Split (3)

train · 30.4k rows

aid stringlengths 9 15	mid stringlengths 7 10	abstract stringlengths 78 2.56k	related_work stringlengths 92 1.77k	ref_abstract dict
1905.10777	2946059079	Recent deep learning based face recognition methods have achieved great performance, but it still remains challenging to recognize very low-resolution query face like 28x28 pixels when CCTV camera is far from the captured subject. Such face with very low-resolution is totally out of detail information of the face identity compared to normal resolution in a gallery and hard to find corresponding faces therein. To this end, we propose a Resolution Invariant Model (RIM) for addressing such cross-resolution face recognition problems, with three distinct novelties. First, RIM is a novel and unified deep architecture, containing a Face Hallucination sub-Net (FHN) and a Heterogeneous Recognition sub-Net (HRN), which are jointly learned end to end. Second, FHN is a well-designed tri-path Generative Adversarial Network (GAN) which simultaneously perceives facial structure and geometry prior information, i.e. landmark heatmaps and parsing maps, incorporated with an unsupervised cross-domain adversarial training strategy to super-resolve very low-resolution query image to its 8x larger ones without requiring them to be well aligned. Third, HRN is a generic Convolutional Neural Network (CNN) for heterogeneous face recognition with our proposed residual knowledge distillation strategy for learning discriminative yet generalized feature representation. Quantitative and qualitative experiments on several benchmarks demonstrate the superiority of the proposed model over the state-of-the-arts. Codes and models will be released upon acceptance.	Knowledge distillation @cite_4 is one of the most efficient methods for model compression and knowledge transfer, which aims at training a smaller network to mimic a more complex teacher network. Most knowledge methods typically apply one teacher network to supervise one student network. For instance, Ashok @cite_30 use reinforcement learning to prune student network under the guidance of teacher network. Belagiannis @cite_3 apply adversarial strategy to knowledge distillation. They further utilize a discriminator to measure whether student model and teacher model are close enough. It is worth considering, however, that there is still a certain discrepancy between the learning capacity of teacher network and student network. Inspired by residual representation @cite_6 , we adopt an additional teacher assistant network to learn the representation gap between teacher network and student network at different-level features.	{ "cite_N": [ "@cite_30", "@cite_4", "@cite_6", "@cite_3" ], "mid": [ "2756085244", "1821462560", "2194775991", "2963341152" ], "abstract": [ "While bigger and deeper neural network architectures continue to advance the state-of-the-art for many computer vision tasks, real-world adoption of these networks is impeded by hardware and speed constraints. Conventional model compression methods attempt to address this problem by modifying the architecture manually or using pre-defined heuristics. Since the space of all reduced architectures is very large, modifying the architecture of a deep neural network in this way is a difficult task. In this paper, we tackle this issue by introducing a principled method for learning reduced network architectures in a data-driven way using reinforcement learning. Our approach takes a larger teacher' network as input and outputs a compressed student' network derived from the teacher' network. In the first stage of our method, a recurrent policy network aggressively removes layers from the large teacher' model. In the second stage, another recurrent policy network carefully reduces the size of each remaining layer. The resulting network is then evaluated to obtain a reward -- a score based on the accuracy and compression of the network. Our approach uses this reward signal with policy gradients to train the policies to find a locally optimal student network. Our experiments show that we can achieve compression rates of more than 10x for models such as ResNet-34 while maintaining similar performance to the input teacher' network. We also present a valuable transfer learning result which shows that policies which are pre-trained on smaller teacher' networks can be used to rapidly speed up training on larger teacher' networks.", "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.", "Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers—8× deeper than VGG nets [40] but still having lower complexity. An ensemble of these residual nets achieves 3.57 error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28 relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions1, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.", "Neural network compression has recently received much attention due to the computational requirements of modern deep models. In this work, our objective is to transfer knowledge from a deep and accurate model to a smaller one. Our contributions are threefold: (i) we propose an adversarial network compression approach to train the small student network to mimic the large teacher, without the need for labels during training; (ii) we introduce a regularization scheme to prevent a trivially-strong discriminator without reducing the network capacity and (iii) our approach generalizes on different teacher-student models." ] }
