different, to better suppress the noise in the high-noise-intensity bands and preserve the detailed information in the low-noise-intensity bands, the denoising strength should be adaptively adjusted with the noise intensity in the different bands. Meanwhile, in the same band, there exist different spatial property regions, such as homogeneous regions and edge or texture regions; to better reduce the noise in the homogeneous regions and preserve the edge and texture information, the denoising strength applied to pixels in different spatial property regions should also be different. Therefore, in this paper, we propose a hyperspectral image denoising algorithm employing a spectral-spatial adaptive total variation (TV) model, in which the spectral noise differences and spatial information differences are both considered in the process of noise reduction. To reduce the computational load in the denoising process, the split Bregman iteration algorithm is employed to optimize the spectral-spatial hyperspectral TV model and accelerate the speed of hyperspectral image denoising. A number of experiments illustrate that the proposed approach can satisfactorily realize the spectral-spatial adaptive mechanism in the denoising process, and superior denoising results are produced.",
"As compared to the conventional RGB or gray-scale images, multispectral images (MSI) can deliver more faithful representation for real scenes, and enhance the performance of many computer vision tasks. In practice, however, an MSI is always corrupted by various noises. In this paper we propose an effective MSI denoising approach by combinatorially considering two intrinsic characteristics underlying an MSI: the nonlocal similarity over space and the global correlation across spectrum. In specific, by explicitly considering spatial self-similarity of an MSI we construct a nonlocal tensor dictionary learning model with a group-block-sparsity constraint, which makes similar full-band patches (FBP) share the same atoms from the spatial and spectral dictionaries. Furthermore, through exploiting spectral correlation of an MSI and assuming over-redundancy of dictionaries, the constrained nonlocal MSI dictionary learning model can be decomposed into a series of unconstrained low-rank tensor approximation problems, which can be readily solved by off-the-shelf higher order statistics. Experimental results show that our method outperforms all state-of-the-art MSI denoising methods under comprehensive quantitative performance measures.",
"Many computer vision problems can be posed as learning a low-dimensional subspace from high dimensional data. The low rank matrix factorization (LRMF) represents a commonly utilized subspace learning strategy. Most of the current LRMF techniques are constructed on the optimization problem using L_1 norm and L_2 norm, which mainly deal with Laplacian and Gaussian noise, respectively. To make LRMF capable of adapting more complex noise, this paper proposes a new LRMF model by assuming noise as Mixture of Exponential Power (MoEP) distributions and proposes a penalized MoEP model by combining the penalized likelihood method with MoEP distributions. Such setting facilitates the learned LRMF model capable of automatically fitting the real noise through MoEP distributions. Each component in this mixture is adapted from a series of preliminary super-or sub-Gaussian candidates. An Expectation Maximization (EM) algorithm is also designed to infer the parameters involved in the proposed PMoEP model. The advantage of our method is demonstrated by extensive experiments on synthetic data, face modeling and hyperspectral image restoration.",
"Hyperspectral imaging is beneficial in a diverse range of applications from diagnostic medicine, to agriculture, to surveillance to name a few. However, hyperspectral images often suffer from degradation such as noise and low resolution. In this paper, we propose an effective model for hyperspectral image (HSI) restoration, specifically image denoising and super-resolution. Our model considers three underlying characteristics of HSIs: sparsity across the spatial-spectral domain, high correlation across spectra, and non-local self-similarity over space. We first exploit high correlation across spectra and non-local self-similarity over space in the degraded HSI to learn an adaptive spatial-spectral dictionary. Then, we employ the local and non-local sparsity of the HSI under the learned spatial-spectral dictionary to design an HSI restoration model, which can be effectively solved by an iterative numerical algorithm with parameters that are adaptively adjusted for different clusters and different noise levels. In experiments on HSI denoising, we show that the proposed method outperforms many state-of-the-art methods under several comprehensive quantitative assessments. We also show that our method performs well on HSI super-resolution.",
"We present an extension of the BM3D filter to volumetric data. The proposed algorithm, BM4D, implements the grouping and collaborative filtering paradigm, where mutually similar d -dimensional patches are stacked together in a (d+1) -dimensional array and jointly filtered in transform domain. While in BM3D the basic data patches are blocks of pixels, in BM4D we utilize cubes of voxels, which are stacked into a 4-D “group.” The 4-D transform applied on the group simultaneously exploits the local correlation present among voxels in each cube and the nonlocal correlation between the corresponding voxels of different cubes. Thus, the spectrum of the group is highly sparse, leading to very effective separation of signal and noise through coefficient shrinkage. After inverse transformation, we obtain estimates of each grouped cube, which are then adaptively aggregated at their original locations. We evaluate the algorithm on denoising of volumetric data corrupted by Gaussian and Rician noise, as well as on reconstruction of volumetric phantom data with non-zero phase from noisy and incomplete Fourier-domain (k-space) measurements. Experimental results demonstrate the state-of-the-art denoising performance of BM4D, and its effectiveness when exploited as a regularizer in volumetric data reconstruction."
]
} |
1709.00192 | 2752896014 | Hyperspectral imaging, providing abundant spatial and spectral information simultaneously, has attracted a lot of interest in recent years. Unfortunately, due to the hardware limitations, the hyperspectral image (HSI) is vulnerable to various degradations, such noises (random noise, HSI denoising), blurs (Gaussian and uniform blur, HSI deblurring), and down-sampled (both spectral and spatial downsample, HSI super-resolution). Previous HSI restoration methods are designed for one specific task only. Besides, most of them start from the 1-D vector or 2-D matrix models and cannot fully exploit the structurally spectral-spatial correlation in 3-D HSI. To overcome these limitations, in this work, we propose a unified low-rank tensor recovery model for comprehensive HSI restoration tasks, in which non-local similarity between spectral-spatial cubic and spectral correlation are simultaneously captured by 3-order tensors. Further, to improve the capability and flexibility, we formulate it as a weighted low-rank tensor recovery (WLRTR) model by treating the singular values differently, and study its analytical solution. We also consider the exclusive stripe noise in HSI as the gross error by extending WLRTR to robust principal component analysis (WLRTR-RPCA). Extensive experiments demonstrate the proposed WLRTR models consistently outperform state-of-the-arts in typical low level vision HSI tasks, including denoising, destriping, deblurring and super-resolution. | Natural image deblurring aims to recover a sharp latent image from a blurred one @cite_88 , which is a classical and active research field within the last decade. Numerous HSI deblurring methods directly learn from the natural single image priors by assuming the widely used sparsity of image gradients, e.g., Huber-Markov prior @cite_19 , the total variational (TV) @cite_60 , and Gaussian mixture model (GMM) @cite_40 . Only recently, the spatial-spectral joint total variational has been introduced to HSI deblurring @cite_22 @cite_55 . In general, most of previous HSI deblurring methods only exploit the spatial information, while none of them have utilized the nonlocal self-similarity presented in HSI. In our work, we focus on the non-blind HSI deblurring, and show that the additional spectral correlation and nonlocal information would significantly improve the HSI deblurring performance. | {
"cite_N": [
"@cite_22",
"@cite_60",
"@cite_55",
"@cite_19",
"@cite_40",
"@cite_88"
],
"mid": [
"1991180630",
"1988868707",
"",
"2024474298",
"2240914251",
"2138204001"
],
"abstract": [
"In this brief, we provide an efficient scheme for performing deconvolution of large hyperspectral images under a positivity constraint, while accounting for spatial and spectral smoothness of the data.",
"The main aim of this paper is to study total variation (TV) regularization in deblurring and sparse unmixing of hyperspectral images. In the model, we also incorporate blurring operators for dealing with blurring effects, particularly blurring operators for hyperspectral imaging whose point spread functions are generally system dependent and formed from axial optical aberrations in the acquisition system. An alternating direction method is developed to solve the resulting optimization problem efficiently. According to the structure of the TV regularization and sparse unmixing in the model, the convergence of the alternating direction method can be guaranteed. Experimental results are reported to demonstrate the effectiveness of the TV and sparsity model and the efficiency of the proposed numerical scheme, and the method is compared to the recent Sparse Unmixing via variable Splitting Augmented Lagrangian and TV method by",
"",
"This letter proposes a blind image restoration method for the deblurring of remote sensing images. A simple but robust identification method of point spread function (PSF) support is proposed, and a joint estimation method is presented to simultaneously solve the PSF coefficients and restoration image. To narrow the solution space for the best possible definition, the Huber-Markov (Huber-Markov random field) prior model is employed to regularize the two series of unknowns. Experiments were performed to demonstrate the effectiveness of the proposed approach.",
"The restoration of clean remote sensing images from the degraded images is usually an ill-posed problem. A feasible solution for the problem is incorporating various priors into restoration procedure as constrained conditions. Recently, the patch-based Gaussian mixture model (GMM) priors became popular. However, the learning of patch-based GMM priors usually assumes the components of the GMM are independent and identically distributed (i.i.d.). The assumption means that two components could model the same training samples, and thus a possible decrease in model’s description ability. This work presents a new learning method to diversify the patch-based GMM priors through modeling the GMM’s parameters as a determinantal point process (DPP). The introduction of DPP allows us to specify a preference for diversity in patch-based GMM using a positive definite kernel function and thus to increase the model’s description ability for a given model complexity. Furthermore, under the expected patch log-likelihood (EPLL) restoration framework, we further introduce the diversified patch-based GMM priors into remote sensing literature and develop new remote sensing image restoration algorithms. The extensive experiments for denoising and deblurring of remote sensing images demonstrated that the proposed algorithms produced better restoration results than other methods that already proven to perform well in the literature.",
"Blind deconvolution is the recovery of a sharp version of a blurred image when the blur kernel is unknown. Recent algorithms have afforded dramatic progress, yet many aspects of the problem remain challenging and hard to understand. The goal of this paper is to analyze and evaluate recent blind deconvolution algorithms both theoretically and experimentally. We explain the previously reported failure of the naive MAP approach by demonstrating that it mostly favors no-blur explanations. On the other hand we show that since the kernel size is often smaller than the image size a MAP estimation of the kernel alone can be well constrained and accurately recover the true blur. The plethora of recent deconvolution techniques makes an experimental evaluation on ground-truth data important. We have collected blur data with ground truth and compared recent algorithms under equal settings. Additionally, our data demonstrates that the shift-invariant blur assumption made by most algorithms is often violated."
]
} |
1709.00192 | 2752896014 | Hyperspectral imaging, providing abundant spatial and spectral information simultaneously, has attracted a lot of interest in recent years. Unfortunately, due to the hardware limitations, the hyperspectral image (HSI) is vulnerable to various degradations, such noises (random noise, HSI denoising), blurs (Gaussian and uniform blur, HSI deblurring), and down-sampled (both spectral and spatial downsample, HSI super-resolution). Previous HSI restoration methods are designed for one specific task only. Besides, most of them start from the 1-D vector or 2-D matrix models and cannot fully exploit the structurally spectral-spatial correlation in 3-D HSI. To overcome these limitations, in this work, we propose a unified low-rank tensor recovery model for comprehensive HSI restoration tasks, in which non-local similarity between spectral-spatial cubic and spectral correlation are simultaneously captured by 3-order tensors. Further, to improve the capability and flexibility, we formulate it as a weighted low-rank tensor recovery (WLRTR) model by treating the singular values differently, and study its analytical solution. We also consider the exclusive stripe noise in HSI as the gross error by extending WLRTR to robust principal component analysis (WLRTR-RPCA). Extensive experiments demonstrate the proposed WLRTR models consistently outperform state-of-the-arts in typical low level vision HSI tasks, including denoising, destriping, deblurring and super-resolution. | HSI super-resolution refers to the fusion of a hyperspectral image (low spatial but high spectral resolution) with a panchromatic multispectral image (high spatial but low spectral resolution, usually RGB image). The most popular sparsity promoting methods mainly include the sparse representation @cite_4 @cite_13 @cite_75 @cite_48 @cite_56 @cite_14 and the matrix factorization approach @cite_30 @cite_100 @cite_23 @cite_74 @cite_20 . In @cite_4 , the authors applied the dictionary learning embedded in the spectral subspace to exploit the sparsity of hyperspectral images. In @cite_35 , proposed a non-negative structured sparse representation (NSSR) approach with the prior knowledge about spatio-spectral sparsity of the hyperspectral image. Analog to classical super-resolution @cite_29 , a sparse matrix factorization method @cite_85 borrowed the idea that both the LR hyperspectral image and HR RGB image share the same coding coefficients. The HR hyperspectral image was then reconstructed by multiplying the learned basis from the HR RGB image and sparse coefficients from the LR hyperspectral image. The interested readers can refer to the survey @cite_63 . | {
"cite_N": [
"@cite_30",
"@cite_35",
"@cite_14",
"@cite_4",
"@cite_75",
"@cite_48",
"@cite_29",
"@cite_85",
"@cite_56",
"@cite_23",
"@cite_74",
"@cite_63",
"@cite_100",
"@cite_13",
"@cite_20"
],
"mid": [
"1990231296",
"2327302159",
"",
"1969352102",
"",
"",
"2121058967",
"2092116045",
"",
"",
"",
"2963442801",
"",
"",
""
],
"abstract": [
"Coupled nonnegative matrix factorization (CNMF) unmixing is proposed for the fusion of low-spatial-resolution hyperspectral and high-spatial-resolution multispectral data to produce fused data with high spatial and spectral resolutions. Both hyperspectral and multispectral data are alternately unmixed into end member and abundance matrices by the CNMF algorithm based on a linear spectral mixture model. Sensor observation models that relate the two data are built into the initialization matrix of each NMF unmixing procedure. This algorithm is physically straightforward and easy to implement owing to its simple update rules. Simulations with various image data sets demonstrate that the CNMF algorithm can produce high-quality fused data both in terms of spatial and spectral domains, which contributes to the accurate identification and classification of materials observed at a high spatial resolution.",
"Hyperspectral imaging has many applications from agriculture and astronomy to surveillance and mineralogy. However, it is often challenging to obtain high-resolution (HR) hyperspectral images using existing hyperspectral imaging techniques due to various hardware limitations. In this paper, we propose a new hyperspectral image super-resolution method from a low-resolution (LR) image and a HR reference image of the same scene. The estimation of the HR hyperspectral image is formulated as a joint estimation of the hyperspectral dictionary and the sparse codes based on the prior knowledge of the spatial-spectral sparsity of the hyperspectral image. The hyperspectral dictionary representing prototype reflectance spectra vectors of the scene is first learned from the input LR image. Specifically, an efficient non-negative dictionary learning algorithm using the block-coordinate descent optimization technique is proposed. Then, the sparse codes of the desired HR hyperspectral image with respect to learned hyperspectral basis are estimated from the pair of LR and HR reference images. To improve the accuracy of non-negative sparse coding, a clustering-based structured sparse coding method is proposed to exploit the spatial correlation among the learned sparse codes. The experimental results on both public datasets and real LR hypspectral images suggest that the proposed method substantially outperforms several existing HR hyperspectral image recovery techniques in the literature in terms of both objective quality metrics and computational efficiency.",
"",
"Abstract In this study, a novel noise reduction algorithm for hyperspectral imagery (HSI) is proposed based on high-order rank-1 tensor decomposition. The hyperspectral data cube is considered as a three-order tensor that is able to jointly treat both the spatial and spectral modes. Subsequently, the rank-1 tensor decomposition (R1TD) algorithm is applied to the tensor data, which takes into account both the spatial and spectral information of the hyperspectral data cube. A noise-reduced hyperspectral image is then obtained by combining the rank-1 tensors using an eigenvalue intensity sorting and reconstruction technique. Compared with the existing noise reduction methods such as the conventional channel-by-channel approaches and the recently developed multidimensional filter, the spatial–spectral adaptive total variation filter, experiments with both synthetic noisy data and real HSI data reveal that the proposed R1TD algorithm significantly improves the HSI data quality in terms of both visual inspection and image quality indices. The subsequent image classification results further validate the effectiveness of the proposed HSI noise reduction algorithm.",
"",
"",
"This paper presents a new approach to single-image superresolution, based upon sparse signal representation. Research on image statistics suggests that image patches can be well-represented as a sparse linear combination of elements from an appropriately chosen over-complete dictionary. Inspired by this observation, we seek a sparse representation for each patch of the low-resolution input, and then use the coefficients of this representation to generate the high-resolution output. Theoretical results from compressed sensing suggest that under mild conditions, the sparse representation can be correctly recovered from the downsampled signals. By jointly training two dictionaries for the low- and high-resolution image patches, we can enforce the similarity of sparse representations between the low-resolution and high-resolution image patch pair with respect to their own dictionaries. Therefore, the sparse representation of a low-resolution image patch can be applied with the high-resolution image patch dictionary to generate a high-resolution image patch. The learned dictionary pair is a more compact representation of the patch pairs, compared to previous approaches, which simply sample a large amount of image patch pairs , reducing the computational cost substantially. The effectiveness of such a sparsity prior is demonstrated for both general image super-resolution (SR) and the special case of face hallucination. In both cases, our algorithm generates high-resolution images that are competitive or even superior in quality to images produced by other similar SR methods. In addition, the local sparse modeling of our approach is naturally robust to noise, and therefore the proposed algorithm can handle SR with noisy inputs in a more unified framework.",
"Hyperspectral imaging is a promising tool for applications in geosensing, cultural heritage and beyond. However, compared to current RGB cameras, existing hyperspectral cameras are severely limited in spatial resolution. In this paper, we introduce a simple new technique for reconstructing a very high-resolution hyperspectral image from two readily obtained measurements: A lower-resolution hyper-spectral image and a high-resolution RGB image. Our approach is divided into two stages: We first apply an unmixing algorithm to the hyperspectral input, to estimate a basis representing reflectance spectra. We then use this representation in conjunction with the RGB input to produce the desired result. Our approach to unmixing is motivated by the spatial sparsity of the hyperspectral input, and casts the unmixing problem as the search for a factorization of the input into a basis and a set of maximally sparse coefficients. Experiments show that this simple approach performs reasonably well on both simulations and real data examples.",
"",
"",
"",
"Pansharpening aims at fusing a panchromatic image with a multispectral one, to generate an image with the high spatial resolution of the former and the high spectral resolution of the latter. In the last decade, many algorithms have been presented in the literatures for pansharpening using multispectral data. With the increasing availability of hyperspectral systems, these methods are now being adapted to hyperspectral images. In this work, we compare new pansharpening techniques designed for hyperspectral data with some of the state-of-the-art methods for multispectral pansharpening, which have been adapted for hyperspectral data. Eleven methods from different classes (component substitution, multiresolution analysis, hybrid, Bayesian and matrix factorization) are analyzed. These methods are applied to three datasets and their effectiveness and robustness are evaluated with widely used performance indicators. In addition, all the pansharpening techniques considered in this paper have been implemented in a MATLAB toolbox that is made available to the community.",
"",
"",
""
]
} |
1709.00192 | 2752896014 | Hyperspectral imaging, providing abundant spatial and spectral information simultaneously, has attracted a lot of interest in recent years. Unfortunately, due to the hardware limitations, the hyperspectral image (HSI) is vulnerable to various degradations, such noises (random noise, HSI denoising), blurs (Gaussian and uniform blur, HSI deblurring), and down-sampled (both spectral and spatial downsample, HSI super-resolution). Previous HSI restoration methods are designed for one specific task only. Besides, most of them start from the 1-D vector or 2-D matrix models and cannot fully exploit the structurally spectral-spatial correlation in 3-D HSI. To overcome these limitations, in this work, we propose a unified low-rank tensor recovery model for comprehensive HSI restoration tasks, in which non-local similarity between spectral-spatial cubic and spectral correlation are simultaneously captured by 3-order tensors. Further, to improve the capability and flexibility, we formulate it as a weighted low-rank tensor recovery (WLRTR) model by treating the singular values differently, and study its analytical solution. We also consider the exclusive stripe noise in HSI as the gross error by extending WLRTR to robust principal component analysis (WLRTR-RPCA). Extensive experiments demonstrate the proposed WLRTR models consistently outperform state-of-the-arts in typical low level vision HSI tasks, including denoising, destriping, deblurring and super-resolution. | Tensor-based methods in HSI super-resolution is not entirely new. Very recently, @cite_80 proposed a non-local sparse Tucker tensor factorization (NLSTF) model for HSI super-resolution. While both NLSTF and our technique tackle this problem from the tensor perspective, NLSTF fails to make use of an auxiliary HR RGB image. Similar to @cite_94 , its realization is relatively heuristic, and hard to be incorporate additional information. Our unified WLRTR method by-passes the tensor dictionary learning process by using high order singular value decomposition (HOSVD) @cite_92 , benefiting us enough flexibility, such as the auxiliary HR RGB image. Besides, NLSTF does not apply any sparsity prior for the underlying HR HSI. In contrast, we introduce the weighted low-rank tensor prior to further refine the solution. | {
"cite_N": [
"@cite_80",
"@cite_92",
"@cite_94"
],
"mid": [
"2748530166",
"2154011501",
"2095906131"
],
"abstract": [
"Hyperspectral image (HSI) super-resolution, which fuses a low-resolution (LR) HSI with a high-resolution (HR) multispectral image (MSI), has recently attracted much attention. Most of the current HSI super-resolution approaches are based on matrix factorization, which unfolds the three-dimensional HSI as a matrix before processing. In general, the matrix data representation obtained after the matrix unfolding operation makes it hard to fully exploit the inherent HSI spatial-spectral structures. In this paper, a novel HSI super-resolution method based on non-local sparse tensor factorization (called as the NLSTF) is proposed. The sparse tensor factorization can directly decompose each cube of the HSI as a sparse core tensor and dictionaries of three modes, which reformulates the HSI super-resolution problem as the estimation of sparse core tensor and dictionaries for each cube. To further exploit the non-local spatial self-similarities of the HSI, similar cubes are grouped together, and they are assumed to share the same dictionaries. The dictionaries are learned from the LR-HSI and HR-MSI for each group, and corresponding sparse core tensors are estimated by spare coding on the learned dictionaries for each cube. Experimental results demonstrate the superiority of the proposed NLSTF approach over several state-of-the-art HSI super-resolution approaches.",
"In this paper, we propose a very simple and elegant patch-based, machine learning technique for image denoising using the higher order singular value decomposition (HOSVD). The technique simply groups together similar patches from a noisy image (with similarity defined by a statistically motivated criterion) into a 3D stack, computes the HOSVD coefficients of this stack, manipulates these coefficients by hard thresholding, and inverts the HOSVD transform to produce the final filtered image. Our technique chooses all required parameters in a principled way, relating them to the noise model. We also discuss our motivation for adopting the HOSVD as an appropriate transform for image denoising. We experimentally demonstrate the excellent performance of the technique on grayscale as well as color images. On color images, our method produces state-of-the-art results, outperforming other color image denoising algorithms at moderately high noise levels. A criterion for optimal patch-size selection and noise variance estimation from the residual images (after denoising) is also presented.",
"As compared to the conventional RGB or gray-scale images, multispectral images (MSI) can deliver more faithful representation for real scenes, and enhance the performance of many computer vision tasks. In practice, however, an MSI is always corrupted by various noises. In this paper we propose an effective MSI denoising approach by combinatorially considering two intrinsic characteristics underlying an MSI: the nonlocal similarity over space and the global correlation across spectrum. In specific, by explicitly considering spatial self-similarity of an MSI we construct a nonlocal tensor dictionary learning model with a group-block-sparsity constraint, which makes similar full-band patches (FBP) share the same atoms from the spatial and spectral dictionaries. Furthermore, through exploiting spectral correlation of an MSI and assuming over-redundancy of dictionaries, the constrained nonlocal MSI dictionary learning model can be decomposed into a series of unconstrained low-rank tensor approximation problems, which can be readily solved by off-the-shelf higher order statistics. Experimental results show that our method outperforms all state-of-the-art MSI denoising methods under comprehensive quantitative performance measures."
]
} |
1709.00155 | 2752442988 | Generating texts from structured data (e.g., a table) is important for various natural language processing tasks such as question answering and dialog systems. In recent studies, researchers use neural language models and encoder-decoder frameworks for table-to-text generation. However, these neural network-based approaches do not model the order of contents during text generation. When a human writes a summary based on a given table, he or she would probably consider the content order before wording. In a biography, for example, the nationality of a person is typically mentioned before occupation in a biography. In this paper, we propose an order-planning text generation model to capture the relationship between different fields and use such relationship to make the generated text more fluent and smooth. We conducted experiments on the WikiBio dataset and achieve significantly higher performance than previous methods in terms of BLEU, ROUGE, and NIST scores. | Text generation has long aroused interest in the NLP community due to is wide applications including automated navigation @cite_2 and weather forecasting @cite_7 . Traditionally, text generation can be divided into several steps @cite_1 : defines what information should be conveyed in the generated sentence; (2) determines what to generate in each sentence; and (3) actually generates those sentences with words. | {
"cite_N": [
"@cite_1",
"@cite_7",
"@cite_2"
],
"mid": [
"2139079654",
"2119874156",
"1721378200"
],
"abstract": [
"A challenging problem for spoken dialog systems is the design of utterance generation modules that are fast, flexible and general, yet produce high quality output in particular domains. A promising approach is trainable generation, which uses general-purpose linguistic knowledge automatically adapted to the application domain. This paper presents a trainable sentence planner for the MATCH dialog system. We show that trainable sentence planning can produce output comparable to that of MATCH's template-based generator even for quite complex information presentations.",
"One of the main challenges in automatically generating textual weather forecasts is choosing appropriate English words to communicate numeric weather data. A corpus-based analysis of how humans write forecasts showed that there were major differences in how individual writers performed this task, that is, in how they translated data into words. These differences included both different preferences between potential near-synonyms that could be used to express information, and also differences in the meanings that individual writers associated with specific words. Because we thought these differences could confuse readers, we built our SumTime-Mousam weather-forecast generator to use consistent data-to-word rules, which avoided words which were only used by a few people, and words which were interpreted differently by different people. An evaluation by forecast users suggested that they preferred SumTime-Mousam's texts to human-generated texts, in part because of better word choice; this may be the first time that an evaluation has shown that nlg texts are better than human-authored texts.",
"In this paper we tackle the problem of generating natural route descriptions on the basis of input obtained from a commercially available way-finding system. Our framework and architecture incorporates the use of generic natural language generation techniques. Through examples we demonstrate that it is possible to bridge the gap between underlying representation and natural sounding descriptions. The work presented contributes both to the area of natural language generation and to the improvement of way-finding system interfaces."
]
} |
1709.00155 | 2752442988 | Generating texts from structured data (e.g., a table) is important for various natural language processing tasks such as question answering and dialog systems. In recent studies, researchers use neural language models and encoder-decoder frameworks for table-to-text generation. However, these neural network-based approaches do not model the order of contents during text generation. When a human writes a summary based on a given table, he or she would probably consider the content order before wording. In a biography, for example, the nationality of a person is typically mentioned before occupation in a biography. In this paper, we propose an order-planning text generation model to capture the relationship between different fields and use such relationship to make the generated text more fluent and smooth. We conducted experiments on the WikiBio dataset and achieve significantly higher performance than previous methods in terms of BLEU, ROUGE, and NIST scores. | In early years, surface realization is often accomplished by templates @cite_3 or statistically learned (shallow) models, e.g., probabilistic context-free grammar @cite_8 and language models @cite_12 , with hand-crafted features or rules. Therefore, these methods are weak in terms of the quality of generated sentences. For planning, researchers also apply (shallow) machine learning approaches. , for example, model it as a collective classification problem, whereas use a generative semi-Markov model to align text segment and assigned meanings. Generally, planning and realization in the above work are separate and have difficulty in capturing the complexity of language due to the nature of shallow models. | {
"cite_N": [
"@cite_12",
"@cite_3",
"@cite_8"
],
"mid": [
"1521413921",
"2159573465",
"2135363470"
],
"abstract": [
"We present a simple, robust generation system which performs content selection and surface realization in a unified, domain-independent framework. In our approach, we break up the end-to-end generation process into a sequence of local decisions, arranged hierarchically and each trained discriminatively. We deployed our system in three different domains---Robocup sportscasting, technical weather forecasts, and common weather forecasts, obtaining results comparable to state-of-the-art domain-specific systems both in terms of BLEU scores and human evaluation.",
"This article challenges the received wisdom that template-based approaches to the generation of language are necessarily inferior to other approaches as regards their maintainability, linguistic well-foundedness, and quality of output. Some recent NLG systems that call themselves ''template-based'' will illustrate our claims.",
"Two important recent trends in natural language generation are (i) probabilistic techniques and (ii) comprehensive approaches that move away from traditional strictly modular and sequential models. This paper reports experiments in which pCRU a generation framework that combines probabilistic generation methodology with a comprehensive model of the generation space was used to semi-automatically create five different versions of a weather forecast generator. The generators were evaluated in terms of output quality, development time and computational efficiency against (i) human forecasters, (ii) a traditional handcrafted pipelined NLG system and (iii) a HALOGEN-style statistical generator. The most striking result is that despite acquiring all decision-making abilities automatically, the best pCRU generators produce outputs of high enough quality to be scored more highly by human judges than forecasts written by experts."
]
} |
1709.00155 | 2752442988 | Generating texts from structured data (e.g., a table) is important for various natural language processing tasks such as question answering and dialog systems. In recent studies, researchers use neural language models and encoder-decoder frameworks for table-to-text generation. However, these neural network-based approaches do not model the order of contents during text generation. When a human writes a summary based on a given table, he or she would probably consider the content order before wording. In a biography, for example, the nationality of a person is typically mentioned before occupation in a biography. In this paper, we propose an order-planning text generation model to capture the relationship between different fields and use such relationship to make the generated text more fluent and smooth. We conducted experiments on the WikiBio dataset and achieve significantly higher performance than previous methods in terms of BLEU, ROUGE, and NIST scores. | Recently, the recurrent neural network (RNN) is playing a key role in natural language generating. As RNN can automatically capture highly complicated patterns during end-to-end training, it has successful applications including machine translation @cite_9 , dialog systems @cite_11 , and text summarization @cite_10 . | {
"cite_N": [
"@cite_9",
"@cite_10",
"@cite_11"
],
"mid": [
"2133564696",
"2741375528",
"2159640018"
],
"abstract": [
"Neural machine translation is a recently proposed approach to machine translation. Unlike the traditional statistical machine translation, the neural machine translation aims at building a single neural network that can be jointly tuned to maximize the translation performance. The models proposed recently for neural machine translation often belong to a family of encoder-decoders and consists of an encoder that encodes a source sentence into a fixed-length vector from which a decoder generates a translation. In this paper, we conjecture that the use of a fixed-length vector is a bottleneck in improving the performance of this basic encoder-decoder architecture, and propose to extend this by allowing a model to automatically (soft-)search for parts of a source sentence that are relevant to predicting a target word, without having to form these parts as a hard segment explicitly. With this new approach, we achieve a translation performance comparable to the existing state-of-the-art phrase-based system on the task of English-to-French translation. Furthermore, qualitative analysis reveals that the (soft-)alignments found by the model agree well with our intuition.",
"",
"We propose Neural Responding Machine (NRM), a neural network-based response generator for Short-Text Conversation. NRM takes the general encoder-decoder framework: it formalizes the generation of response as a decoding process based on the latent representation of the input text, while both encoding and decoding are realized with recurrent neural networks (RNN). The NRM is trained with a large amount of one-round conversation data collected from a microblogging service. Empirical study shows that NRM can generate grammatically correct and content-wise appropriate responses to over 75 of the input text, outperforming state-of-the-arts in the same setting, including retrieval-based and SMT-based models."
]
} |
1709.00023 | 2753329127 | In recent years researchers have achieved considerable success applying neural network methods to question answering (QA). These approaches have achieved state of the art results in simplified closed-domain settings such as the SQuAD (, 2016) dataset, which provides a pre-selected passage, from which the answer to a given question may be extracted. More recently, researchers have begun to tackle open-domain QA, in which the model is given a question and access to a large corpus (e.g., wikipedia) instead of a pre-selected passage (, 2017a). This setting is more complex as it requires large-scale search for relevant passages by an information retrieval component, combined with a reading comprehension model that "reads" the passages to generate an answer to the question. Performance in this setting lags considerably behind closed-domain performance. In this paper, we present a novel open-domain QA system called Reinforced Ranker-Reader @math , based on two algorithmic innovations. First, we propose a new pipeline for open-domain QA with a Ranker component, which learns to rank retrieved passages in terms of likelihood of generating the ground-truth answer to a given question. Second, we propose a novel method that jointly trains the Ranker along with an answer-generation Reader model, based on reinforcement learning. We report extensive experimental results showing that our method significantly improves on the state of the art for multiple open-domain QA datasets. | Very recently, IR plus machine reading comprehension () showed promise for open-domain QA, especially after datasets created specifically for the multiple-passage RC setting @cite_11 @cite_9 @cite_18 @cite_3 @cite_22 . These datasets deal with the end-to-end open-domain QA setting, where only question-answer pairs provide supervision. Similarly to previous work on open-domain QA, existing deep learning based solutions to the above datasets also rely on a document retrieval module to retrieve a list of passages for RC models to extract answers. Therefore, these approaches suffer from the limitation that the passage ranking scores are determined by n-gram matching (with tf-idf weighting), which is not ideal for QA. | {
"cite_N": [
"@cite_18",
"@cite_22",
"@cite_9",
"@cite_3",
"@cite_11"
],
"mid": [
"2612431505",
"2734823783",
"2962985038",
"2609826708",
"2951534261"
],
"abstract": [
"We present TriviaQA, a challenging reading comprehension dataset containing over 650K question-answer-evidence triples. TriviaQA includes 95K question-answer pairs authored by trivia enthusiasts and independently gathered evidence documents, six per question on average, that provide high quality distant supervision for answering the questions. We show that, in comparison to other recently introduced large-scale datasets, TriviaQA (1) has relatively complex, compositional questions, (2) has considerable syntactic and lexical variability between questions and corresponding answer-evidence sentences, and (3) requires more cross sentence reasoning to find answers. We also present two baseline algorithms: a feature-based classifier and a state-of-the-art neural network, that performs well on SQuAD reading comprehension. Neither approach comes close to human performance (23 and 40 vs. 80 ), suggesting that TriviaQA is a challenging testbed that is worth significant future study. Data and code available at -- this http URL",
"We present two new large-scale datasets aimed at evaluating systems designed to comprehend a natural language query and extract its answer from a large corpus of text. The Quasar-S dataset consists of 37000 cloze-style (fill-in-the-gap) queries constructed from definitions of software entity tags on the popular website Stack Overflow. The posts and comments on the website serve as the background corpus for answering the cloze questions. The Quasar-T dataset consists of 43000 open-domain trivia questions and their answers obtained from various internet sources. ClueWeb09 serves as the background corpus for extracting these answers. We pose these datasets as a challenge for two related subtasks of factoid Question Answering: (1) searching for relevant pieces of text that include the correct answer to a query, and (2) reading the retrieved text to answer the query. We also describe a retrieval system for extracting relevant sentences and documents from the corpus given a query, and include these in the release for researchers wishing to only focus on (2). We evaluate several baselines on both datasets, ranging from simple heuristics to powerful neural models, and show that these lag behind human performance by 16.4 and 32.1 for Quasar-S and -T respectively. The datasets are available at this https URL .",
"",
"We publicly release a new large-scale dataset, called SearchQA, for machine comprehension, or question-answering. Unlike recently released datasets, such as DeepMind CNN DailyMail and SQuAD, the proposed SearchQA was constructed to reflect a full pipeline of general question-answering. That is, we start not from an existing article and generate a question-answer pair, but start from an existing question-answer pair, crawled from J! Archive, and augment it with text snippets retrieved by Google. Following this approach, we built SearchQA, which consists of more than 140k question-answer pairs with each pair having 49.6 snippets on average. Each question-answer-context tuple of the SearchQA comes with additional meta-data such as the snippet's URL, which we believe will be valuable resources for future research. We conduct human evaluation as well as test two baseline methods, one simple word selection and the other deep learning based, on the SearchQA. We show that there is a meaningful gap between the human and machine performances. This suggests that the proposed dataset could well serve as a benchmark for question-answering.",
"We introduce a large scale MAchine Reading COmprehension dataset, which we name MS MARCO. The dataset comprises of 1,010,916 anonymized questions---sampled from Bing's search query logs---each with a human generated answer and 182,669 completely human rewritten generated answers. In addition, the dataset contains 8,841,823 passages---extracted from 3,563,535 web documents retrieved by Bing---that provide the information necessary for curating the natural language answers. A question in the MS MARCO dataset may have multiple answers or no answers at all. Using this dataset, we propose three different tasks with varying levels of difficulty: (i) predict if a question is answerable given a set of context passages, and extract and synthesize the answer as a human would (ii) generate a well-formed answer (if possible) based on the context passages that can be understood with the question and passage context, and finally (iii) rank a set of retrieved passages given a question. The size of the dataset and the fact that the questions are derived from real user search queries distinguishes MS MARCO from other well-known publicly available datasets for machine reading comprehension and question-answering. We believe that the scale and the real-world nature of this dataset makes it attractive for benchmarking machine reading comprehension and question-answering models."
]
} |
1709.00340 | 2909384861 | Fine-grained visual categorization is to recognize hundreds of subcategories belonging to the same basic-level category, which is a highly challenging task due to the quite subtle and local visual distinctions among similar subcategories. Most existing methods generally learn part detectors to discover discriminative regions for better categorization performance. However, not all parts are beneficial and indispensable for visual categorization, and the setting of part detector number heavily relies on prior knowledge as well as experimental validation. As is known to all, when we describe the object of an image via textual descriptions, we mainly focus on the pivotal characteristics, and rarely pay attention to common characteristics as well as the background areas. This is an involuntary transfer from human visual attention to textual attention, which leads to the fact that textual attention tells us how many and which parts are discriminative and significant to categorization. So textual attention could help us to discover visual attention in image. Inspired by this, we propose a fine-grained visual-textual representation learning (VTRL) approach, and its main contributions are: (1) Fine-grained visual-textual pattern mining devotes to discovering discriminative visual-textual pairwise information for boosting categorization performance through jointly modeling vision and text with generative adversarial networks (GANs), which automatically and adaptively discovers discriminative parts. (2) Visual-textual representation learning jointly combines visual and textual information, which preserves the intra-modality and inter-modality information to generate complementary fine-grained representation, as well as further improves categorization performance. | Since the discriminative regions of image is crucial for fine-grained visual categorization, most existing methods @cite_14 @cite_60 first localize the discriminative regions of image, such as the object and its parts, and then extract their discriminative features for fine-grained categorization. Some methods directly use the annotations of the object @cite_45 @cite_46 and parts @cite_3 @cite_33 to localize the discriminative regions. However, it is not available to obtain the annotations in practical applications, some researchers begin to use the annotations of the object and parts only in the training phase. @cite_0 propose the Deformable Part-based Model (DPM) to localize the discriminative regions with the object and part annotations as the supervised information in the training phase. Further more, PG Alignment @cite_65 is proposed to train part detectors only with object annotation, and localize the discriminative parts in an automatic manner in the testing phase. | {
"cite_N": [
"@cite_14",
"@cite_33",
"@cite_60",
"@cite_65",
"@cite_3",
"@cite_0",
"@cite_45",
"@cite_46"
],
"mid": [
"56385144",
"1980526845",
"1928906481",
"1898560071",
"2118696714",
"2079789819",
"2135706578",
"2169501191"
],
"abstract": [
"Semantic part localization can facilitate fine-grained categorization by explicitly isolating subtle appearance differences associated with specific object parts. Methods for pose-normalized representations have been proposed, but generally presume bounding box annotations at test time due to the difficulty of object detection. We propose a model for fine-grained categorization that overcomes these limitations by leveraging deep convolutional features computed on bottom-up region proposals. Our method learns whole-object and part detectors, enforces learned geometric constraints between them, and predicts a fine-grained category from a pose-normalized representation. Experiments on the Caltech-UCSD bird dataset confirm that our method outperforms state-of-the-art fine-grained categorization methods in an end-to-end evaluation without requiring a bounding box at test time.",
"As a special topic in computer vision, fine-grained visual categorization (FGVC) has been attracting growing attention these years. Different with traditional image classification tasks in which objects have large inter-class variation, the visual concepts in the fine-grained datasets, such as hundreds of bird species, often have very similar semantics. Due to the large inter-class similarity, it is very difficult to classify the objects without locating really discriminative features, therefore it becomes more important for the algorithm to make full use of the part information in order to train a robust model. In this paper, we propose a powerful flowchart named Hierarchical Part Matching (HPM) to cope with fine-grained classification tasks. We extend the Bag-of-Features (BoF) model by introducing several novel modules to integrate into image representation, including foreground inference and segmentation, Hierarchical Structure Learning (HSL), and Geometric Phrase Pooling (GPP). We verify in experiments that our algorithm achieves the state-of-the-art classification accuracy in the Caltech-UCSD-Birds-200-2011 dataset by making full use of the ground-truth part annotations.",
"Fine-grained classification is challenging because categories can only be discriminated by subtle and local differences. Variances in the pose, scale or rotation usually make the problem more difficult. Most fine-grained classification systems follow the pipeline of finding foreground object or object parts (where) to extract discriminative features (what).",
"Scaling up fine-grained recognition to all domains of fine-grained objects is a challenge the computer vision community will need to face in order to realize its goal of recognizing all object categories. Current state-of-the-art techniques rely heavily upon the use of keypoint or part annotations, but scaling up to hundreds or thousands of domains renders this annotation cost-prohibitive for all but the most important categories. In this work we propose a method for fine-grained recognition that uses no part annotations. Our method is based on generating parts using co-segmentation and alignment, which we combine in a discriminative mixture. Experimental results show its efficacy, demonstrating state-of-the-art results even when compared to methods that use part annotations during training.",
"From a set of images in a particular domain, labeled with part locations and class, we present a method to automatically learn a large and diverse set of highly discriminative intermediate features that we call Part-based One-vs.-One Features (POOFs). Each of these features specializes in discrimination between two particular classes based on the appearance at a particular part. We demonstrate the particular usefulness of these features for fine-grained visual categorization with new state-of-the-art results on bird species identification using the Caltech UCSD Birds (CUB) dataset and parity with the best existing results in face verification on the Labeled Faces in the Wild (LFW) dataset. Finally, we demonstrate the particular advantage of POOFs when training data is scarce.",
"Recognizing objects in fine-grained domains can be extremely challenging due to the subtle differences between subcategories. Discriminative markings are often highly localized, leading traditional object recognition approaches to struggle with the large pose variation often present in these domains. Pose-normalization seeks to align training exemplars, either piecewise by part or globally for the whole object, effectively factoring out differences in pose and in viewing angle. Prior approaches relied on computationally-expensive filter ensembles for part localization and required extensive supervision. This paper proposes two pose-normalized descriptors based on computationally-efficient deformable part models. The first leverages the semantics inherent in strongly-supervised DPM parts. The second exploits weak semantic annotations to learn cross-component correspondences, computing pose-normalized descriptors from the latent parts of a weakly-supervised DPM. These representations enable pooling across pose and viewpoint, in turn facilitating tasks such as fine-grained recognition and attribute prediction. Experiments conducted on the Caltech-UCSD Birds 200 dataset and Berkeley Human Attribute dataset demonstrate significant improvements of our approach over state-of-art algorithms.",
"We propose a new method for the task of fine-grained visual categorization. The method builds a model of the base-level category that can be fitted to images, producing high-quality foreground segmentation and mid-level part localizations. The model can be learnt from the typical datasets available for fine-grained categorization, where the only annotation provided is a loose bounding box around the instance (e.g. bird) in each image. Both segmentation and part localizations are then used to encode the image content into a highly-discriminative visual signature. The model is symbiotic in that part discovery localization is helped by segmentation and, conversely, the segmentation is helped by the detection (e.g. part layout). Our model builds on top of the part-based object category detector of , and also on the powerful Grab Cut segmentation algorithm of , and adds a simple spatial saliency coupling between them. In our evaluation, the model improves the categorization accuracy over the state-of-the-art. It also improves over what can be achieved with an analogous system that runs segmentation and part-localization independently.",
"Fine-grained recognition refers to a subordinate level of recognition, such as recognizing different species of animals and plants. It differs from recognition of basic categories, such as humans, tables, and computers, in that there are global similarities in shape and structure shared cross different categories, and the differences are in the details of object parts. We suggest that the key to identifying the fine-grained differences lies in finding the right alignment of image regions that contain the same object parts. We propose a template model for the purpose, which captures common shape patterns of object parts, as well as the cooccurrence relation of the shape patterns. Once the image regions are aligned, extracted features are used for classification. Learning of the template model is efficient, and the recognition results we achieve significantly outperform the state-of-the-art algorithms."