1905.10822	2945886414	Face performance capture and reenactment techniques use multiple cameras and sensors, positioned at a distance from the face or mounted on heavy wearable devices. This limits their applications in mobile and outdoor environments. We present EgoFace, a radically new lightweight setup for face performance capture and front-view videorealistic reenactment using a single egocentric RGB camera. Our lightweight setup allows operations in uncontrolled environments, and lends itself to telepresence applications such as video-conferencing from dynamic environments. The input image is projected into a low dimensional latent space of the facial expression parameters. Through careful adversarial training of the parameter-space synthetic rendering, a videorealistic animation is produced. Our problem is challenging as the human visual system is sensitive to the smallest face irregularities that could occur in the final results. This sensitivity is even stronger for video results. Our solution is trained in a pre-processing stage, through a supervised manner without manual annotations. EgoFace captures a wide variety of facial expressions, including mouth movements and asymmetrical expressions. It works under varying illuminations, background, movements, handles people from different ethnicities and can operate in real time.	Face capture and reenactment techniques can be classified into two categories; using wearable devices such as VR headsets @cite_2 @cite_22 @cite_19 @cite_16 @cite_30 @cite_37 and using non-wearable devices @cite_25 @cite_18 @cite_36 @cite_29 @cite_23 @cite_20 . Wearable device techniques deal with partial facial occlusions caused by the head mounted displays. On the other hand, techniques for non-wearable devices require full facial coverage at once. We focus on methods using consumer level setups and sensors such as RGB, IR and depth cameras. Techniques utilizing production-level setups are out the scope of this work. Please refer to @cite_17 for more detail.	{ "cite_N": [ "@cite_30", "@cite_37", "@cite_18", "@cite_22", "@cite_36", "@cite_29", "@cite_19", "@cite_23", "@cite_2", "@cite_16", "@cite_25", "@cite_20", "@cite_17" ], "mid": [ "2811426698", "2886439287", "2806833697", "2039239232", "2301937176", "2004068758", "2556634673", "2211242044", "2810084975", "2890415768", "2806609126", "2135666716", "1961571273" ], "abstract": [ "We introduce Deep Appearance Models for capturing human facial appearance with a multi-view camera system, real-time realistic rendering in VR, and performance-driven animation from HMD-mounted cameras.", "Facial performance capture is used for animation production that projects a performer's facial expression to a computer graphics model. Retro-reflective markers and cameras are widely used for the performance capture. To capture expressions, we need to place markers on the performer's face and calibrate the intrinsic and extrinsic parameters of cameras in advance. However, the measurable space is limited to the calibrated area. In this study, we propose a system to capture facial performance using a smart eyewear with photo-reflective sensors and machine learning technique. Also, we show a result of principal components analysis of facial geometry to determine a good estimation parameter set.", "Our novel deep video portrait approach enables full control over a target actor by transferring head pose, facial expressions, and eye motion with a high level of photorealism.", "There are currently no solutions for enabling direct face-to-face interaction between virtual reality (VR) users wearing head-mounted displays (HMDs). The main challenge is that the headset obstructs a significant portion of a user's face, preventing effective facial capture with traditional techniques. To advance virtual reality as a next-generation communication platform, we develop a novel HMD that enables 3D facial performance-driven animation in real-time. Our wearable system uses ultra-thin flexible electronic materials that are mounted on the foam liner of the headset to measure surface strain signals corresponding to upper face expressions. These strain signals are combined with a head-mounted RGB-D camera to enhance the tracking in the mouth region and to account for inaccurate HMD placement. To map the input signals to a 3D face model, we perform a single-instance offline training session for each person. For reusable and accurate online operation, we propose a short calibration step to readjust the Gaussian mixture distribution of the mapping before each use. The resulting animations are visually on par with cutting-edge depth sensor-driven facial performance capture systems and hence, are suitable for social interactions in virtual worlds.", "We present a novel approach for real-time facial reenactment of a monocular target video sequence (e.g., Youtube video). The source sequence is also a monocular video stream, captured live with a commodity webcam. Our goal is to animate the facial expressions of the target video by a source actor and re-render the manipulated output video in a photo-realistic fashion. To this end, we first address the under-constrained problem of facial identity recovery from monocular video by non-rigid model-based bundling. At run time, we track facial expressions of both source and target video using a dense photometric consistency measure. Reenactment is then achieved by fast and efficient deformation transfer between source and target. The mouth interior that best matches the re-targeted expression is retrieved from the target sequence and warped to produce an accurate fit. Finally, we convincingly re-render the synthesized target face on top of the corresponding video stream such that it seamlessly blends with the real-world illumination. We demonstrate our method in a live setup, where Youtube videos are reenacted in real time.", "We present a