]
} |
1709.00340 | 2909384861 | Fine-grained visual categorization is to recognize hundreds of subcategories belonging to the same basic-level category, which is a highly challenging task due to the quite subtle and local visual distinctions among similar subcategories. Most existing methods generally learn part detectors to discover discriminative regions for better categorization performance. However, not all parts are beneficial and indispensable for visual categorization, and the setting of part detector number heavily relies on prior knowledge as well as experimental validation. As is known to all, when we describe the object of an image via textual descriptions, we mainly focus on the pivotal characteristics, and rarely pay attention to common characteristics as well as the background areas. This is an involuntary transfer from human visual attention to textual attention, which leads to the fact that textual attention tells us how many and which parts are discriminative and significant to categorization. So textual attention could help us to discover visual attention in image. Inspired by this, we propose a fine-grained visual-textual representation learning (VTRL) approach, and its main contributions are: (1) Fine-grained visual-textual pattern mining devotes to discovering discriminative visual-textual pairwise information for boosting categorization performance through jointly modeling vision and text with generative adversarial networks (GANs), which automatically and adaptively discovers discriminative parts. (2) Visual-textual representation learning jointly combines visual and textual information, which preserves the intra-modality and inter-modality information to generate complementary fine-grained representation, as well as further improves categorization performance. | Only using object annotation is still not promising in the practical applications. Recently, some works @cite_60 @cite_24 @cite_34 are proposed to localize the discriminative regions in a weakly-supervised manner, which means that neither object nor part annotations are used in both training and testing phases. @cite_60 combine the object and part level attentions to select the discriminative image proposals, which is the first work to localize the discriminative regions without using object and part annotations. @cite_24 also propose to combine the two complementary object-level and part-level visual descriptions for better performance. A neural activation constellation (NAC) part model @cite_19 is proposed to train part detectors with constellation model. He and Peng @cite_56 integrate two spatial constraints to select more discriminative proposals and achieve better categorization accuracy. The aforementioned methods mostly set the detector number due to the prior knowledge or experimental validation, which is highly limited in flexibility and difficult for generalizing to the other domains. Therefore, we attempt to automatically learn how many and which parts really make sense to categorization via fine-grained visual-textual pattern mining. | {
"cite_N": [
"@cite_60",
"@cite_56",
"@cite_24",
"@cite_19",
"@cite_34"
],
"mid": [
"1928906481",
"2604134068",
"2755775504",
"1496650988",
""
],
"abstract": [
"Fine-grained classification is challenging because categories can only be discriminated by subtle and local differences. Variances in the pose, scale or rotation usually make the problem more difficult. Most fine-grained classification systems follow the pipeline of finding foreground object or object parts (where) to extract discriminative features (what).",
"Fine-grained image classification is challenging due to the large intra-class variance and small inter-class variance, aiming at recognizing hundreds of sub-categories belonging to the same basic-level category. Since two different sub-categories is distinguished only by the subtle differences in some specific parts, semantic part localization is crucial for fine-grained image classification. Most previous works improve the accuracy by looking for the semantic parts, but rely heavily upon the use of the object or part annotations of images whose labeling are costly. Recently, some researchers begin to focus on recognizing sub-categories via weakly supervised part detection instead of using the expensive annotations. However, these works ignore the spatial relationship between the object and its parts as well as the interaction of the parts, both of them are helpful to promote part selection. Therefore, this paper proposes a weakly supervised part selection method with spatial constraints for fine-grained image classification, which is free of using any bounding box or part annotations. We first learn a whole-object detector automatically to localize the object through jointly using saliency extraction and co-segmentation. Then two spatial constraints are proposed to select the distinguished parts. The first spatial constraint, called box constraint, defines the relationship between the object and its parts, and aims to ensure that the selected parts are definitely located in the object region, and have the largest overlap with the object region. The second spatial constraint, called parts constraint, defines the relationship of the object's parts, is to reduce the parts' overlap with each other to avoid the information redundancy and ensure the selected parts are the most distinguishing parts from other categories. Combining two spatial constraints promotes parts selection significantly as well as achieves a notable improvement on fine-grained image classification. Experimental results on CUB-200-2011 dataset demonstrate the superiority of our method even compared with those methods using expensive annotations.",
"Compared with traditional image classification, fine-grained visual categorization is a more challenging task, because it targets to classify objects belonging to the same species, e.g. , classify hundreds of birds or cars. In the past several years, researchers have made many achievements on this topic. However, most of them are heavily dependent on the artificial annotations, e.g., bounding boxes, part annotations, and so on . The requirement of artificial annotations largely hinders the scalability and application. Motivated to release such dependence, this paper proposes a robust and discriminative visual description named Automated Bi-level Description (AutoBD). “Bi-level” denotes two complementary part-level and object-level visual descriptions, respectively. AutoBD is “automated,” because it only requires the image-level labels of training images and does not need any annotations for testing images. Compared with the part annotations labeled by the human, the image-level labels can be easily acquired, which thus makes AutoBD suitable for large-scale visual categorization. Specifically, the part-level description is extracted by identifying the local region saliently representing the visual distinctiveness. The object-level description is extracted from object bounding boxes generated with a co-localization algorithm. Although only using the image-level labels, AutoBD outperforms the recent studies on two public benchmark, i.e. , classification accuracy achieves 81.6 on CUB-200–2011 and 88.9 on Car-196, respectively. On the large-scale Birdsnap data set, AutoBD achieves the accuracy of 68 , which is currently the best performance to the best of our knowledge.",
"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, 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.",
""
]
} |
1709.00340 | 2909384861 | Fine-grained visual categorization is to recognize hundreds of subcategories belonging to the same basic-level category, which is a highly challenging task due to the quite subtle and local visual distinctions among similar subcategories. Most existing methods generally learn part detectors to discover discriminative regions for better categorization performance. However, not all parts are beneficial and indispensable for visual categorization, and the setting of part detector number heavily relies on prior knowledge as well as experimental validation. As is known to all, when we describe the object of an image via textual descriptions, we mainly focus on the pivotal characteristics, and rarely pay attention to common characteristics as well as the background areas. This is an involuntary transfer from human visual attention to textual attention, which leads to the fact that textual attention tells us how many and which parts are discriminative and significant to categorization. So textual attention could help us to discover visual attention in image. Inspired by this, we propose a fine-grained visual-textual representation learning (VTRL) approach, and its main contributions are: (1) Fine-grained visual-textual pattern mining devotes to discovering discriminative visual-textual pairwise information for boosting categorization performance through jointly modeling vision and text with generative adversarial networks (GANs), which automatically and adaptively discovers discriminative parts. (2) Visual-textual representation learning jointly combines visual and textual information, which preserves the intra-modality and inter-modality information to generate complementary fine-grained representation, as well as further improves categorization performance. | Frequent patterns are itemsets, subsequences, or substructures that appear in a data set with frequency no less than a user-specified threshold @cite_66 . For example, diaper and beer appear frequently together in sales data of a supermarket, which is a frequent pattern. Frequent pattern mining is first proposed by @cite_18 for market basket analysis. Agrawal and Strikant propose Apriori algorithm @cite_61 to mine frequent patterns in a large transaction database. For textual mining, frequent patterns may be sequential patterns, frequent itemsets, or multiple grams. While for visual mining, frequent patterns may be middle-level feature representation or high-level semantic representation. @cite_57 propose to mine visual patterns using low-level features. @cite_26 propose to combine CNN features and association rule mining for discovering visual patterns. @cite_27 propose a novel multi-modal pattern mining method, which takes textual pattern and visual pattern into consideration at the same time. In this paper, we first utilize Apriori algorithm to discover the textual patterns, and then employ generative adversarial networks (GANs) to mine the relationships between part proposals and textual patterns for better categorization accuracy, which discovers visual and textual patterns at the same time as well as mines the intrinsic correlation between them. | {
"cite_N": [
"@cite_61",
"@cite_18",
"@cite_26",
"@cite_57",
"@cite_27",
"@cite_66"
],
"mid": [
"1484413656",
"2166559705",
"",
"2064853889",
"2526486375",
"2110893883"
],
"abstract": [
"We consider the problem of discovering association rules between items in a large database of sales transactions. We present two new algorithms for solving thii problem that are fundamentally different from the known algorithms. Empirical evaluation shows that these algorithms outperform the known algorithms by factors ranging from three for small problems to more than an order of magnitude for large problems. We also show how the best features of the two proposed algorithms can be combined into a hybrid algorithm, called AprioriHybrid. Scale-up experiments show that AprioriHybrid scales linearly with the number of transactions. AprioriHybrid also has excellent scale-up properties with respect to the transaction size and the number of items in the database.",
"We are given a large database of customer transactions. Each transaction consists of items purchased by a customer in a visit. We present an efficient algorithm that generates all significant association rules between items in the database. The algorithm incorporates buffer management and novel estimation and pruning techniques. We also present results of applying this algorithm to sales data obtained from a large retailing company, which shows the effectiveness of the algorithm.",
"",
"Mining frequent patterns in transaction databases, time-series databases, and many other kinds of databases has been studied popularly in data mining research. Most of the previous studies adopt an Apriori-like candidate set generation-and-test approach. However, candidate set generation is still costly, especially when there exist prolific patterns and or long patterns. In this study, we propose a novel frequent pattern tree (FP-tree) structure, which is an extended prefix-tree structure for storing compressed, crucial information about frequent patterns, and develop an efficient FP-tree-based mining method, FP-growth, for mining the complete set of frequent patterns by pattern fragment growth. Efficiency of mining is achieved with three techniques: (1) a large database is compressed into a highly condensed, much smaller data structure, which avoids costly, repeated database scans, (2) our FP-tree-based mining adopts a pattern fragment growth method to avoid the costly generation of a large number of candidate sets, and (3) a partitioning-based, divide-and-conquer method is used to decompose the mining task into a set of smaller tasks for mining confined patterns in conditional databases, which dramatically reduces the search space. Our performance study shows that the FP-growth method is efficient and scalable for mining both long and short frequent patterns, and is about an order of magnitude faster than the Apriori algorithm and also faster than some recently reported new frequent pattern mining methods.",
"Knowledge bases, which consist of a collection of entities, attributes, and the relations between them are widely used and important for many information retrieval tasks. Knowledge base schemas are often constructed manually using experts with specific domain knowledge for the field of interest. Once the knowledge base is generated then many tasks such as automatic content extraction and knowledge base population can be performed, which have so far been robustly studied by the Natural Language Processing community. However, the current approaches ignore visual information that could be used to build or populate these structured ontologies. Preliminary work on visual knowledge base construction only explores limited basic objects and scene relations. In this paper, we propose a novel multimodal pattern mining approach towards constructing a high-level \"event\" schema semi-automatically, which has the capability to extend text only methods for schema construction. We utilize a large unconstrained corpus of weakly-supervised image-caption pairs related to high-level events such as \"attack\" and \"demonstration\" to both discover visual aspects of an event, and name these visual components automatically. We compare our method with several state-of-the-art visual pattern mining approaches and demonstrate that our proposed method can achieve dramatic improvements in terms of the number of concepts discovered (33 gain), semantic consistence of visual patterns (52 gain), and correctness of pattern naming (150 gain).",
"Frequent pattern mining has been a focused theme in data mining research for over a decade. Abundant literature has been dedicated to this research and tremendous progress has been made, ranging from efficient and scalable algorithms for frequent itemset mining in transaction databases to numerous research frontiers, such as sequential pattern mining, structured pattern mining, correlation mining, associative classification, and frequent pattern-based clustering, as well as their broad applications. In this article, we provide a brief overview of the current status of frequent pattern mining and discuss a few promising research directions. We believe that frequent pattern mining research has substantially broadened the scope of data analysis and will have deep impact on data mining methodologies and applications in the long run. However, there are still some challenging research issues that need to be solved before frequent pattern mining can claim a cornerstone approach in data mining applications."
]
} |
1709.00340 | 2909384861 | Fine-grained visual categorization is to recognize hundreds of subcategories belonging to the same basic-level category, which is a highly challenging task due to the quite subtle and local visual distinctions among similar subcategories. Most existing methods generally learn part detectors to discover discriminative regions for better categorization performance. However, not all parts are beneficial and indispensable for visual categorization, and the setting of part detector number heavily relies on prior knowledge as well as experimental validation. As is known to all, when we describe the object of an image via textual descriptions, we mainly focus on the pivotal characteristics, and rarely pay attention to common characteristics as well as the background areas. This is an involuntary transfer from human visual attention to textual attention, which leads to the fact that textual attention tells us how many and which parts are discriminative and significant to categorization. So textual attention could help us to discover visual attention in image. Inspired by this, we propose a fine-grained visual-textual representation learning (VTRL) approach, and its main contributions are: (1) Fine-grained visual-textual pattern mining devotes to discovering discriminative visual-textual pairwise information for boosting categorization performance through jointly modeling vision and text with generative adversarial networks (GANs), which automatically and adaptively discovers discriminative parts. (2) Visual-textual representation learning jointly combines visual and textual information, which preserves the intra-modality and inter-modality information to generate complementary fine-grained representation, as well as further improves categorization performance. | Nowadays, multi-modal data, e.g. image, text, video and audio, has been widely available on the Internet. They contains different kinds of information, which are complementary to help achieving comprehensive results in many real-world applications. So it is significant to learn multi-modal representation for boosting the signal-modal tasks @cite_4 @cite_29 . Canonical correlation analysis (CCA) @cite_28 is proposed to learn linear projection matrices, which project features of different modalities into the common space and obtain the common representation. It is widely used for modeling multi-modal data @cite_2 @cite_5 @cite_49 . propose the joint representation learning method (JRL) to learn projection matrices considering the semantic and correlation information. Due to the advances of deep learning, deep learning based methods have been proposed to boost the performance of multi-modal representation learning. @cite_6 propose the bimodal autoencoders (Bimodal AE) to model multi-modal data via minimizing the reconstruction error, and learn a shared representation across modalities. | {
"cite_N": [
"@cite_4",
"@cite_28",
"@cite_29",
"@cite_6",
"@cite_49",
"@cite_2",
"@cite_5"
],
"mid": [
"2042708146",
"2025341678",
"2620666060",
"2184188583",
"1957706851",
"1997231916",
"2100235303"
],
"abstract": [
"The explosive growth of Web videos brings out the challenge of how to efficiently browse hundreds or even thousands of videos at a glance. Given an event-driven query, social media Web sites usually return a large number of videos that are diverse and noisy in a ranking list. Exploring such results will be time-consuming and thus degrades user experience. This article presents a novel scheme that is able to summarize the content of video search results by mining and threading “key” shots, such that users can get an overview of main content of these videos at a glance. The proposed framework mainly comprises four stages. First, given an event query, a set of Web videos is collected associated with their ranking order and tags. Second, key-shots are established and ranked based on near-duplicate keyframe detection and they are threaded in a chronological order. Third, we analyze the tags associated with key-shots. Irrelevant tags are filtered out via a representativeness and descriptiveness analysis, whereas the remaining tags are propagated among key-shots by random walk. Finally, summarization is formulated as an optimization framework that compromises relevance of key-shots and user-defined skimming ratio. We provide two types of summarization: video skimming and visual-textual storyboard. We conduct user studies on twenty event queries for over hundred hours of videos crawled from YouTube. The evaluation demonstrates the feasibility and effectiveness of the proposed solution.",
"Concepts of correlation and regression may be applied not only to ordinary one-dimensional variates but also to variates of two or more dimensions. Marksmen side by side firing simultaneous shots at targets, so that the deviations are in part due to independent individual errors and in part to common causes such as wind, provide a familiar introduction to the theory of correlation; but only the correlation of the horizontal components is ordinarily discussed, whereas the complex consisting of horizontal and vertical deviations may be even more interesting. The wind at two places may be compared, using both components of the velocity in each place. A fluctuating vector is thus matched at each moment with another fluctuating vector. The study of individual differences in mental and physical traits calls for a detailed study of the relations between sets of correlated variates. For example the scores on a number of mental tests may be compared with physical measurements on the same persons. The questions then arise of determining the number and nature of the independent relations of mind and body shown by these data to exist, and of extracting from the multiplicity of correlations in the system suitable characterizations of these independent relations. As another example, the inheritance of intelligence in rats might be studied by applying not one but s different mental tests to N mothers and to a daughter of each",
"The rapidly increasing number of images on the internet has further increased the need for efficient indexing for digital image searching of large databases. The design of a cloud service that provides high efficiency but compact image indexing remains challenging, partly due to the well-known semantic gap between user queries and the rich semantics of large-scale data sets. In this paper, we construct a novel joint semantic-visual space by leveraging visual descriptors and semantic attributes, which narrows the semantic gap by combining both attributes and indexing into a single framework. Such a joint space embraces the flexibility of coherent semantic-visual indexing, which employs binary codes to boost retrieval speed while maintaining accuracy. To solve the proposed model, we make the following contributions. First, we propose an interactive optimization method to find the joint semantic and visual descriptor space. Second, we prove convergence of our optimization algorithm, which guarantees a good solution after a certain number of iterations. Third, we integrate the semantic-visual joint space system with spectral hashing, which finds an efficient solution to search up to billion-scale data sets. Finally, we design an online cloud service to provide a more efficient online multimedia service. Experiments on two standard retrieval datasets ( i.e., Holidays1M, Oxford5K ) show that the proposed method is promising compared with the current state-of-the-art and that the cloud system significantly improves performance.",
"Deep networks have been successfully applied to unsupervised feature learning for single modalities (e.g., text, images or audio). In this work, we propose a novel application of deep networks to learn features over multiple modalities. We present a series of tasks for multimodal learning and show how to train deep networks that learn features to address these tasks. In particular, we demonstrate cross modality feature learning, where better features for one modality (e.g., video) can be learned if multiple modalities (e.g., audio and video) are present at feature learning time. Furthermore, we show how to learn a shared representation between modalities and evaluate it on a unique task, where the classifier is trained with audio-only data but tested with video-only data and vice-versa. Our models are validated on the CUAVE and AVLetters datasets on audio-visual speech classification, demonstrating best published visual speech classification on AVLetters and effective shared representation learning.",
"In recent years, the problem of associating a sentence with an image has gained a lot of attention. This work continues to push the envelope and makes further progress in the performance of image annotation and image search by a sentence tasks. In this work, we are using the Fisher Vector as a sentence representation by pooling the word2vec embedding of each word in the sentence. The Fisher Vector is typically taken as the gradients of the log-likelihood of descriptors, with respect to the parameters of a Gaussian Mixture Model (GMM). In this work we present two other Mixture Models and derive their Expectation-Maximization and Fisher Vector expressions. The first is a Laplacian Mixture Model (LMM), which is based on the Laplacian distribution. The second Mixture Model presented is a Hybrid Gaussian-Laplacian Mixture Model (HGLMM) which is based on a weighted geometric mean of the Gaussian and Laplacian distribution. Finally, by using the new Fisher Vectors derived from HGLMMs to represent sentences, we achieve state-of-the-art results for both the image annotation and the image search by a sentence tasks on four benchmarks: Pascal1K, Flickr8K, Flickr30K, and COCO.",
"We investigate the use of audio-visual speech synchrony measure in the framework of identity verification based on talking faces. Two synchrony measures based on canonical correlation analysis and co-inertia analysis respectively are introduced and their performances are evaluated on the specific task of detecting synchronized and not-synchronized audio-visual speech sequences. The notion of high-effort impostor attacks is also introduced as a dangerous threat for current biometric system based on speaker verification and face recognition. A novel biometric modality based on synchrony measures is introduced in order to improve the overall performance of identity verification, and more specifically its robustness to replay attacks.",
"We present a general method using kernel canonical correlation analysis to learn a semantic representation to web images and their associated text. The semantic space provides a common representation and enables a comparison between the text and images. In the experiments, we look at two approaches of retrieving images based on only their content from a text query. We compare orthogonalization approaches against a standard cross-representation retrieval technique known as the generalized vector space model."
]
} |
1708.09789 | 2741278639 | Informal first-person narratives are a unique resource for computational models of everyday events and people's affective reactions to them. People blogging about their day tend not to explicitly say I am happy. Instead they describe situations from which other humans can readily infer their affective reactions. However current sentiment dictionaries are missing much of the information needed to make similar inferences. We build on recent work that models affect in terms of lexical predicate functions and affect on the predicate's arguments. We present a method to learn proxies for these functions from first-person narratives. We construct a novel fine-grained test set, and show that the patterns we learn improve our ability to predict first-person affective reactions to everyday events, from a Stanford sentiment baseline of .67F to .75F. | use the narratives produced by the ISEAR questionnaire @cite_14 for first-person examples of particular emotions and extract sequences of subject-verb-object triples, which they annotate for basic emotions. | {
"cite_N": [
"@cite_14"
],
"mid": [
"2136201510"
],
"abstract": [
"We present a weakly supervised approach for learning hashtags, hashtag patterns, and phrases associated with five emotions: AFFECTION, ANGER RAGE, FEAR ANXIETY, JOY, and SADNESS DISAPPOINTMENT. Starting with seed hashtags to label an initial set of tweets, we train emotion classifiers and use them to learn new emotion hashtags and hashtag patterns. This process then repeats in a bootstrapping framework. Emotion phrases are also extracted from the learned hashtags and used to create phrase-based emotion classifiers. We show that the learned set of emotion indicators yields a substantial improvement in F-scores, ranging from + 5 to + 18 over baseline classifiers."
]
} |
1708.09496 | 2751501689 | To understand narrative, humans draw inferences about the underlying relations between narrative events. Cognitive theories of narrative understanding define these inferences as four different types of causality, that include pairs of events A, B where A physically causes B (X drop, X break), to pairs of events where A causes emotional state B (Y saw X, Y felt fear). Previous work on learning narrative relations from text has either focused on "strict" physical causality, or has been vague about what relation is being learned. This paper learns pairs of causal events from a corpus of film scene descriptions which are action rich and tend to be told in chronological order. We show that event pairs induced using our methods are of high quality and are judged to have a stronger causal relation than event pairs from Rel-grams. | used a minimally supervised approach, based on focused distributional similarity methods and discourse connectives, to identify causality relations between events in PDTB in context (both verbs and nouns) @cite_14 . They present a detailed formula for calculating contingency causality that takes into account several different kinds of argument overlap between adjacent events. However they do not provide any evidence that all the components of this formula actually contribute to their results. | {
"cite_N": [
"@cite_14"
],
"mid": [
"2166957049"
],
"abstract": [
"We present the second version of the Penn Discourse Treebank, PDTB-2.0, describing its lexically-grounded annotations of discourse relations and their two abstract object arguments over the 1 million word Wall Street Journal corpus. We describe all aspects of the annotation, including (a) the argument structure of discourse relations, (b) the sense annotation of the relations, and (c) the attribution of discourse relations and each of their arguments. We list the differences between PDTB-1.0 and PDTB-2.0. We present representative statistics for several aspects of the annotation in the corpus."
]
} |
1708.09496 | 2751501689 | To understand narrative, humans draw inferences about the underlying relations between narrative events. Cognitive theories of narrative understanding define these inferences as four different types of causality, that include pairs of events A, B where A physically causes B (X drop, X break), to pairs of events where A causes emotional state B (Y saw X, Y felt fear). Previous work on learning narrative relations from text has either focused on "strict" physical causality, or has been vague about what relation is being learned. This paper learns pairs of causal events from a corpus of film scene descriptions which are action rich and tend to be told in chronological order. We show that event pairs induced using our methods are of high quality and are judged to have a stronger causal relation than event pairs from Rel-grams. | used event ngrams and discourse cues to learn causal relations from first person stories posted on weblogs and evaluated them with respect to the COPA SEM-EVAL task. Other related work learns likely sequences of temporally ordered events but does not explicitly model causality @cite_20 @cite_12 @cite_1 . | {
"cite_N": [
"@cite_1",
"@cite_12",
"@cite_20"
],
"mid": [
"1846895676",
"2252139350",
"2158794898"
],
"abstract": [
"One of the central problems in building broad-coverage story understanding systems is generating expectations about event sequences, i.e. predicting what happens next given some arbitrary narrative context. In this paper, we describe how a large corpus of stories extracted from Internet weblogs was used to learn a probabilistic model of event sequences using statistical language modeling techniques. Our approach was to encode weblog stories as sequences of events, one per sentence in the story, where each event was represented as a pair of descriptive key words extracted from the sentence. We then applied statistical language modeling techniques to each of the event sequences in the corpus. We evaluated the utility of the resulting model for the tasks of narrative event ordering and event prediction.",
"Chambers and Jurafsky (2009) demonstrated that event schemas can be automatically induced from text corpora. However, our analysis of their schemas identifies several weaknesses, e.g., some schemas lack a common topic and distinct roles are incorrectly mixed into a single actor. It is due in part to their pair-wise representation that treats subjectverb independently from verb-object. This often leads to subject-verb-object triples that are not meaningful in the real-world. We present a novel approach to inducing open-domain event schemas that overcomes these limitations. Our approach uses cooccurrence statistics of semantically typed relational triples, which we call Rel-grams (relational n-grams). In a human evaluation, our schemas outperform Chambers’s schemas by wide margins on several evaluation criteria. Both Rel-grams and event schemas are freely available to the research community.",
"We describe an unsupervised system for learning narrative schemas, coherent sequences or sets of events (arrested(POLICE, SUSPECT), convicted(JUDGE, SUSPECT)) whose arguments are filled with participant semantic roles defined over words (Judge = judge, jury, court , Police = police, agent, authorities ). Unlike most previous work in event structure or semantic role learning, our system does not use supervised techniques, hand-built knowledge, or predefined classes of events or roles. Our unsupervised learning algorithm uses coreferring arguments in chains of verbs to learn both rich narrative event structure and argument roles. By jointly addressing both tasks, we improve on previous results in narrative frame learning and induce rich frame-specific semantic roles."
]
} |
1708.09561 | 2949121352 | Distributed cloud networking enables the deployment of a wide range of services in the form of interconnected software functions instantiated over general purpose hardware at multiple cloud locations distributed throughout the network. We consider the problem of optimal service delivery over a distributed cloud network, in which nodes are equipped with both communication and computation resources. We address the design of distributed online solutions that drive flow processing and routing decisions, along with the associated allocation of cloud and network resources. For a given set of services, each described by a chain of service functions, we characterize the cloud network capacity region and design a family of dynamic cloud network control (DCNC) algorithms that stabilize the underlying queuing system, while achieving arbitrarily close to minimum cost with a tradeoff in network delay. The proposed DCNC algorithms make local decisions based on the online minimization of linear and quadratic metrics obtained from an upper bound on the Lyapunov drift-plus-penalty of the cloud network queuing system. Minimizing a quadratic vs. a linear metric is shown to improve the cost-delay tradeoff at the expense of increased computational complexity. Our algorithms are further enhanced with a shortest transmission-plus-processing distance bias that improves delay performance without compromising throughput or overall cloud network cost. We provide throughput and cost optimality guarantees, convergence time analysis, and extensive simulations in representative cloud network scenarios. | LDP control strategies have shown effective in optimizing traditional multi-hop communication networks (as opposed to computation networks). Different versions of LDP-based algorithms have been developed. Most of them are based on the minimization of a linear metric obtained from an upper bound expression of the queueing system LDP function @cite_2 - @cite_12 . Subsequently, the inclusion of a bias term, indicative of network distance, into this linear metric was shown to reduce network delay (especially in low congested scenarios) @cite_11 , @cite_14 . Furthermore, @cite_13 proposed a control algorithm for single-commodity multi-hop networks based on the minimization of a quadratic metric from the LDP upper bound, shown to improve delay performance in the scenarios explored in @cite_13 . | {
"cite_N": [
"@cite_14",
"@cite_2",
"@cite_13",
"@cite_12",
"@cite_11"
],
"mid": [
"2070317877",
"2138993731",
"2091511579",
"",
"2086002250"
],
"abstract": [
"We consider dynamic routing and power allocation for a wireless network with time-varying channels. The network consists of power constrained nodes that transmit over wireless links with adaptive transmission rates. Packets randomly enter the system at each node and wait in output queues to be transmitted through the network to their destinations. We establish the capacity region of all rate matrices ( spl lambda sub ij ) that the system can stably support-where spl lambda sub ij represents the rate of traffic originating at node i and destined for node j. A joint routing and power allocation policy is developed that stabilizes the system and provides bounded average delay guarantees whenever the input rates are within this capacity region. Such performance holds for general arrival and channel state processes, even if these processes are unknown to the network controller. We then apply this control algorithm to an ad hoc wireless network, where channel variations are due to user mobility. Centralized and decentralized implementations are compared, and the stability region of the decentralized algorithm is shown to contain that of the mobile relay strategy developed by Grossglauser and Tse (2002).",
"Information flow in a telecommunication network is accomplished through the interaction of mechanisms at various design layers with the end goal of supporting the information exchange needs of the applications. In wireless networks in particular, the different layers interact in a nontrivial manner in order to support information transfer. In this text we will present abstract models that capture the cross-layer interaction from the physical to transport layer in wireless network architectures including cellular, ad-hoc and sensor networks as well as hybrid wireless-wireline. The model allows for arbitrary network topologies as well as traffic forwarding modes, including datagrams and virtual circuits. Furthermore the time varying nature of a wireless network, due either to fading channels or to changing connectivity due to mobility, is adequately captured in our model to allow for state dependent network control policies. Quantitative performance measures that capture the quality of service requirements in these systems depending on the supported applications are discussed, including throughput maximization, energy consumption minimization, rate utility function maximization as well as general performance functionals. Cross-layer control algorithms with optimal or suboptimal performance with respect to the above measures are presented and analyzed. A detailed exposition of the related analysis and design techniques is provided.",
"An information collection problem in a wireless network with random events is considered. Wireless devices report on each event using one of multiple reporting formats. Each format has a different quality and uses different data lengths. Delivering all data in the highest-quality format can overload system resources. The goal is to make intelligent format selection and routing decisions to maximize time-averaged information quality subject to network stability. Lyapunov optimization theory can be used to solve such a problem by repeatedly minimizing the linear terms of a quadratic drift-plus-penalty expression. To reduce delays, this paper proposes a novel extension of this technique that preserves the quadratic nature of the drift minimization while maintaining a fully separable structure. In addition, to avoid high queuing delay, paths are restricted to at most 2 hops. The resulting algorithm can push average information quality arbitrarily close to optimum, with a tradeoff in queue backlog. The algorithm compares favorably to the basic drift-plus-penalty scheme in terms of backlog and delay.",
"",
"We consider the problem of optimal scheduling and routing in an ad-hoc wireless network with multiple traffic streams and time varying channel reliability. Each packet transmission can be overheard by a subset of receiver nodes, with a transmission success probability that may vary from receiver to receiver and may also vary with time. We develop a simple backpressure routing algorithm that maximizes network throughput and expends an average power that can be pushed arbitrarily close to the minimum average power required for network stability, with a corresponding tradeoff in network delay. When channels are orthogonal, the algorithm can be implemented in a distributed manner using only local link error probability information, and supports a ''blind transmission'' mode (where error probabilities are not required) in special cases when the power metric is neglected and when there is only a single destination for all traffic streams. For networks with general inter-channel interference, we present a distributed algorithm with constant-factor optimality guarantees."
]
} |
1708.09398 | 2752487800 | In 1969, Strassen shocked the world by showing that two n x n matrices could be multiplied in time asymptotically less than @math . While the recursive construction in his algorithm is very clear, the key gain was made by showing that 2 x 2 matrix multiplication could be performed with only 7 multiplications instead of 8. The latter construction was arrived at by a process of elimination and appears to come out of thin air. Here, we give the simplest and most transparent proof of Strassen's algorithm that we are aware of, using only a simple unitary 2-design and a few easy lines of calculation. Moreover, using basic facts from the representation theory of finite groups, we use 2-designs coming from group orbits to generalize our construction to all n (although the resulting algorithms aren't optimal for n at least 3). | Although the method of construction suggested in @cite_5 , and independently in @cite_11 , is more general than this, the constructions we ended up finding in @cite_5 were in fact all instances of a single design-based construction yielding @math multiplications for @math matrix multiplication. The proof that this construction works is the simplest and most transparent proof of Strassen's algorithm that we are aware of. | {
"cite_N": [
"@cite_5",
"@cite_11"
],
"mid": [
"2576976944",
"2950875855"
],
"abstract": [
"We show how to construct highly symmetric algorithms for matrix multiplication. In particular, we consider algorithms which decompose the matrix multiplication tensor into a sum of rank-1 tensors, where the decomposition itself consists of orbits under some finite group action. We show how to use the representation theory of the corresponding group to derive simple constraints on the decomposition, which we solve by hand for n=2,3,4,5, recovering Strassen's algorithm (in a particularly symmetric form) and new algorithms for larger n. While these new algorithms do not improve the known upper bounds on tensor rank or the matrix multiplication exponent, they are beautiful in their own right, and we point out modifications of this idea that could plausibly lead to further improvements. Our constructions also suggest further patterns that could be mined for new algorithms, including a tantalizing connection with lattices. In particular, using lattices we give the most transparent proof to date of Strassen's algorithm; the same proof works for all n, to yield a decomposition with @math terms.",
"This is the first in a series of papers on rank decompositions of the matrix multiplication tensor. In this paper we: establish general facts about rank decompositions of tensors, describe potential ways to search for new matrix multiplication decompositions, give a geometric proof of the theorem of Burichenko's theorem establishing the symmetry group of Strassen's algorithm, and present two particularly nice subfamilies in the Strassen family of decompositions."
]
} |
1708.09398 | 2752487800 | In 1969, Strassen shocked the world by showing that two n x n matrices could be multiplied in time asymptotically less than @math . While the recursive construction in his algorithm is very clear, the key gain was made by showing that 2 x 2 matrix multiplication could be performed with only 7 multiplications instead of 8. The latter construction was arrived at by a process of elimination and appears to come out of thin air. Here, we give the simplest and most transparent proof of Strassen's algorithm that we are aware of, using only a simple unitary 2-design and a few easy lines of calculation. Moreover, using basic facts from the representation theory of finite groups, we use 2-designs coming from group orbits to generalize our construction to all n (although the resulting algorithms aren't optimal for n at least 3). | One may also reasonably wonder whether there is any relationship between our group-based construction and the family of group-based constructions suggested by Cohn and Umans @cite_16 , including the constructions given in @cite_9 and generalizations in @cite_4 . While there may be a common generalization that captures both methods, at the moment we don't know of any direct relationship between the two. Indeed, one cannot use the group-theoretic approach of @cite_16 to explain Strassen's result, even though the constructions of @cite_9 get a better exponent: The only way to use their approach for the @math case is to embed @math into the cyclic group @math , but Cohn and Umans showed that one could not beat @math using only abelian groups. (Some of their more complicated constructions can beat @math in abelian groups, but those involve embedding multiple copies of @math into the same group simultaneously, whereas here we are explicitly talking about embedding a single copy of @math .) | {
"cite_N": [
"@cite_9",
"@cite_16",
"@cite_4"
],
"mid": [
"",
"2111612671",
"2951997246"
],
"abstract": [
"",
"We develop a new, group-theoretic approach to bounding the exponent of matrix multiplication. There are two components to this approach: (1) identifying groups G that admit a certain type of embedding of matrix multiplication into the group algebra spl Copf [G], and (2) controlling the dimensions of the irreducible representations of such groups. We present machinery and examples to support (1), including a proof that certain families of groups of order n sup 2+o(1) support n spl times n matrix multiplication, a necessary condition for the approach to yield exponent 2. Although we cannot yet completely achieve both (1) and (2), we hope that it may be possible, and we suggest potential routes to that result using the constructions in this paper.",
"We introduce a relaxation of the notion of tensor rank, called s-rank, and show that upper bounds on the s-rank of the matrix multiplication tensor imply upper bounds on the ordinary rank. In particular, if the \"s-rank exponent of matrix multiplication\" equals 2, then omega = 2. This connection between the s-rank exponent and the ordinary exponent enables us to significantly generalize the group-theoretic approach of Cohn and Umans, from group algebras to general algebras. Embedding matrix multiplication into general algebra multiplication yields bounds on s-rank (not ordinary rank) and, prior to this paper, that had been a barrier to working with general algebras. We identify adjacency algebras of coherent configurations as a promising family of algebras in the generalized framework. Coherent configurations are combinatorial objects that generalize groups and group actions; adjacency algebras are the analogue of group algebras and retain many of their important features. As with groups, coherent configurations support matrix multiplication when a natural combinatorial condition is satisfied, involving triangles of points in their underlying geometry. Finally, we prove a closure property involving symmetric powers of adjacency algebras, which enables us to prove nontrivial bounds on omega using commutative coherent configurations and suggests that commutative coherent configurations may be sufficient to prove omega = 2. Altogether, our results show that bounds on omega can be established by embedding large matrix multiplication instances into small commutative coherent configurations, while avoiding the representation-theoretic complications that were present in the group-theoretic approach."