method for the real-time transfer of facial expressions from an actor in a source video to an actor in a target video, thus enabling the ad-hoc control of the facial expressions of the target actor. The novelty of our approach lies in the transfer and photorealistic re-rendering of facial deformations and detail into the target video in a way that the newly-synthesized expressions are virtually indistinguishable from a real video. To achieve this, we accurately capture the facial performances of the source and target subjects in real-time using a commodity RGB-D sensor. For each frame, we jointly fit a parametric model for identity, expression, and skin reflectance to the input color and depth data, and also reconstruct the scene lighting. For expression transfer, we compute the difference between the source and target expressions in parameter space, and modify the target parameters to match the source expressions. A major challenge is the convincing re-rendering of the synthesized target face into the corresponding video stream. This requires a careful consideration of the lighting and shading design, which both must correspond to the real-world environment. We demonstrate our method in a live setup, where we modify a video conference feed such that the facial expressions of a different person (e.g., translator) are matched in real-time.", "Significant challenges currently prohibit expressive interaction in virtual reality (VR). Occlusions introduced by head-mounted displays (HMDs) make existing facial tracking techniques intractable, and even state-of-the-art techniques used for real-time facial tracking in unconstrained environments fail to capture subtle details of the user's facial expressions that are essential for compelling speech animation. We introduce a novel system for HMD users to control a digital avatar in real-time while producing plausible speech animation and emotional expressions. Using a monocular camera attached to an HMD, we record multiple subjects performing various facial expressions and speaking several phonetically-balanced sentences. These images are used with artist-generated animation data corresponding to these sequences to train a convolutional neural network (CNN) to regress images of a user's mouth region to the parameters that control a digital avatar. To make training this system more tractable, we use audio-based alignment techniques to map images of multiple users making the same utterance to the corresponding animation parameters. We demonstrate that this approach is also feasible for tracking the expressions around the user's eye region with an internal infrared (IR) camera, thereby enabling full facial tracking. This system requires no user-specific calibration, uses easily obtainable consumer hardware, and produces high-quality animations of speech and emotional expressions. Finally, we demonstrate the quality of our system on a variety of subjects and evaluate its performance against state-of-the-art real-time facial tracking techniques.", "We reconstruct a controllable model of a person from a large photo collection that captures his or her persona, i.e., physical appearance and behavior. The ability to operate on unstructured photo collections enables modeling a huge number of people, including celebrities and other well photographed people without requiring them to be scanned. Moreover, we show the ability to drive or puppeteer the captured person B using any other video of a different person A. In this scenario, B acts out the role of person A, but retains his her own personality and character. Our system is based on a novel combination of 3D face reconstruction, tracking, alignment, and multi-texture modeling, applied to the puppeteering problem. We demonstrate convincing results on a large variety of celebrities derived from Internet imagery and video.", "FaceVR virtually removes VR-goggles using facial reenactment, which is of paramount importance in a VR-teleconferencing scenario. To this end, we transfer expressions as well as eye motions to a stereo target video.", "We propose a new approach for 3D reconstruction of dynamic indoor and outdoor scenes in everyday environments, leveraging only cameras worn by a user. This approach allows 3D reconstruction of experiences at any location and virtual tours from anywhere. The key innovation of the proposed ego-centric reconstruction system is to capture the wearer's body pose and facial expression from near-body views, e.g. cameras on the user's glasses, and to capture the surrounding environment using outward-facing views. The main challenge of the ego-centric reconstruction, however, is the poor coverage of the near-body views – that is, the user's body and face are observed from vantage points that are convenient for wear but inconvenient for capture. To overcome these challenges, we propose a parametric-model-based approach to user motion estimation. This approach utilizes convolutional neural networks (CNNs) for near-view body pose estimation, and we introduce a CNN-based approach for facial expression estimation that combines audio and video. For each time-point during capture, the intermediate model-based reconstructions from these systems are used to re-target a high-fidelity pre-scanned model of the user. We demonstrate that the proposed self-sufficient, head-worn capture system is capable of reconstructing the wearer's movements and their surrounding environment in both indoor and outdoor situations without any additional views. As a proof of concept, we show how the resulting 3D-plus-time reconstruction can be immersively experienced within a virtual reality system (e.g., the HTC Vive). We expect