]
} |
1708.09522 | 2752655580 | Inspired by recent advances in neural machine translation, that jointly align and translate using encoder-decoder networks equipped with attention, we propose an attentionbased LSTM model for human activity recognition. Our model jointly learns to classify actions and highlight frames associated with the action, by attending to salient visual information through a jointly learned soft-attention networks. We explore attention informed by various forms of visual semantic features, including those encoding actions, objects and scenes. We qualitatively show that soft-attention can learn to effectively attend to important objects and scene information correlated with specific human actions. Further, we show that, quantitatively, our attention-based LSTM outperforms the vanilla LSTM and CNN models used by stateof-the-art methods. On a large-scale youtube video dataset, ActivityNet, our model outperforms competing methods in action classification. | Traditional action recognition: Traditional action recognition approaches largely focused on global video representations that achieved good results on smaller and simpler datasets, , KTH @cite_0 , HMDB51 @cite_51 and UCF101 @cite_23 . Such approaches focused on pach-based local motion and appearance information in the form of Histogram of Oriented Gradients (HOG), Histogram of Optical Flow (HOF) @cite_24 , Motion Boundary Histogram (MBH) @cite_38 or dense-trajectories @cite_31 @cite_34 . Classification was typically achieved by further aggregating these local representations across videos or using spatio-temporal pyramids @cite_25 with Bag-of-Words(BoW) @cite_24 or Fisher vector based encodings @cite_34 followed by traditional classifiers ( , SVMs). | {
"cite_N": [
"@cite_38",
"@cite_0",
"@cite_24",
"@cite_23",
"@cite_31",
"@cite_34",
"@cite_51",
"@cite_25"
],
"mid": [
"2162670331",
"2034328688",
"2142194269",
"24089286",
"2126574503",
"2105101328",
"2126579184",
"2162915993"
],
"abstract": [
"We present a discriminative part-based approach for human action recognition from video sequences using motion features. Our model is based on the recently proposed hidden conditional random field (HCRF) for object recognition. Similarly to HCRF for object recognition, we model a human action by a flexible constellation of parts conditioned on image observations. Differently from object recognition, our model combines both large-scale global features and local patch features to distinguish various actions. Our experimental results show that our model is comparable to other state-of-the-art approaches in action recognition. In particular, our experimental results demonstrate that combining large-scale global features and local patch features performs significantly better than directly applying HCRF on local patches alone. We also propose an alternative for learning the parameters of an HCRF model in a max-margin framework. We call this method the max-margin hidden conditional random field (MMHCRF). We demonstrate that MMHCRF outperforms HCRF in human action recognition. In addition, MMHCRF can handle a much broader range of complex hidden structures arising in various problems in computer vision.",
"Local space-time features capture local events in video and can be adapted to the size, the frequency and the velocity of moving patterns. In this paper, we demonstrate how such features can be used for recognizing complex motion patterns. We construct video representations in terms of local space-time features and integrate such representations with SVM classification schemes for recognition. For the purpose of evaluation we introduce a new video database containing 2391 sequences of six human actions performed by 25 people in four different scenarios. The presented results of action recognition justify the proposed method and demonstrate its advantage compared to other relative approaches for action recognition.",
"The aim of this paper is to address recognition of natural human actions in diverse and realistic video settings. This challenging but important subject has mostly been ignored in the past due to several problems one of which is the lack of realistic and annotated video datasets. Our first contribution is to address this limitation and to investigate the use of movie scripts for automatic annotation of human actions in videos. We evaluate alternative methods for action retrieval from scripts and show benefits of a text-based classifier. Using the retrieved action samples for visual learning, we next turn to the problem of action classification in video. We present a new method for video classification that builds upon and extends several recent ideas including local space-time features, space-time pyramids and multi-channel non-linear SVMs. The method is shown to improve state-of-the-art results on the standard KTH action dataset by achieving 91.8 accuracy. Given the inherent problem of noisy labels in automatic annotation, we particularly investigate and show high tolerance of our method to annotation errors in the training set. We finally apply the method to learning and classifying challenging action classes in movies and show promising results.",
"We introduce UCF101 which is currently the largest dataset of human actions. It consists of 101 action classes, over 13k clips and 27 hours of video data. The database consists of realistic user uploaded videos containing camera motion and cluttered background. Additionally, we provide baseline action recognition results on this new dataset using standard bag of words approach with overall performance of 44.5 . To the best of our knowledge, UCF101 is currently the most challenging dataset of actions due to its large number of classes, large number of clips and also unconstrained nature of such clips.",
"Feature trajectories have shown to be efficient for representing videos. Typically, they are extracted using the KLT tracker or matching SIFT descriptors between frames. However, the quality as well as quantity of these trajectories is often not sufficient. Inspired by the recent success of dense sampling in image classification, we propose an approach to describe videos by dense trajectories. We sample dense points from each frame and track them based on displacement information from a dense optical flow field. Given a state-of-the-art optical flow algorithm, our trajectories are robust to fast irregular motions as well as shot boundaries. Additionally, dense trajectories cover the motion information in videos well. We, also, investigate how to design descriptors to encode the trajectory information. We introduce a novel descriptor based on motion boundary histograms, which is robust to camera motion. This descriptor consistently outperforms other state-of-the-art descriptors, in particular in uncontrolled realistic videos. We evaluate our video description in the context of action classification with a bag-of-features approach. Experimental results show a significant improvement over the state of the art on four datasets of varying difficulty, i.e. KTH, YouTube, Hollywood2 and UCF sports.",
"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.",
"With nearly one billion online videos viewed everyday, an emerging new frontier in computer vision research is recognition and search in video. While much effort has been devoted to the collection and annotation of large scalable static image datasets containing thousands of image categories, human action datasets lag far behind. Current action recognition databases contain on the order of ten different action categories collected under fairly controlled conditions. State-of-the-art performance on these datasets is now near ceiling and thus there is a need for the design and creation of new benchmarks. To address this issue we collected the largest action video database to-date with 51 action categories, which in total contain around 7,000 manually annotated clips extracted from a variety of sources ranging from digitized movies to YouTube. We use this database to evaluate the performance of two representative computer vision systems for action recognition and explore the robustness of these methods under various conditions such as camera motion, viewpoint, video quality and occlusion.",
"This paper presents a method for recognizing scene categories based on approximate global geometric correspondence. This technique works by partitioning the image into increasingly fine sub-regions and computing histograms of local features found inside each sub-region. The resulting \"spatial pyramid\" is a simple and computationally efficient extension of an orderless bag-of-features image representation, and it shows significantly improved performance on challenging scene categorization tasks. Specifically, our proposed method exceeds the state of the art on the Caltech-101 database and achieves high accuracy on a large database of fifteen natural scene categories. The spatial pyramid framework also offers insights into the success of several recently proposed image descriptions, including Torralbas \"gist\" and Lowes SIFT descriptors."
]
} |
1708.09522 | 2752655580 | Inspired by recent advances in neural machine translation, that jointly align and translate using encoder-decoder networks equipped with attention, we propose an attentionbased LSTM model for human activity recognition. Our model jointly learns to classify actions and highlight frames associated with the action, by attending to salient visual information through a jointly learned soft-attention networks. We explore attention informed by various forms of visual semantic features, including those encoding actions, objects and scenes. We qualitatively show that soft-attention can learn to effectively attend to important objects and scene information correlated with specific human actions. Further, we show that, quantitatively, our attention-based LSTM outperforms the vanilla LSTM and CNN models used by stateof-the-art methods. On a large-scale youtube video dataset, ActivityNet, our model outperforms competing methods in action classification. | Temporal models and LSTM: The core idea that temporal progression of an activity is important, has a long history in the field. A variety of early models have been utilized for modeling of temporal context, including Hidden semi-Markov Models (HSMM) @cite_45 , CRFs @cite_5 , and finite-state-machines @cite_32 . Most recently, RNNs have also been explored; LSTM, a form of RNN, has recently become particularly popular due to its ability to deal with vanishing and exploding gradients during training. RNN LSTM have a number of notable benefits with respect to, for example, HMMs or CRFs, including fewer conditional independence assumptions and possible end-to-end training when combined with CNN. Several recent works incorporate spatial optical-flow CNN features with vanilla LSTM models for global temporal modeling of videos @cite_12 @cite_6 @cite_27 @cite_1 @cite_42 . These works showed improvements with respect to action recognition @cite_6 @cite_25 @cite_1 or video description tasks @cite_29 @cite_42 . Similar to these methods, we also incorporate CNN features with LSTM for action recognition. However, we use a new variant of LSTM model equipped with an attention network that allows it to focus and highlight discriminative frames and use object and scene features as context. | {
"cite_N": [
"@cite_29",
"@cite_42",
"@cite_1",
"@cite_32",
"@cite_6",
"@cite_27",
"@cite_45",
"@cite_5",
"@cite_25",
"@cite_12"
],
"mid": [
"2950307714",
"2950019618",
"2161565164",
"2156342794",
"2951183276",
"",
"2161029186",
"2036502167",
"2162915993",
"2952453038"
],
"abstract": [
"Recent progress in using recurrent neural networks (RNNs) for image description has motivated the exploration of their application for video description. However, while images are static, working with videos requires modeling their dynamic temporal structure and then properly integrating that information into a natural language description. In this context, we propose an approach that successfully takes into account both the local and global temporal structure of videos to produce descriptions. First, our approach incorporates a spatial temporal 3-D convolutional neural network (3-D CNN) representation of the short temporal dynamics. The 3-D CNN representation is trained on video action recognition tasks, so as to produce a representation that is tuned to human motion and behavior. Second we propose a temporal attention mechanism that allows to go beyond local temporal modeling and learns to automatically select the most relevant temporal segments given the text-generating RNN. Our approach exceeds the current state-of-art for both BLEU and METEOR metrics on the Youtube2Text dataset. We also present results on a new, larger and more challenging dataset of paired video and natural language descriptions.",
"Real-world videos often have complex dynamics; and methods for generating open-domain video descriptions should be sensitive to temporal structure and allow both input (sequence of frames) and output (sequence of words) of variable length. To approach this problem, we propose a novel end-to-end sequence-to-sequence model to generate captions for videos. For this we exploit recurrent neural networks, specifically LSTMs, which have demonstrated state-of-the-art performance in image caption generation. Our LSTM model is trained on video-sentence pairs and learns to associate a sequence of video frames to a sequence of words in order to generate a description of the event in the video clip. Our model naturally is able to learn the temporal structure of the sequence of frames as well as the sequence model of the generated sentences, i.e. a language model. We evaluate several variants of our model that exploit different visual features on a standard set of YouTube videos and two movie description datasets (M-VAD and MPII-MD).",
"Classifying videos according to content semantics is an important problem with a wide range of applications. In this paper, we propose a hybrid deep learning framework for video classification, which is able to model static spatial information, short-term motion, as well as long-term temporal clues in the videos. Specifically, the spatial and the short-term motion features are extracted separately by two Convolutional Neural Networks (CNN). These two types of CNN-based features are then combined in a regularized feature fusion network for classification, which is able to learn and utilize feature relationships for improved performance. In addition, Long Short Term Memory (LSTM) networks are applied on top of the two features to further model longer-term temporal clues. The main contribution of this work is the hybrid learning framework that can model several important aspects of the video data. We also show that (1) combining the spatial and the short-term motion features in the regularized fusion network is better than direct classification and fusion using the CNN with a softmax layer, and (2) the sequence-based LSTM is highly complementary to the traditional classification strategy without considering the temporal frame orders. Extensive experiments are conducted on two popular and challenging benchmarks, the UCF-101 Human Actions and the Columbia Consumer Videos (CCV). On both benchmarks, our framework achieves very competitive performance: 91.3 on the UCF-101 and 83.5 on the CCV.",
"We describe a method of representing human activities that allows a collection of motions to be queried without examples, using a simple and effective query language. Our approach is based on units of activity at segments of the body, that can be composed across space and across the body to produce complex queries. The presence of search units is inferred automatically by tracking the body, lifting the tracks to 3D and comparing to models trained using motion capture data. We show results for a large range of queries applied to a collection of complex motion and activity. Our models of short time scale limb behaviour are built using labelled motion capture set. We compare with discriminative methods applied to tracker data; our method offers significantly improved performance. We show experimental evidence that our method is robust to view direction and is unaffected by the changes of clothing.",
"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.",
"",
"This paper addresses the problem of learning and recognizing human activities of daily living (ADL), which is an important research issue in building a pervasive and smart environment. In dealing with ADL, we argue that it is beneficial to exploit both the inherent hierarchical organization of the activities and their typical duration. To this end, we introduce the switching hidden semi-markov model (S-HSMM), a two-layered extension of the hidden semi-Markov model (HSMM) for the modeling task. Activities are modeled in the S-HSMM in two ways: the bottom layer represents atomic activities and their duration using HSMMs; the top layer represents a sequence of high-level activities where each high-level activity is made of a sequence of atomic activities. We consider two methods for modeling duration: the classic explicit duration model using multinomial distribution, and the novel use of the discrete Coxian distribution. In addition, we propose an effective scheme to detect abnormality without the need for training on abnormal data. Experimental results show that the S-HSMM performs better than existing models including the flat HSMM and the hierarchical hidden Markov model in both classification and abnormality detection tasks, alleviating the need for presegmented training data. Furthermore, our discrete Coxian duration model yields better computation time and generalization error than the classic explicit duration model.",
"We present algorithms for recognizing human motion in monocular video sequences, based on discriminative conditional random field (CRF) and maximum entropy Markov models (MEMM). Existing approaches to this problem typically use generative (joint) structures like the hidden Markov model (HMM). Therefore they have to make simplifying, often unrealistic assumptions on the conditional independence of observations given the motion class labels and cannot accommodate overlapping features or long term contextual dependencies in the observation sequence. In contrast, conditional models like the CRFs seamlessly represent contextual dependencies, support efficient, exact inference using dynamic programming, and their parameters can be trained using convex optimization. We introduce conditional graphical models as complementary tools for human motion recognition and present an extensive set of experiments that show how these typically outperform HMMs in classifying not only diverse human activities like walking, jumping. running, picking or dancing, but also for discriminating among subtle motion styles like normal walk and wander walk",
"This paper presents a method for recognizing scene categories based on approximate global geometric correspondence. This technique works by partitioning the image into increasingly fine sub-regions and computing histograms of local features found inside each sub-region. The resulting \"spatial pyramid\" is a simple and computationally efficient extension of an orderless bag-of-features image representation, and it shows significantly improved performance on challenging scene categorization tasks. Specifically, our proposed method exceeds the state of the art on the Caltech-101 database and achieves high accuracy on a large database of fifteen natural scene categories. The spatial pyramid framework also offers insights into the success of several recently proposed image descriptions, including Torralbas \"gist\" and Lowes SIFT descriptors.",
"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."
]
} |
1708.09522 | 2752655580 | Inspired by recent advances in neural machine translation, that jointly align and translate using encoder-decoder networks equipped with attention, we propose an attentionbased LSTM model for human activity recognition. Our model jointly learns to classify actions and highlight frames associated with the action, by attending to salient visual information through a jointly learned soft-attention networks. We explore attention informed by various forms of visual semantic features, including those encoding actions, objects and scenes. We qualitatively show that soft-attention can learn to effectively attend to important objects and scene information correlated with specific human actions. Further, we show that, quantitatively, our attention-based LSTM outperforms the vanilla LSTM and CNN models used by stateof-the-art methods. On a large-scale youtube video dataset, ActivityNet, our model outperforms competing methods in action classification. | Temporal and spatial attention: There have been previous attempts to ask whether information from all frames is equally important for video recognition understanding @cite_37 . To this end a number of keyframe-based representations have been proposed. For example, @cite_20 rank all frames based on holistic information theoretic measure; @cite_36 rely on spatio-temporal localization and AdaBoost to select keyframes; @cite_30 automatically select compact keyframe representation using a latent structured SVM formulation. In complement, low-level spatial attention models, trained from eye-gaze in video, have been studied in @cite_13 @cite_41 , with relatively limited success. More recently, Shapovalova al @cite_39 use eye-gaze information as weak-supervision for action localization. Our model also has an attention mechanism, but compared with @cite_30 or @cite_39 that suffer from complexities imposed by discrete latent variables, avoids scalability issues by employing soft-attention coupled with LSTM. Our dynamic soft-attention mechanism is also in stark contrast with static low-level attention models of @cite_13 and @cite_41 . | {
"cite_N": [
"@cite_30",
"@cite_37",
"@cite_36",
"@cite_41",
"@cite_39",
"@cite_13",
"@cite_20"
],
"mid": [
"2025508903",
"2136853139",
"2081773958",
"1850164289",
"",
"33363299",
"2079169833"
],
"abstract": [
"In this paper, we develop a new model for recognizing human actions. An action is modeled as a very sparse sequence of temporally local discriminative key frames - collections of partial key-poses of the actor(s), depicting key states in the action sequence. We cast the learning of key frames in a max-margin discriminative framework, where we treat key frames as latent variables. This allows us to (jointly) learn a set of most discriminative key frames while also learning the local temporal context between them. Key frames are encoded using a spatially-localizable pose let-like representation with HoG and BoW components learned from weak annotations, we rely on structured SVM formulation to align our components and mine for hard negatives to boost localization performance. This results in a model that supports spatio-temporal localization and is insensitive to dropped frames or partial observations. We show classification performance that is competitive with the state of the art on the benchmark UT-Interaction dataset and illustrate that our model outperforms prior methods in an on-line streaming setting.",
"Visual recognition of human actions in video clips has been an active field of research in recent years. However, most published methods either analyse an entire video and assign it a single action label, or use relatively large look-ahead to classify each frame. Contrary to these strategies, human vision proves that simple actions can be recognised almost instantaneously. In this paper, we present a system for action recognition from very short sequences (ldquosnippetsrdquo) of 1-10 frames, and systematically evaluate it on standard data sets. It turns out that even local shape and optic flow for a single frame are enough to achieve ap90 correct recognitions, and snippets of 5-7 frames (0.3-0.5 seconds of video) are enough to achieve a performance similar to the one obtainable with the entire video sequence.",
"In this paper we propose a novel method for human action recognition based on boosted key-frame selection and correlated pyramidal motion feature representations. Instead of using an unsupervised method to detect interest points, a Pyramidal Motion Feature (PMF), which combines optical flow with a biologically inspired feature, is extracted from each frame of a video sequence. The AdaBoost learning algorithm is then applied to select the most discriminative frames from a large feature pool. In this way, we obtain the top-ranked boosted frames of each video sequence as the key frames which carry the most representative motion information. Furthermore, we utilise the correlogram which focuses not only on probabilistic distributions within one frame but also on the temporal relationships of the action sequence. In the classification phase, a Support-Vector Machine (SVM) is adopted as the final classifier for human action recognition. To demonstrate generalizability, our method has been systematically tested on a variety of datasets and shown to be more effective and accurate for action recognition compared to the previous work. We obtain overall accuracies of: 95.5 , 93.7 , and 36.5 with our proposed method on the KTH, the multiview IXMAS and the challenging HMDB51 datasets, respectively.",
"Algorithms using \"bag of features\"-style video representations currently achieve state-of-the-art performance on action recognition tasks, such as the challenging Hollywood2 benchmark [1,2,3]. These algorithms are based on local spatiotemporal descriptors that can be extracted either sparsely (at interest points) or densely (on regular grids), with dense sampling typically leading to the best performance [1]. Here, we investigate the benefit of space-variant processing of inputs, inspired by attentional mechanisms in the human visual system. We employ saliency-mapping algorithms to find informative regions and descriptors corresponding to these regions are either used exclusively, or are given greater representational weight (additional codebook vectors). This approach is evaluated with three state-of-the-art action recognition algorithms [1,2,3], and using several saliency algorithms. We also use saliency maps derived from human eye movements to probe the limits of the approach. Saliency-based pruning allows up to 70 of descriptors to be discarded, while maintaining high performance on Hollywood2. Meanwhile, pruning of 20-50 (depending on model) can even improve recognition. Further improvements can be obtained by combining representations learned separately on salience-pruned and unpruned descriptor sets. Not surprisingly, using the human eye movement data gives the best mean Average Precision (mAP; 61.9 ), providing an upper bound on what is possible with a high-quality saliency map. Even without such external data, the Dense Trajectories model [1] enhanced by automated saliency-based descriptor sampling achieves the best mAP (60.0 ) reported on Hollywood2 to date.",
"",
"Systems based on bag-of-words models operating on image features collected at maxima of sparse interest point operators have been extremely successful for both computer-based visual object and action recognition tasks. While the sparse, interest-point based approach to recognition is not inconsistent with visual processing in biological systems that operate in \"saccade and fixate\" regimes, the knowledge, methodology, and emphasis in the human and the computer vision communities remains sharply distinct. Here, we make three contributions aiming to bridge this gap. First, we complement existing state-of-the art large-scale dynamic computer vision datasets like Hollywood-2[1] and UCF Sports[2] with human eye movements collected under the ecological constraints of the visual action recognition task. To our knowledge these are the first massive human eye tracking datasets of significant size to be collected for video (497,107 frames, each viewed by 16 subjects), unique in terms of their (a) large scale and computer vision relevance, (b) dynamic, video stimuli, (c) task control, as opposed to free-viewing. Second, we introduce novel dynamic consistency and alignment models, which underline the remarkable stability of patterns of visual search among subjects. Third, we leverage the massive amounts of collected data in order to pursue studies and build automatic, end-to-end trainable computer vision systems based on human eye movements. Our studies not only shed light on the differences between computer vision spatio-temporal interest point image sampling strategies and human fixations, as well as their impact for visual recognition performance, but also demonstrate that human fixations can be accurately predicted, and when used in an end-to-end automatic system, leveraging some of the most advanced computer vision practice, can lead to state of the art results.",
"This paper presents an approach for human action recognition by finding the discriminative key frames from a video sequence and representing them with the distribution of local motion features and their spatiotemporal arrangements. In this approach, the key frames of the video sequence are selected by their discriminative power and represented by the local motion features detected in them and integrated from their temporal neighbors. In the key frame’s representation, the spatial arrangements of the motion features are captured in a hierarchical spatial pyramid structure. By using frame by frame voting for the recognition, experiments have demonstrated improved performances over most of the other known methods on the popular benchmark data sets. Recognizinghumanactionfromimagesequencesis an appealingyet challengingproblem in computer vision with many applications including motion capture, human-computer interaction, environment control, and security surveillance. In this paper, we focus on recognizing the activities of a person in an image sequence from local motion features and their spatiotemporal arrangements. Our approach is motivated by the recent success of “bag-of-words” model for general object recognition in computer vision[21, 14]. This representation, which is adapted from the text retrieval literature, models the object by the distribution of words from a fixed visual code book, which is usually obtained by vector quantization of local image visual features. However, this method discards the spatial and the temporal relations among these visual features, which could be helpful in object recognition. Addressing this problem, our approach uses a hierarchical representation for the key frames of a given video sequence to integrate information from both the spatial and the temporal domains. We first apply a spatiotemporal feature detector to the video sequence and obtain the local motion features. Then we generate a visual word code book by quantization of the local motion features and assign a word label to each of them. Next we select key frames of the video sequence by their discriminative power. Then, for each key frame, we integrate the visual words from its nearby frames, divide the key frame spatially into finer subdivisions and compute in each cell the histograms of the visual words detected in this key frame and its temporal neighbors. Finally, we concatenate the histograms from all cells and use"
]
} |
1708.09522 | 2752655580 | Inspired by recent advances in neural machine translation, that jointly align and translate using encoder-decoder networks equipped with attention, we propose an attentionbased LSTM model for human activity recognition. Our model jointly learns to classify actions and highlight frames associated with the action, by attending to salient visual information through a jointly learned soft-attention networks. We explore attention informed by various forms of visual semantic features, including those encoding actions, objects and scenes. We qualitatively show that soft-attention can learn to effectively attend to important objects and scene information correlated with specific human actions. Further, we show that, quantitatively, our attention-based LSTM outperforms the vanilla LSTM and CNN models used by stateof-the-art methods. On a large-scale youtube video dataset, ActivityNet, our model outperforms competing methods in action classification. | LSTM models with attention: Models that combine neural network architectures with attention are relatively recent (perhaps the earliest mention is in the context of Boltzmann Machine @cite_52 ). In vision, Xu al @cite_8 use spatial soft-attention that learns to fix gaze on salient image parts while generating description for images. Yao al @cite_29 also use soft-attention mechanism on video frames for video description. Perhaps in the closest work to ours, Yeung al @cite_46 propose a multiLSTM for action detection, where as input-data their model aggregates few frames CNN features using soft-attention and generates multiple outputs at each time step corresponding to multiple frames. The results show that adding soft-attention for weighted averaging of few input frames only marginally ( @math | {
"cite_N": [
"@cite_46",
"@cite_29",
"@cite_52",
"@cite_8"
],
"mid": [
"2952835694",
"2950307714",
"2141399712",
"2950178297"
],
"abstract": [
"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.",
"Recent progress in using recurrent neural networks (RNNs) for image description has motivated the exploration of their application for video description. However, while images are static, working with videos requires modeling their dynamic temporal structure and then properly integrating that information into a natural language description. In this context, we propose an approach that successfully takes into account both the local and global temporal structure of videos to produce descriptions. First, our approach incorporates a spatial temporal 3-D convolutional neural network (3-D CNN) representation of the short temporal dynamics. The 3-D CNN representation is trained on video action recognition tasks, so as to produce a representation that is tuned to human motion and behavior. Second we propose a temporal attention mechanism that allows to go beyond local temporal modeling and learns to automatically select the most relevant temporal segments given the text-generating RNN. Our approach exceeds the current state-of-art for both BLEU and METEOR metrics on the Youtube2Text dataset. We also present results on a new, larger and more challenging dataset of paired video and natural language descriptions.",
"We describe a model based on a Boltzmann machine with third-order connections that can learn how to accumulate information about a shape over several fixations. The model uses a retina that only has enough high resolution pixels to cover a small area of the image, so it must decide on a sequence of fixations and it must combine the \"glimpse\" at each fixation with the location of the fixation before integrating the information with information from other glimpses of the same object. We evaluate this model on a synthetic dataset and two image classification datasets, showing that it can perform at least as well as a model trained on whole images.",
"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."
]
} |
1708.09417 | 2950614765 | LangPro is an automated theorem prover for natural language (this https URL). Given a set of premises and a hypothesis, it is able to prove semantic relations between them. The prover is based on a version of analytic tableau method specially designed for natural logic. The proof procedure operates on logical forms that preserve linguistic expressions to a large extent. This property makes the logical forms easily obtainable from syntactic trees. , in particular, Combinatory Categorial Grammar derivation trees. The nature of proofs is deductive and transparent. On the FraCaS and SICK textual entailment datasets, the prover achieves high results comparable to state-of-the-art. | Theorem proving techniques @cite_2 or ideas from Natural Logic @cite_4 were already used in recognizing textual entailment (RTE). But the combination of these two is a novel approach to RTE. The underlying higher-order logic of guarantees sound reasoning over several premises, including some complex semantic phenomena. This is in contrast to the RTE systems that cannot reason over several premises or cannot account for Booleans and quantifiers, including the ones @cite_4 inspired by Natural Logic, and in contrast to those ones that use FOL representations and cannot cover higher-order phenomena like generalized quantifiers or subsective adjectives. | {
"cite_N": [
"@cite_4",
"@cite_2"
],
"mid": [
"2228826686",
"2045254372"
],
"abstract": [
"Inference has been a central topic in artificial intelligence from the start, but while automatic methods for formal deduction have advanced tremendously, comparatively little progress has been made on the problem of natural language inference (NLI), that is, determining whether a natural language hypothesis h can justifiably be inferred from a natural language premise p. The challenges of NLI are quite different from those encountered in formal deduction: the emphasis is on informal reasoning, lexical semantic knowledge, and variability of linguistic expression. This dissertation explores a range of approaches to NLI, beginning with methods which are robust but approximate, and proceeding to progressively more precise approaches. We first develop a baseline system based on overlap between bags of words. Despite its extreme simplicity, this model achieves surprisingly good results on a standard NLI evaluation, the PASCAL RTE Challenge. However, its effectiveness is limited by its failure to represent semantic structure. To remedy this lack, we next introduce the Stanford RTE system, which uses typed dependency trees as a proxy for semantic structure, and seeks a low-cost alignment between trees for p and h, using a cost model which incorporates both lexical and structural matching costs. This system is typical of a category of approaches to NLI based on approximate graph matching. We argue, however, that such methods work best when the entailment decision is based, not merely on the degree of alignment, but also on global features of the aligned 〈p, h〉 pair motivated by semantic theory. Seeking still greater precision, we devote the largest part of the dissertation to developing an approach to NLI based on a model of natural logic. We greatly extend past work in natural logic, which has focused solely on semantic containment and monotonicity, to incorporate both semantic exclusion and implicativity. Our system decomposes an inference problem into a sequence of atomic edits which transforms p into h; predicts a lexical entailment relation for each edit using a statistical classifier; propagates these relations upward through a syntax tree according to semantic properties of intermediate nodes; and composes the resulting entailment relations across the edit sequence. Finally, we address the problem of alignment for NLI, by developing a model of phrase-based alignment inspired by analogous work in machine translation, including an alignment scoring function, inference algorithms for finding good alignments, and training algorithms for choosing feature weights.",
"We use logical inference techniques for recognising textual entailment. As the performance of theorem proving turns out to be highly dependent on not readily available background knowledge, we incorporate model building, a technique borrowed from automated reasoning, and show that it is a useful robust method to approximate entailment. Finally, we use machine learning to combine these deep semantic analysis techniques with simple shallow word overlap; the resulting hybrid model achieves high accuracy on the RTE testset, given the state of the art. Our results also show that the different techniques that we employ perform very differently on some of the subsets of the RTE corpus and as a result, it is useful to use the nature of the dataset as a feature."
]
} |
1708.09332 | 2619968118 | With the increased use of Internet, governments and large companies store and share massive amounts of personal data in such a way that leaves no space for transparency. When a user needs to achieve a simple task like applying for college or a driving license, he needs to visit a lot of institutions and organizations, thus leaving a lot of private data in many places. The same happens when using the Internet. These privacy issues raised by the centralized architectures along with the recent developments in the area of serverless applications demand a decentralized private data layer under user control. | Although many authors perceive choreographies as a mechanism to describe in a more formal way the contracts among several organizations @cite_7 , the academic research proposed the concept of executable choreographies @cite_8 @cite_1 @cite_13 . They suggest transforming the descriptions of the choreographies in code that is executed inside each organization participating in the choreography. As such, a choreography is not only a formal description of a contract among organizations but also a description of a workflow in an executable way. The same description (choreography) gets to run in several organizations in a decentralized manner (without the need for a centralized conductor) and therefore any need to translate the choreography into other programming languages disappears. | {
"cite_N": [
"@cite_13",
"@cite_1",
"@cite_7",
"@cite_8"
],
"mid": [
"165177164",
"2213903010",
"",
"2141865897"
],
"abstract": [
"The Lightweight Coordination Calculus (LCC) is a compact choreography language based on process calculus. LCC is a directly executable specification and can therefore be dynamically distributed to a group of peers for enactment at run-time; this offers flexibility and allows peers to coordinate in open systems without prior knowledge of an interaction. This paper contributes to the body of choreography research by proposing two extensions to LCC covering parallel composition and choreography abstraction. These language extensions are evaluated against a subset of the Service Interaction Patterns, a benchmark in the process modelling community.",
"Enforcing in code privacy laws, internal company rules and principles like Privacy by Design is recognized as a challenge for the IT industry. In this paper we analyze the steps required and propose a guide towards this major goal. Our proposal is to emphasize the need to overcome the limits of service orchestration and create strong privacy and security enabling architectures based on two main ideas. The first idea is to use a semantic firewall that is capable to check privacy properties for the communication between applications and cloud and between cloud's sub-systems. The second idea is to improve current SOA architectures with architectures based on executable choreographies that can be formally verified. In this paper we identify three types of executable choreographies. New types of abstraction which machines can verify and humans can trust are enabled by executable choreographies that act like truly verifiable environments for cloud applications.",
"",
"Interoperability and loose coupling requirements are pushing the next generation of distributed applications towards more decentralized and more dynamic interaction schemes, which the classic request response communication paradigm can hardly accommodate. Hence, sound foundations and mechanisms for the establishment of unanticipated peer-to-peer interactions across organizational boundaries are of significant importance to upcoming middleware platforms. The executable choreography framework (ECF) is a middleware-level framework that targets dynamic and decentralized service compositions. The ECF combines transparent context propagation with aspect-oriented software composition techniques to dynamically refine the default control and data flow of service invocations. The framework provides a ground for experimentation with dynamic and distributed workflows, and a base to assess their safety and applicability when deployed across organizational boundaries."
]
} |
1708.09332 | 2619968118 | With the increased use of Internet, governments and large companies store and share massive amounts of personal data in such a way that leaves no space for transparency. When a user needs to achieve a simple task like applying for college or a driving license, he needs to visit a lot of institutions and organizations, thus leaving a lot of private data in many places. The same happens when using the Internet. These privacy issues raised by the centralized architectures along with the recent developments in the area of serverless applications demand a decentralized private data layer under user control. | In regards to the advances related to data sovereignty, we would like to mention @cite_4 , which proposes storing encrypted data in cloud federations and @cite_11 , which proposes sovereign information sharing in order to integrate the information belonging to autonomous entities. Queries are executed on the databases and reveal only the results. The work is continued in @cite_14 which enables sovereign information sharing using web services. This work applies to service providers which want to allow queries on their databases without sharing the content on which the queries are executed. Our work focuses on the average user which needs to own and store his data in a single place and provide revoke access to it to various service providers as needed. | {
"cite_N": [
"@cite_14",
"@cite_4",
"@cite_11"
],
"mid": [
"2135388600",
"2291549475",
"2139979322"
],
"abstract": [
"Sovereign information sharing allows autonomous entities to compute queries across their databases in such a way that nothing apart from the result is revealed. We describe an implementation of this model using web services infrastructure. Each site participating in sovereign sharing offers a data service that allows database operations to be applied on the tables they own. Of particular interest is the provision for binary operations such as relational joins. Applications are developed by combining these data services. We present performance measurements that show the promise of a new breed of practical applications based on the paradigm of sovereign information integration.",
"The rapidly growing demand for cloud services in the current business practice has favored the success of the hybrid clouds and the advent of cloud federation. The available literature of this topic has focused on middleware abstraction to interoperate heterogeneous cloud platforms and orchestrate different management and business models. However, cloud federation implies serious security and privacy issues with respect to data sovereignty when data is outsourced across different judicial and legal systems. This column describes a solution that applies encryption to protect data sovereignty in federated clouds rather than restricting the elasticity and migration of data across federated clouds.",
"Literature on information integration across databases tacitly assumes that the data in each database can be revealed to the other databases. However, there is an increasing need for sharing information across autonomous entities in such a way that no information apart from the answer to the query is revealed. We formalize the notion of minimal information sharing across private databases, and develop protocols for intersection, equijoin, intersection size, and equijoin size. We also show how new applications can be built using the proposed protocols."
]
} |
1708.09332 | 2619968118 | With the increased use of Internet, governments and large companies store and share massive amounts of personal data in such a way that leaves no space for transparency. When a user needs to achieve a simple task like applying for college or a driving license, he needs to visit a lot of institutions and organizations, thus leaving a lot of private data in many places. The same happens when using the Internet. These privacy issues raised by the centralized architectures along with the recent developments in the area of serverless applications demand a decentralized private data layer under user control. | Note that @cite_9 introduces the data sovereignty notion for establishing the nation-state where the cloud storage service providers are storing the data physically in order to ensure they are meeting their contractual geographic obligations. In this paper, we consider data sovereignty to be the ability of the user to have full control over his data and the entities to which it is shared or revoked. | {
"cite_N": [
"@cite_9"
],
"mid": [
"1550039286"
],
"abstract": [
"In this paper we define the problem and scope of data sovereignty - the coupling of stored data authenticity and geographical location in the cloud. Establishing sovereignty is an especially important concern amid legal and policy constraints when data and resources are virtualized and widely distributed. We identify the key challenges that need to be solved to achieve an effective and un-cheatable solution as well as propose an initial technique for data sovereignty."
]
} |
1708.09234 | 2752195334 | Graph-based synset induction methods, such as MaxMax and Watset, induce synsets by performing a global clustering of a synonymy graph. However, such methods are sensitive to the structure of the input synonymy graph: sparseness of the input dictionary can substantially reduce the quality of the extracted synsets. In this paper, we propose two different approaches designed to alleviate the incompleteness of the input dictionaries. The first one performs a pre-processing of the graph by adding missing edges, while the second one performs a post-processing by merging similar synset clusters. We evaluate these approaches on two datasets for the Russian language and discuss their impact on the performance of synset induction methods. Finally, we perform an extensive error analysis of each approach and discuss prominent alternative methods for coping with the problem of sparsity of the synonymy dictionaries. | Hope and Keller @cite_29 introduced the clustering algorithm particularly designed for the word sense induction task. In a nutshell, pairs of nodes are grouped if they have a maximal mutual affinity. The algorithm starts by converting the undirected input graph into a directed graph by keeping the maximal affinity nodes of each node. Next, all nodes are marked as root nodes. Finally, for each root node, the following procedure is repeated: all transitive children of this root form a cluster and the root are marked as non-root nodes; a root node together with all its transitive children form a fuzzy cluster. | {
"cite_N": [
"@cite_29"
],
"mid": [
"6380806"
],
"abstract": [
"This paper introduces a linear time graph-based soft clustering algorithm. The algorithm applies a simple idea: given a graph, vertex pairs are assigned to the same cluster if either vertex has maximal affinity to the other. Clusters of varying size, shape, and density are found automatically making the algorithm suited to tasks such Word Sense Induction (WSI), where the number of classes is unknown and where class distributions may be skewed. The algorithm is applied to two WSI tasks, obtaining results comparable with those of systems adopting existing, state-of-the-art methods."
]
} |
1708.09234 | 2752195334 | Graph-based synset induction methods, such as MaxMax and Watset, induce synsets by performing a global clustering of a synonymy graph. However, such methods are sensitive to the structure of the input synonymy graph: sparseness of the input dictionary can substantially reduce the quality of the extracted synsets. In this paper, we propose two different approaches designed to alleviate the incompleteness of the input dictionaries. The first one performs a pre-processing of the graph by adding missing edges, while the second one performs a post-processing by merging similar synset clusters. We evaluate these approaches on two datasets for the Russian language and discuss their impact on the performance of synset induction methods. Finally, we perform an extensive error analysis of each approach and discuss prominent alternative methods for coping with the problem of sparsity of the synonymy dictionaries. | Van Dongen @cite_13 presented the Markov Clustering (MCL) algorithm for graphs based on simulation of stochastic flow in graphs. MCL simulates random walks within a graph by alternation of two operators called expansion and inflation, which recompute the class labels. This approch has been successfully used for the word sense induction task @cite_15 . | {
"cite_N": [
"@cite_15",
"@cite_13"
],
"mid": [
"2000951787",
"2121564430"
],
"abstract": [
"This paper presents an unsupervised algorithm which automatically discovers word senses from text. The algorithm is based on a graph model representing words and relationships between them. Sense clusters are iteratively computed by clustering the local graph of similar words around an ambiguous word. Discrimination against previously extracted sense clusters enables us to discover new senses. We use the same data for both recognising and resolving ambiguity.",
"Dit proefschrift heeft als onderwerp het clusteren van grafen door middel van simulatie van stroming, een probleem dat in zijn algemeenheid behoort tot het gebied der clusteranalyse. In deze tak van wetenschap ontwerpt en onderzoekt men methoden die gegeven bepaalde data een onderverdeling in groepen genereren, waarbij het oogmerk is een onderverdeling in groepen te vinden die natuurlijk is. Dat wil zeggen dat verschillende data-elementen in dezelfde groep idealiter veel op elkaar lijken, en dat data-elementen uit verschillende groepen idealiter veel van elkaar verschillen. Soms ontbreken zulke groepjes helemaal; dan is er weinig patroon te herkennen in de data. Het idee is dat de aanwezigheid van natuurlijke groepjes het mogelijk maakt de data te categoriseren. Een voorbeeld is het clusteren van gegevens (over symptomen of lichaamskarakteristieken) van patienten die aan dezelfde ziekte lijden. Als er duidelijke groepjes bestaan in die gegevens, kan dit tot extra inzicht leiden in de ziekte. Clusteranalyse kan aldus gebruikt worden voor exploratief onderzoek. Verdere voorbeelden komen uit de scheikunde, taxonomie, psychiatrie, archeologie, marktonderzoek en nog vele andere disicplines. Taxonomie, de studie van de classificatie van organismen, heeft een rijke geschiedenis beginnend bij Aristoteles en culminerend in de werken van Linnaeus. In feite kan de clusteranalyse gezien worden als het resultaat van een steeds meer systematische en abstracte studie van de diverse methoden ontworpen in verschillende toepassingsgebieden, waarbij methode zowel wordt gescheiden van data en toepassingsgebied als van berekeningswijze. In de cluster analyse kunnen grofweg twee richtingen onderscheiden worden, naar gelang het type data dat geclassificeerd moet worden. De data-elementen in het voorbeeld hierboven worden beschreven door vectoren (lijstjes van scores of metingen), en het verschil tussen twee elementen wordt bepaald door het verschil van de vectoren. Deze dissertatie betreft cluster analyse toegepast op data van het type graaf'. Voorbeelden komen uit de patroonherkenning, het computer ondersteund ontwerpen, databases voorzien van hyperlinks en het World Wide Web. In al deze gevallen is er sprake van punten' die verbonden zijn of niet. Een stelsel van punten samen met hun verbindingen heet een graaf. Een goede clustering van een graaf deelt de punten op in groepjes zodanig dat er weinig verbindingen lopen tussen (punten uit) verschillende groepjes en er veel verbindingen zijn in elk groepje afzonderlijk."
]
} |
1708.09234 | 2752195334 | Graph-based synset induction methods, such as MaxMax and Watset, induce synsets by performing a global clustering of a synonymy graph. However, such methods are sensitive to the structure of the input synonymy graph: sparseness of the input dictionary can substantially reduce the quality of the extracted synsets. In this paper, we propose two different approaches designed to alleviate the incompleteness of the input dictionaries. The first one performs a pre-processing of the graph by adding missing edges, while the second one performs a post-processing by merging similar synset clusters. We evaluate these approaches on two datasets for the Russian language and discuss their impact on the performance of synset induction methods. Finally, we perform an extensive error analysis of each approach and discuss prominent alternative methods for coping with the problem of sparsity of the synonymy dictionaries. | Biemann @cite_14 introduced Chinese Whispers, a clustering algorithm for weighted graphs that can be considered as a special case of MCL with a simplified class update step. At each iteration, the labels of all the nodes are updated according to the majority labels among the neighboring nodes. The author showed usefulness of the algorithm for induction of word senses based on corpus-induced graphs. | {
"cite_N": [
"@cite_14"
],
"mid": [
"1987302197"
],
"abstract": [
"We introduce Chinese Whispers, a randomized graph-clustering algorithm, which is time-linear in the number of edges. After a detailed definition of the algorithm and a discussion of its strengths and weaknesses, the performance of Chinese Whispers is measured on Natural Language Processing (NLP) problems as diverse as language separation, acquisition of syntactic word classes and word sense disambiguation. At this, the fact is employed that the small-world property holds for many graphs in NLP."
]
} |
1708.09234 | 2752195334 | Graph-based synset induction methods, such as MaxMax and Watset, induce synsets by performing a global clustering of a synonymy graph. However, such methods are sensitive to the structure of the input synonymy graph: sparseness of the input dictionary can substantially reduce the quality of the extracted synsets. In this paper, we propose two different approaches designed to alleviate the incompleteness of the input dictionaries. The first one performs a pre-processing of the graph by adding missing edges, while the second one performs a post-processing by merging similar synset clusters. We evaluate these approaches on two datasets for the Russian language and discuss their impact on the performance of synset induction methods. Finally, we perform an extensive error analysis of each approach and discuss prominent alternative methods for coping with the problem of sparsity of the synonymy dictionaries. | The ECO approach @cite_23 was applied to induce a WordNet of the Portuguese language. http: ontopt.dei.uc.pt In its core, ECO is based on a clustering algorithm that was used to induce synsets from synonymy dictionaries. The algorithm starts by adding random noise to edge weights. Then, the approach applies Markov Clustering of this graph several times to estimate the probability of each word pair being in the same synset. Finally, candidate pairs over a certain threshold are added to output synsets. | {
"cite_N": [
"@cite_23"
],
"mid": [
"2061429721"
],
"abstract": [
"A wordnet is an important tool for developing natural language processing applications for a language. However, most wordnets are handcrafted by experts, which limits their growth. In this article, we propose an automatic approach to create wordnets by exploiting textual resources, dubbed ECO. After extracting semantic relation instances, identified by discriminating textual patterns, ECO discovers synonymy clusters, used as synsets, and attaches the remaining relations to suitable synsets. Besides introducing each step of ECO, we report on how it was implemented to create Onto.PT, a public lexical ontology for Portuguese. Onto.PT is the result of the automatic exploitation of Portuguese dictionaries and thesauri, and it aims to minimise the main limitations of existing Portuguese lexical knowledge bases."
]
} |
1708.09234 | 2752195334 | Graph-based synset induction methods, such as MaxMax and Watset, induce synsets by performing a global clustering of a synonymy graph. However, such methods are sensitive to the structure of the input synonymy graph: sparseness of the input dictionary can substantially reduce the quality of the extracted synsets. In this paper, we propose two different approaches designed to alleviate the incompleteness of the input dictionaries. The first one performs a pre-processing of the graph by adding missing edges, while the second one performs a post-processing by merging similar synset clusters. We evaluate these approaches on two datasets for the Russian language and discuss their impact on the performance of synset induction methods. Finally, we perform an extensive error analysis of each approach and discuss prominent alternative methods for coping with the problem of sparsity of the synonymy dictionaries. | In our experiments, we rely on the synset induction method @cite_8 based on a graph meta-clustering algorithm that combines local and global hard clustering to obtain a fuzzy graph clustering. The authors shown that this approach outperforms all methods mentioned above on the synset induction task and therefore we use it as the strongest baseline to date. | {
"cite_N": [
"@cite_8"
],
"mid": [
"2608041596"
],
"abstract": [
"This paper presents a new graph-based approach that induces synsets using synonymy dictionaries and word embeddings. First, we build a weighted graph of synonyms extracted from commonly available resources, such as Wiktionary. Second, we apply word sense induction to deal with ambiguous words. Finally, we cluster the disambiguated version of the ambiguous input graph into synsets. Our meta-clustering approach lets us use an efficient hard clustering algorithm to perform a fuzzy clustering of the graph. Despite its simplicity, our approach shows excellent results, outperforming five competitive state-of-the-art methods in terms of F-score on three gold standard datasets for English and Russian derived from large-scale manually constructed lexical resources."
]
} |
1708.09234 | 2752195334 | Graph-based synset induction methods, such as MaxMax and Watset, induce synsets by performing a global clustering of a synonymy graph. However, such methods are sensitive to the structure of the input synonymy graph: sparseness of the input dictionary can substantially reduce the quality of the extracted synsets. In this paper, we propose two different approaches designed to alleviate the incompleteness of the input dictionaries. The first one performs a pre-processing of the graph by adding missing edges, while the second one performs a post-processing by merging similar synset clusters. We evaluate these approaches on two datasets for the Russian language and discuss their impact on the performance of synset induction methods. Finally, we perform an extensive error analysis of each approach and discuss prominent alternative methods for coping with the problem of sparsity of the synonymy dictionaries. | Meyer and Gurevich @cite_1 presented an approach for construction of an ontologized version of Wiktionary, by formation of ontological concepts and relationships between them from the ambiguous input dictionary, yet their approach does not involve graph clustering. | {
"cite_N": [
"@cite_1"
],
"mid": [
"2482986266"
],
"abstract": [
"The semi-automatic development of ontologies is an important field of research, since existing ontologies often suffer from their small size, unaffordable construction cost, and limited quality of ontology learning systems. The main objective of this chapter is to introduce Wiktionary, which is a collaborative online dictionary encoding information about words, word senses, and relations between them, as a resource for ontology construction. The authors find that a Wiktionary-based ontology can exceed the size of, for example, OpenCyc and OntoWordNet. One particular advantage of Wiktionary is its multilingual nature, which allows the construction of ontologies for different languages. Additionally, its collaborative construction approach means that novel concepts and domain-specific knowledge are quick to appear in the dictionary. For constructing their ontology OntoWiktionary, the authors present a two-step approach that involves (1) harvesting structured knowledge from Wiktionary and (2) ontologizing this knowledge (i.e., the formation of ontological concepts and relationships from the harvested knowledge). They evaluate their approach based on human judgments and find their new ontology to be of overall good quality. To encourage further research in this field, the authors make the final OntoWiktionary publicly available and suggest integrating this novel resource with the linked data cloud as well as other existing ontology projects."
]
} |
1708.09087 | 2752668377 | Recent studies have suggested that the stability of peer-to-peer networks may rely on persistent peers, who dwell on the network after they obtain the entire file. In the absence of such peers, one piece becomes extremely rare in the network, which leads to instability. Technological developments, however, are poised to reduce the incidence of persistent peers, giving rise to a need for a protocol that guarantees stability with non-persistent peers. We propose a novel peer-to-peer protocol, the group suppression protocol, to ensure the stability of peer-to-peer networks under the scenario that all the peers adopt non-persistent behavior. Using a suitable Lyapunov potential function, the group suppression protocol is proven to be stable when the file is broken into two pieces, and detailed experiments demonstrate the stability of the protocol for arbitrary number of pieces. We define and simulate a decentralized version of this protocol for practical applications. Straightforward incorporation of the group suppression protocol into BitTorrent while retaining most of BitTorrent's core mechanisms is also presented. Subsequent simulations show that under certain assumptions, BitTorrent with the official protocol cannot escape from the missing piece syndrome, but BitTorrent with group suppression does. | Massouli 'e and Vojnovic @cite_26 proved global stability in a model with non-persistent peers if all new peers arrive into the system with one uniformly random piece of the file uploaded by the fixed seed. The fixed seed's upload rate, however, may not be large enough to allow this in practice. | {
"cite_N": [
"@cite_26"
],
"mid": [
"2171217544"
],
"abstract": [
"Motivated by the study of peer-to-peer file swarming systems a la BitTorrent, we introduce a probabilistic model of coupon replication systems. These systems consist of users aiming to complete a collection of distinct coupons. Users enter the system with an initial coupon provided by a bootstrap server, acquire other coupons from other users, and leave once they complete their coupon collection. For open systems, with exogenous user arrivals, we derive stability condition for a layered scenario, where encounters are between users holding the same number of coupons. We also consider a system where encounters are between users chosen uniformly at random from the whole population. We show that sojourn time in both systems is asymptotically optimal as the number of coupon types becomes large. We also consider closed systems with no exogenous user arrivals. In a special scenario where users have only one missing coupon, we evaluate the size of the population ultimately remaining in the system, as the initial number of users N goes to infinity. We show that this size decreases geometrically with the number of coupons K. In particular, when the ratio K log(N) is above a critical threshold, we prove that this number of leftovers is of order log(log(N)). These results suggest that, under the assumption that the bootstrap server is not a bottleneck, the performance does not depend critically on either altruistic user behavior or on load-balancing strategies such as rarest first."
]
} |
1708.09157 | 2753398138 | Even for common NLP tasks, sufficient supervision is not available in many languages -- morphological tagging is no exception. In the work presented here, we explore a transfer learning scheme, whereby we train character-level recurrent neural taggers to predict morphological taggings for high-resource languages and low-resource languages together. Learning joint character representations among multiple related languages successfully enables knowledge transfer from the high-resource languages to the low-resource ones, improving accuracy by up to 30 | Most cross-lingual work in NLP---focusing on morphology or otherwise---has concentrated on indirect supervision, rather than transfer learning. The goal in such a regime is to provide noisy labels for training the tagger in the low-resource language through annotations projected over aligned bitext with a high-resource language. This method of projection was first introduced by for the projection of POS annotation. While follow-up work @cite_1 @cite_33 @cite_11 has continually demonstrated the efficacy of projecting simple part-of-speech annotations, were the first to show the use of bitext-based projection for the training of a morphological tagger for low-resource languages. | {
"cite_N": [
"@cite_1",
"@cite_33",
"@cite_11"
],
"mid": [
"1506525786",
"2142523187",
"1818534184"
],
"abstract": [
"We implement a variant of the algorithm described by Yarowsky and Ngai in [21] to induce an HMM POS tagger for an arbitrary target language using only an existing POS tagger for a source language and an unannotated parallel corpus between the source and target languages. We extend this work by projecting from multiple source languages onto a single target language. We hypothesize that systematic transfer errors from differing source languages will cancel out, improving the quality of bootstrapped resources in the target language. Our experiments confirm the hypothesis. Each experiment compares three cases: (a) source data comes from a single language A, (b) source data comes from a single language B, and (c) source data comes from both A and B, but half as much from each. Apart from the source language, other conditions are held constant in all three cases – including the total amount of source data used. The null hypothesis is that performance in the mixed case would be an average of performance in the single-language cases, but in fact, mixed-case performance always exceeds the maximum of the single-language cases. We observed this effect in all six experiments we ran, involving three different source-language pairs and two different target languages.",
"We describe a novel approach for inducing unsupervised part-of-speech taggers for languages that have no labeled training data, but have translated text in a resource-rich language. Our method does not assume any knowledge about the target language (in particular no tagging dictionary is assumed), making it applicable to a wide array of resource-poor languages. We use graph-based label propagation for cross-lingual knowledge transfer and use the projected labels as features in an unsupervised model (Berg-, 2010). Across eight European languages, our approach results in an average absolute improvement of 10.4 over a state-of-the-art baseline, and 16.7 over vanilla hidden Markov models induced with the Expectation Maximization algorithm.",
"It has been established that incorporating word cluster features derived from large unlabeled corpora can significantly improve prediction of linguistic structure. While previous work has focused primarily on English, we extend these results to other languages along two dimensions. First, we show that these results hold true for a number of languages across families. Second, and more interestingly, we provide an algorithm for inducing cross-lingual clusters and we show that features derived from these clusters significantly improve the accuracy of cross-lingual structure prediction. Specifically, we show that by augmenting direct-transfer systems with cross-lingual cluster features, the relative error of delexicalized dependency parsers, trained on English treebanks and transferred to foreign languages, can be reduced by up to 13 . When applying the same method to direct transfer of named-entity recognizers, we observe relative improvements of up to 26 ."
]
} |
1708.09157 | 2753398138 | Even for common NLP tasks, sufficient supervision is not available in many languages -- morphological tagging is no exception. In the work presented here, we explore a transfer learning scheme, whereby we train character-level recurrent neural taggers to predict morphological taggings for high-resource languages and low-resource languages together. Learning joint character representations among multiple related languages successfully enables knowledge transfer from the high-resource languages to the low-resource ones, improving accuracy by up to 30 | As we also discuss the training of a morphological tagger, our work is most closely related to in terms of the task itself. We contrast the approaches. The main difference lies therein, that our approach is not projection-based and, thus, does not require the construction of a bilingual lexicon for projection based on bitext. Rather, our method jointly learns multiple taggers and forces them to share features---a true transfer learning scenario. In contrast to projection-based methods, our procedure always requires a minimal amount of annotated data in the low-resource target language---in practice, however, this distinction is non-critical as projection-based methods without a small mount of seed target language data perform poorly @cite_39 . | {
"cite_N": [
"@cite_39"
],
"mid": [
"2426917359"
],
"abstract": [
"Morphologically rich languages often lack the annotated linguistic resources required to develop accurate natural language processing tools. We propose models suitable for training morphological taggers with rich tagsets for low-resource languages without using direct supervision. Our approach extends existing approaches of projecting part-of-speech tags across languages, using bitext to infer constraints on the possible tags for a given word type or token. We propose a tagging model using Wsabie, a discriminative embeddingbased model with rank-based learning. In our evaluation on 11 languages, on average this model performs on par with a baseline weakly-supervised HMM, while being more scalable. Multilingual experiments show that the method performs best when projecting between related language pairs. Despite the inherently lossy projection, we show that the morphological tags predicted by our models improve the downstream performance of a parser by +0.6 LAS on average."
]
} |
1708.09157 | 2753398138 | Even for common NLP tasks, sufficient supervision is not available in many languages -- morphological tagging is no exception. In the work presented here, we explore a transfer learning scheme, whereby we train character-level recurrent neural taggers to predict morphological taggings for high-resource languages and low-resource languages together. Learning joint character representations among multiple related languages successfully enables knowledge transfer from the high-resource languages to the low-resource ones, improving accuracy by up to 30 | Our work also follows a recent trend in NLP, whereby traditional word-level neural representations are being replaced by character-level representations for a myriad tasks, e.g., POS tagging , parsing @cite_6 , language modeling @cite_41 , sentiment analysis @cite_7 as well as the tagger of , whose work we build upon. Our work is also related to recent work on character-level morphological generation using neural architectures @cite_19 @cite_40 . | {
"cite_N": [
"@cite_7",
"@cite_41",
"@cite_6",
"@cite_19",
"@cite_40"
],
"mid": [
"2963012544",
"2949563612",
"1860935423",
"2963965928",
"2470595162"
],
"abstract": [
"This article offers an empirical exploration on the use of character-level convolutional networks (ConvNets) for text classification. We constructed several large-scale datasets to show that character-level convolutional networks could achieve state-of-the-art or competitive results. Comparisons are offered against traditional models such as bag of words, n-grams and their TFIDF variants, and deep learning models such as word-based ConvNets and recurrent neural networks.",
"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 present extensions to a continuousstate dependency parsing method that makes it applicable to morphologically rich languages. Starting with a highperformance transition-based parser that uses long short-term memory (LSTM) recurrent neural networks to learn representations of the parser state, we replace lookup-based word representations with representations constructed from the orthographic representations of the words, also using LSTMs. This allows statistical sharing across word forms that are similar on the surface. Experiments for morphologically rich languages show that the parsing model benefits from incorporating the character-based encodings of words.",
"Morphological inflection generation is the task of generating the inflected form of a given lemma corresponding to a particular linguistic transformation. We model the problem of inflection generation as a character sequence to sequence learning problem and present a variant of the neural encoder-decoder model for solving it. Our model is language independent and can be trained in both supervised and semi-supervised settings. We evaluate our system on seven datasets of morphologically rich languages and achieve either better or comparable results to existing state-of-the-art models of inflection generation.",
""
]
} |
1708.08985 | 2752775682 | Generative models are widely used for unsupervised learning with various applications, including data compression and signal restoration. Training methods for such systems focus on the generality of the network given limited amount of training data. A less researched type of techniques concerns generation of only a single type of input. This is useful for applications such as constraint handling, noise reduction and anomaly detection. In this paper we present a technique to limit the generative capability of the network using negative learning. The proposed method searches the solution in the gradient direction for the desired input and in the opposite direction for the undesired input. One of the application can be anomaly detection where the undesired inputs are the anomalous data. In the results section we demonstrate the features of the algorithm using MNIST handwritten digit dataset and latter apply the technique to a real-world obstacle detection problem. The results clearly show that the proposed learning technique can significantly improve the performance for anomaly detection. | There are a large number of literatures on noise reduction and anomaly detection using generative models. M.N. @cite_2 uses non-negative sparse coding to reduce the wind noise in speech data. They rely on a system that have the source model for the wind noise but not for the speech to reduce the noise. The work done on denoising autoencoder by V. @cite_8 is also very important in this area. L. Gondara @cite_14 presents an application of such denoising system to remove noise from medical images. These techniques can reduce the noise the input data but they do not limit the generative capabilities of the network. Due to the generalization property of such networks, they can also generate the data that is very different from the data shown during training. | {
"cite_N": [
"@cite_8",
"@cite_14",
"@cite_2"
],
"mid": [
"",
"2510850936",
"2156448167"
],
"abstract": [
"",
"Image denoising is an important pre-processing step in medical image analysis. Different algorithms have been proposed in past three decades with varying denoising performances. More recently, having outperformed all conventional methods, deep learning based models have shown a great promise. These methods are however limited for requirement of large training sample size and high computational costs. In this paper we show that using small sample size, denoising autoencoders constructed using convolutional layers can be used for efficient denoising of medical images. Heterogeneous images can be combined to boost sample size for increased denoising performance. Simplest of networks can reconstruct images with corruption levels so high that noise and signal are not differentiable to human eye.",
"We introduce a new speaker independent method for reducing wind noise in single-channel recordings of noisy speech. The method is based on non-negative sparse coding and relies on a wind noise dictionary which is estimated from an isolated noise recording. We estimate the parameters of the model and discuss their sensitivity. We then compare the algorithm with the classical spectral subtraction method and the Qualcomm-ICSI-OGI noise reduction method. We optimize the sound quality in terms of signal-to-noise ratio and provide results on a noisy speech recognition task."
]
} |
1708.08985 | 2752775682 | Generative models are widely used for unsupervised learning with various applications, including data compression and signal restoration. Training methods for such systems focus on the generality of the network given limited amount of training data. A less researched type of techniques concerns generation of only a single type of input. This is useful for applications such as constraint handling, noise reduction and anomaly detection. In this paper we present a technique to limit the generative capability of the network using negative learning. The proposed method searches the solution in the gradient direction for the desired input and in the opposite direction for the undesired input. One of the application can be anomaly detection where the undesired inputs are the anomalous data. In the results section we demonstrate the features of the algorithm using MNIST handwritten digit dataset and latter apply the technique to a real-world obstacle detection problem. The results clearly show that the proposed learning technique can significantly improve the performance for anomaly detection. | Our proposed method uses a similar approach to C. Creusot and A. Munawar @cite_1 . They use an extremely compressive Restricted Boltzmann Machine (RBM) to form a deep feature representation. But rather than training a classifier in the feature space, anomaly detection is performed by reconstructing the data back to the original image space and use conventional image difference as a metric. The extreme compression in autoencoders can severely effect the reconstruction of input appearance in case the non-anomalous data have a non-trivial appearance. | {
"cite_N": [
"@cite_1"
],
"mid": [
"1567041090"
],
"abstract": [
"Small objects on the road can become hazardous obstacles when driving at high speed. Detecting such obstacles is vital to guaranty the safety of self-driving car users, especially on highways. Such tasks cannot be performed using existing active sensors such as radar or LIDAR due to their limited range and resolution at long distances. In this paper we propose a technique to detect anomalous patches on the road from color images using a Restricted Boltzman Machine neural network specifically trained to reconstruct the appearance of the road. The differences between the observed and reconstructed road patches yield a more relevant segmentation of anomalies than classic image processing techniques. We evaluated our technique on texture-based synthetic datasets as well as on real video footage of anomalous objects on highways."
]
} |
1708.09083 | 2751833026 | Incorporating additional knowledge in the learning process can be beneficial for several computer vision and machine learning tasks. Whether privileged information originates from a source domain that is adapted to a target domain, or as additional features available at training time only, using such privileged (i.e., auxiliary) information is of high importance as it improves the recognition performance and generalization. However, both primary and privileged information are rarely derived from the same distribution, which poses an additional challenge to the recognition task. To address these challenges, we present a novel learning paradigm that leverages privileged information in a domain adaptation setup to perform visual recognition tasks. The proposed framework, named Adaptive SVM+, combines the advantages of both the learning using privileged information (LUPI) paradigm and the domain adaptation framework, which are naturally embedded in the objective function of a regular SVM. We demonstrate the effectiveness of our approach on the publicly available Animals with Attributes and INTERACT datasets and report state-of-the-art results in both of them. | : The idea of leveraging additional information during the learning phase is not a new concept as it has previously been addressed in the literature in many contexts. The choice of different types of privileged information in the context of object recognition implemented in a max-margin scheme was proposed by Sharmanska al @cite_6 . Furthermore, Wang and Ji @cite_14 proposed two different loss functions that exploit privileged information and can be used with any classifier. The first model encoded privileged information as an additional feature during training, while the second approach considered that privileged information can be represented as secondary labels. An interesting method that discusses the auxiliary view ( , auxiliary information) from an information theoretic perspective was introduced by Motiian al @cite_9 and was also extended to unsupervised domain transfer @cite_40 . Lapin @cite_0 related the privileged information framework to the importance of sample weighting and showed that prior knowledge can be encoded using weights in a regular SVM. Recently, the LUPI paradigm has been employed with applications on gender classification facial expression recognition as well as age and height estimation @cite_5 @cite_32 @cite_16 @cite_12 . | {
"cite_N": [
"@cite_14",
"@cite_9",
"@cite_32",
"@cite_6",
"@cite_0",
"@cite_40",
"@cite_5",
"@cite_16",
"@cite_12"
],
"mid": [
"2100913937",
"2473618152",
"",
"2126942721",
"2169408065",
"2520426842",
"",
"",
""
],
"abstract": [
"Self-paced learning (SPL) is a recently proposed learning regime inspired by the learning process of humans and animals that gradually incorporates easy to more complex samples into training. Existing methods are limited in that they ignore an important aspect in learning: diversity. To incorporate this information, we propose an approach called self-paced learning with diversity (SPLD) which formalizes the preference for both easy and diverse samples into a general regularizes This regularization term is independent of the learning objective, and thus can be easily generalized into various learning tasks. Albeit non-convex, the optimization of the variables included in this SPLD regularization term for sample selection can be globally solved in linearithmic time. We demonstrate that our method significantly outperforms the conventional SPL on three real-world datasets. Specifically, SPLD achieves the best MAP so far reported in literature on the Hollywood2 and Olympic Sports datasets.",
"We explore the visual recognition problem from a main data view when an auxiliary data view is available during training. This is important because it allows improving the training of visual classifiers when paired additional data is cheaply available, and it improves the recognition from multi-view data when there is a missing view at testing time. The problem is challenging because of the intrinsic asymmetry caused by the missing auxiliary view during testing. We account for such view during training by extending the information bottleneck method, and by combining it with risk minimization. In this way, we establish an information theoretic principle for leaning any type of visual classifier under this particular setting. We use this principle to design a large-margin classifier with an efficient optimization in the primal space. We extensively compare our method with the state-of-the-art on different visual recognition datasets, and with different types of auxiliary data, and show that the proposed framework has a very promising potential.",
"",
"Many computer vision problems have an asymmetric distribution of information between training and test time. In this work, we study the case where we are given additional information about the training data, which however will not be available at test time. This situation is called learning using privileged information (LUPI). We introduce two maximum-margin techniques that are able to make use of this additional source of information, and we show that the framework is applicable to several scenarios that have been studied in computer vision before. Experiments with attributes, bounding boxes, image tags and rationales as additional information in object classification show promising results.",
"Abstract Prior knowledge can be used to improve predictive performance of learning algorithms or reduce the amount of data required for training. The same goal is pursued within the learning using privileged information paradigm which was recently introduced by and is aimed at utilizing additional information available only at training time—a framework implemented by SVM+. We relate the privileged information to importance weighting and show that the prior knowledge expressible with privileged features can also be encoded by weights associated with every training example. We show that a weighted SVM can always replicate an SVM+ solution, while the converse is not true and we construct a counterexample highlighting the limitations of SVM+. Finally, we touch on the problem of choosing weights for weighted SVMs when privileged features are not available.",
"We address the unsupervised domain adaptation problem for visual recognition when an auxiliary data view is available during training. This is important because it allows improving the training of visual classifiers on a new target visual domain when paired additional source data is cheaply available. This is the case when we learn from a source of RGB plus depth data, for then test on a new RGB domain. The problem is challenging because of the intrinsic asymmetry caused by the missing auxiliary view during testing and from which discriminative information should be carried over to the new domain. We jointly account for the auxiliary view during training and for the domain shift by extending the information bottleneck method, and by combining it with risk minimization. In this way, we establish an information theoretic principle for learning any type of visual classifier under this particular settings. We use this principle to design a multi-class large-margin classifier with an efficient optimization in the primal space. We extensively compare our method with the state-of-the-art on several datasets, by effectively learning from RGB plus depth data to recognize objects and gender from a new RGB domain.",
"",
"",
""
]
} |
1708.09083 | 2751833026 | Incorporating additional knowledge in the learning process can be beneficial for several computer vision and machine learning tasks. Whether privileged information originates from a source domain that is adapted to a target domain, or as additional features available at training time only, using such privileged (i.e., auxiliary) information is of high importance as it improves the recognition performance and generalization. However, both primary and privileged information are rarely derived from the same distribution, which poses an additional challenge to the recognition task. To address these challenges, we present a novel learning paradigm that leverages privileged information in a domain adaptation setup to perform visual recognition tasks. The proposed framework, named Adaptive SVM+, combines the advantages of both the learning using privileged information (LUPI) paradigm and the domain adaptation framework, which are naturally embedded in the objective function of a regular SVM. We demonstrate the effectiveness of our approach on the publicly available Animals with Attributes and INTERACT datasets and report state-of-the-art results in both of them. | : Transfer learning seeks to leverage the knowledge obtained while learning some tasks and applying it to new unseen, and possibly unrelated, tasks. It has been applied with great success in applications ranging from cross-domain setups @cite_2 @cite_28 @cite_4 , to facial action unit detection in transductive learning setup @cite_8 , to deep neural networks @cite_3 @cite_1 . Lopez-Paz al @cite_31 introduced generalized distillation, a method that unifies the LUPI framework with the knowledge distillation paradigm. Bengio al @cite_22 argued that instead of employing samples at random it is better to present samples organized in a meaningful way so that less complex examples are presented first. Curriculum learning @cite_22 @cite_7 @cite_37 , which is the learning strategy that implements this paradigm, employs the prior knowledge learned from the first easier'' tasks to improve the performance on harder'' ones that are learned at a later stage. Such a learning process can exploit prior knowledge to improve subsequent classification tasks but it does not scale up to many tasks since each subsequent task has to be learned individually. | {
"cite_N": [
"@cite_37",
"@cite_4",
"@cite_22",
"@cite_7",
"@cite_8",
"@cite_28",
"@cite_1",
"@cite_3",
"@cite_2",
"@cite_31"
],
"mid": [
"",
"",
"2952897246",
"",
"2008635359",
"",
"",
"1821462560",
"1981658663",
"2253986341"
],
"abstract": [
"",
"",
"Sharing information between multiple tasks enables algorithms to achieve good generalization performance even from small amounts of training data. However, in a realistic scenario of multi-task learning not all tasks are equally related to each other, hence it could be advantageous to transfer information only between the most related tasks. In this work we propose an approach that processes multiple tasks in a sequence with sharing between subsequent tasks instead of solving all tasks jointly. Subsequently, we address the question of curriculum learning of tasks, i.e. finding the best order of tasks to be learned. Our approach is based on a generalization bound criterion for choosing the task order that optimizes the average expected classification performance over all tasks. Our experimental results show that learning multiple related tasks sequentially can be more effective than learning them jointly, the order in which tasks are being solved affects the overall performance, and that our model is able to automatically discover the favourable order of tasks.",
"",
"Automatic facial action unit (AFA) detection from video is a long-standing problem in facial expression analysis. Most approaches emphasize choices of features and classifiers. They neglect individual differences in target persons. People vary markedly in facial morphology (e.g., heavy versus delicate brows, smooth versus deeply etched wrinkles) and behavior. Individual differences can dramatically influence how well generic classifiers generalize to previously unseen persons. While a possible solution would be to train person-specific classifiers, that often is neither feasible nor theoretically compelling. The alternative that we propose is to personalize a generic classifier in an unsupervised manner (no additional labels for the test subjects are required). We introduce a transductive learning method, which we refer to Selective Transfer Machine (STM), to personalize a generic classifier by attenuating person-specific biases. STM achieves this effect by simultaneously learning a classifier and re-weighting the training samples that are most relevant to the test subject. To evaluate the effectiveness of STM, we compared STM to generic classifiers and to cross-domain learning methods in three major databases: CK+, GEMEP-FERA and RU-FACS. STM outperformed generic classifiers in all.",
"",
"",
"A very simple way to improve the performance of almost any machine learning algorithm is to train many different models on the same data and then to average their predictions. Unfortunately, making predictions using a whole ensemble of models is cumbersome and may be too computationally expensive to allow deployment to a large number of users, especially if the individual models are large neural nets. Caruana and his collaborators have shown that it is possible to compress the knowledge in an ensemble into a single model which is much easier to deploy and we develop this approach further using a different compression technique. We achieve some surprising results on MNIST and we show that we can significantly improve the acoustic model of a heavily used commercial system by distilling the knowledge in an ensemble of models into a single model. We also introduce a new type of ensemble composed of one or more full models and many specialist models which learn to distinguish fine-grained classes that the full models confuse. Unlike a mixture of experts, these specialist models can be trained rapidly and in parallel.",
"In this paper, we propose a new framework called domain adaptation machine (DAM) for the multiple source domain adaption problem. Under this framework, we learn a robust decision function (referred to as target classifier) for label prediction of instances from the target domain by leveraging a set of base classifiers which are prelearned by using labeled instances either from the source domains or from the source domains and the target domain. With the base classifiers, we propose a new domain-dependent regularizer based on smoothness assumption, which enforces that the target classifier shares similar decision values with the relevant base classifiers on the unlabeled instances from the target domain. This newly proposed regularizer can be readily incorporated into many kernel methods (e.g., support vector machines (SVM), support vector regression, and least-squares SVM (LS-SVM)). For domain adaptation, we also develop two new domain adaptation methods referred to as FastDAM and UniverDAM. In FastDAM, we introduce our proposed domain-dependent regularizer into LS-SVM as well as employ a sparsity regularizer to learn a sparse target classifier with the support vectors only from the target domain, which thus makes the label prediction on any test instance very fast. In UniverDAM, we additionally make use of the instances from the source domains as Universum to further enhance the generalization ability of the target classifier. We evaluate our two methods on the challenging TRECIVD 2005 dataset for the large-scale video concept detection task as well as on the 20 newsgroups and email spam datasets for document retrieval. Comprehensive experiments demonstrate that FastDAM and UniverDAM outperform the existing multiple source domain adaptation methods for the two applications.",
"Distillation (, 2015) and privileged information (Vapnik & Izmailov, 2015) are two techniques that enable machines to learn from other machines. This paper unifies these two techniques into generalized distillation, a framework to learn from multiple machines and data representations. We provide theoretical and causal insight about the inner workings of generalized distillation, extend it to unsupervised, semisupervised and multitask learning scenarios, and illustrate its efficacy on a variety of numerical simulations on both synthetic and real-world data."
]
} |
1708.09083 | 2751833026 | Incorporating additional knowledge in the learning process can be beneficial for several computer vision and machine learning tasks. Whether privileged information originates from a source domain that is adapted to a target domain, or as additional features available at training time only, using such privileged (i.e., auxiliary) information is of high importance as it improves the recognition performance and generalization. However, both primary and privileged information are rarely derived from the same distribution, which poses an additional challenge to the recognition task. To address these challenges, we present a novel learning paradigm that leverages privileged information in a domain adaptation setup to perform visual recognition tasks. The proposed framework, named Adaptive SVM+, combines the advantages of both the learning using privileged information (LUPI) paradigm and the domain adaptation framework, which are naturally embedded in the objective function of a regular SVM. We demonstrate the effectiveness of our approach on the publicly available Animals with Attributes and INTERACT datasets and report state-of-the-art results in both of them. | @cite_34 : In this section, we provide some theoretical background on Adaptive SVM @cite_34 and highlight its differences from a regular SVM, and then we formulate SVM+ @cite_36 , which employs the LUPI paradigm. In the standard paradigm of supervised learning for binary classification, the training set consists of (N ) tuples of feature vectors ( x _i ), along with their respective labels (y_i ), represented as ( ( x _i, y_i) _ i=1 ^ N , x _i R ^d ), where (d ) is the number of features of each sample and (y_i -1,+1 ) . The standard SVM classifier finds a maximum-margin separating hyperplane between the two classes and solves the following constrained optimization problem: where ( w ) represents the weight vector, (|| w ||^2 ) is the size of the margin, (b ) is the bias parameter, ( ) is the slack variable for one training sample that indicates the deviation from the margin borders and (C ) denotes the penalty parameter. | {
"cite_N": [
"@cite_36",
"@cite_34"
],
"mid": [
"2506404347",
"1978920452"
],
"abstract": [
"In the Afterword to the second edition of the book \"Estimation of Dependences Based on Empirical Data\" by V. Vapnik, an advanced learning paradigm called Learning Using Hidden Information (LUHI) was introduced. This Afterword also suggested an extension of the SVM method (the so called SVM γ + method) to implement algorithms which address the LUHI paradigm (Vapnik, 1982-2006, Sections 2.4.2 and 2.5.3 of the Afterword). See also (Vapnik, Vashist, & Pavlovitch, 2008, 2009) for further development of the algorithms. In contrast to the existing machine learning paradigm where a teacher does not play an important role, the advanced learning paradigm considers some elements of human teaching. In the new paradigm along with examples, a teacher can provide students with hidden information that exists in explanations, comments, comparisons, and so on. This paper discusses details of the new paradigm 1 and corresponding algorithms, introduces some new algorithms, considers several specific forms of privileged information, demonstrates superiority of the new learning paradigm over the classical learning paradigm when solving practical problems, and discusses general questions related to the new ideas.",
"Many multimedia applications can benefit from techniques for adapting existing classifiers to data with different distributions. One example is cross-domain video concept detection which aims to adapt concept classifiers across various video domains. In this paper, we explore two key problems for classifier adaptation: (1) how to transform existing classifier(s) into an effective classifier for a new dataset that only has a limited number of labeled examples, and (2) how to select the best existing classifier(s) for adaptation. For the first problem, we propose Adaptive Support Vector Machines (A-SVMs) as a general method to adapt one or more existing classifiers of any type to the new dataset. It aims to learn the \"delta function\" between the original and adapted classifier using an objective function similar to SVMs. For the second problem, we estimate the performance of each existing classifier on the sparsely-labeled new dataset by analyzing its score distribution and other meta features, and select the classifiers with the best estimated performance. The proposed method outperforms several baseline and competing methods in terms of classification accuracy and efficiency in cross-domain concept detection in the TRECVID corpus."
]
} |
1708.09006 | 2753371338 | An efficient, fully automatic method for 3D face shape and pose estimation in unconstrained 2D imagery is presented. The proposed method jointly estimates a dense set of 3D landmarks and facial geometry using a single pass of a modified version of the popular "U-Net" neural network architecture. Additionally, we propose a method for directly estimating a set of 3D Morphable Model (3DMM) parameters, using the estimated 3D landmarks and geometry as constraints in a simple linear system. Qualitative modeling results are presented, as well as quantitative evaluation of predicted 3D face landmarks in unconstrained video sequences. | Many approaches to in the wild'' 3D face modeling leverage a parameterized model of 3D face geometry and optimize the parameter values based on 2D image-level constraints. By far the most popular such model is the 3D Morphable Model (3DMM) of Blanz and Vetter @cite_8 . The 3DMM represents deviations from a canonical mean face'' geometry and appearance using a linear model based on principal components analysis (PCA) of 200 3D face scans. Recently, @cite_1 released a larger scale 3DMM based on 10,000 face scans with improved ethnic and age diversity. | {
"cite_N": [
"@cite_1",
"@cite_8"
],
"mid": [
"2431101926",
"2237250383"
],
"abstract": [
"We present Large Scale Facial Model (LSFM) — a 3D Morphable Model (3DMM) automatically constructed from 9,663 distinct facial identities. To the best of our knowledge LSFM is the largest-scale Morphable Model ever constructed, containing statistical information from a huge variety of the human population. To build such a large model we introduce a novel fully automated and robust Morphable Model construction pipeline. The dataset that LSFM is trained on includes rich demographic information about each subject, allowing for the construction of not only a global 3DMM but also models tailored for specific age, gender or ethnicity groups. As an application example, we utilise the proposed model to perform age classification from 3D shape alone. Furthermore, we perform a systematic analysis of the constructed 3DMMs that showcases their quality and descriptive power. The presented extensive qualitative and quantitative evaluations reveal that the proposed 3DMM achieves state-of-the-art results, outperforming existing models by a large margin. Finally, for the benefit of the research community, we make publicly available the source code of the proposed automatic 3DMM construction pipeline. In addition, the constructed global 3DMM and a variety of bespoke models tailored by age, gender and ethnicity are available on application to researchers involved in medically oriented research.",
"In this paper, a new technique for modeling textured 3D faces is introduced. 3D faces can either be generated automatically from one or more photographs, or modeled directly through an intuitive user interface. Users are assisted in two key problems of computer aided face modeling. First, new face images or new 3D face models can be registered automatically by computing dense one-to-one correspondence to an internal face model. Second, the approach regulates the naturalness of modeled faces avoiding faces with an “unlikely” appearance. Starting from an example set of 3D face models, we derive a morphable face model by transforming the shape and texture of the examples into a vector space representation. New faces and expressions can be modeled by forming linear combinations of the prototypes. Shape and texture constraints derived from the statistics of our example faces are used to guide manual modeling or automated matching algorithms. We show 3D face reconstructions from single images and their applications for photo-realistic image manipulations. We also demonstrate face manipulations according to complex parameters such as gender, fullness of a face or its distinctiveness."
]
} |
1708.09006 | 2753371338 | An efficient, fully automatic method for 3D face shape and pose estimation in unconstrained 2D imagery is presented. The proposed method jointly estimates a dense set of 3D landmarks and facial geometry using a single pass of a modified version of the popular "U-Net" neural network architecture. Additionally, we propose a method for directly estimating a set of 3D Morphable Model (3DMM) parameters, using the estimated 3D landmarks and geometry as constraints in a simple linear system. Qualitative modeling results are presented, as well as quantitative evaluation of predicted 3D face landmarks in unconstrained video sequences. | The originally proposed approach to 3DMM coefficient estimation @cite_8 relied on a non-linear optimization based on minimizing the differences between predicted and observed image intensities. Because of this, the method required careful initialization and was computationally expensive to optimize. More recent optimization approaches @cite_6 @cite_14 rely on sparse 2D landmarks, which can be computed efficiently and robustly @cite_3 . Because the detected landmarks are inherently 2D, however, these methods must account for occlusion and pose effects by incorporating an iterative optimization that alternates between updating camera, pose and geometry parameters. | {
"cite_N": [
"@cite_14",
"@cite_3",
"@cite_6",
"@cite_8"
],
"mid": [
"2605638678",
"2087681821",
"",
"2237250383"
],
"abstract": [
"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.",
"This paper addresses the problem of Face Alignment for a single image. We show how an ensemble of regression trees can be used to estimate the face's landmark positions directly from a sparse subset of pixel intensities, achieving super-realtime performance with high quality predictions. We present a general framework based on gradient boosting for learning an ensemble of regression trees that optimizes the sum of square error loss and naturally handles missing or partially labelled data. We show how using appropriate priors exploiting the structure of image data helps with efficient feature selection. Different regularization strategies and its importance to combat overfitting are also investigated. In addition, we analyse the effect of the quantity of training data on the accuracy of the predictions and explore the effect of data augmentation using synthesized data.",
"",
"In this paper, a new technique for modeling textured 3D faces is introduced. 3D faces can either be generated automatically from one or more photographs, or modeled directly through an intuitive user interface. Users are assisted in two key problems of computer aided face modeling. First, new face images or new 3D face models can be registered automatically by computing dense one-to-one correspondence to an internal face model. Second, the approach regulates the naturalness of modeled faces avoiding faces with an “unlikely” appearance. Starting from an example set of 3D face models, we derive a morphable face model by transforming the shape and texture of the examples into a vector space representation. New faces and expressions can be modeled by forming linear combinations of the prototypes. Shape and texture constraints derived from the statistics of our example faces are used to guide manual modeling or automated matching algorithms. We show 3D face reconstructions from single images and their applications for photo-realistic image manipulations. We also demonstrate face manipulations according to complex parameters such as gender, fullness of a face or its distinctiveness."
]
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.