that the size of the proposed egocentric capture-and-reconstruction system will eventually be reduced to fit within future AR glasses, and will be widely useful for immersive 3D telepresence, virtual tours, and general use-anywhere 3D content creation.", "HeadOn is the first real-time source-to-target reenactment approach for human portrait videos that enables the transfer of torso and head motion, facial expression, and eye gaze.", "Face Transfer is a method for mapping videorecorded performances of one individual to facial animations of another. It extracts visemes (speech-related mouth articulations), expressions, and three-dimensional (3D) pose from monocular video or film footage. These parameters are then used to generate and drive a detailed 3D textured face mesh for a target identity, which can be seamlessly rendered back into target footage. The underlying face model automatically adjusts for how the target performs facial expressions and visemes. The performance data can be easily edited to change the visemes, expressions, pose, or even the identity of the target---the attributes are separably controllable. This supports a wide variety of video rewrite and puppetry applications.Face Transfer is based on a multilinear model of 3D face meshes that separably parameterizes the space of geometric variations due to different attributes (e.g., identity, expression, and viseme). Separability means that each of these attributes can be independently varied. A multilinear model can be estimated from a Cartesian product of examples (identities × expressions × visemes) with techniques from statistical analysis, but only after careful preprocessing of the geometric data set to secure one-to-one correspondence, to minimize cross-coupling artifacts, and to fill in any missing examples. Face Transfer offers new solutions to these problems and links the estimated model with a face-tracking algorithm to extract pose, expression, and viseme parameters.", "Facial appearance capture is now firmly established within academic research and used extensively across various application domains, perhaps most prominently in the entertainment industry through the design of virtual characters in video games and films. While significant progress has occurred over the last two decades, no single survey currently exists that discusses the similarities, differences, and practical considerations of the available appearance capture techniques as applied to human faces. A central difficulty of facial appearance capture is the way light interacts with skin-which has a complex multi-layered structure-and the interactions that occur below the skin surface can, by definition, only be observed indirectly. In this report, we distinguish between two broad strategies for dealing with this complexity. \"Image-based methods\" try to exhaustively capture the exact face appearance under different lighting and viewing conditions, and then render the face through weighted image combinations. \"Parametric methods\" instead fit the captured reflectance data to some parametric appearance model used during rendering, allowing for a more lightweight and flexible representation but at the cost of potentially increased rendering complexity or inexact reproduction. The goal of this report is to provide an overview that can guide practitioners and researchers in assessing the tradeoffs between current approaches and identifying directions for future advances in facial appearance capture." ] }
1905.10822	2945886414	Face performance capture and reenactment techniques use multiple cameras and sensors, positioned at a distance from the face or mounted on heavy wearable devices. This limits their applications in mobile and outdoor environments. We present EgoFace, a radically new lightweight setup for face performance capture and front-view videorealistic reenactment using a single egocentric RGB camera. Our lightweight setup allows operations in uncontrolled environments, and lends itself to telepresence applications such as video-conferencing from dynamic environments. The input image is projected into a low dimensional latent space of the facial expression parameters. Through careful adversarial training of the parameter-space synthetic rendering, a videorealistic animation is produced. Our problem is challenging as the human visual system is sensitive to the smallest face irregularities that could occur in the final results. This sensitivity is even stronger for video results. Our solution is trained in a pre-processing stage, through a supervised manner without manual annotations. EgoFace captures a wide variety of facial expressions, including mouth movements and asymmetrical expressions. It works under varying illuminations, background, movements, handles people from different ethnicities and can operate in real time.	@cite_29 presented the first real-time expression transfer approach using an RGB-D camera. For each frame, a parametric model for identity, expression, skin reflectance and lighting was fitted to the input. Expression is transferred by computing the difference between the source and target expressions in parameter space, and modifying the target parameters to match the source expressions. Later, Face2Face @cite_36 introduced expression transfer in real-time from a single RGB camera. Facial expressions of both source and target are tracked using a dense photometric consistency loss. Reenactment is achieved by fast and efficient deformation transfer between source and target. A learning-based approach is used to estimate the mouth interior for the retargeted expression.	{ "cite_N": [ "@cite_36", "@cite_29" ], "mid": [ "2301937176", "2004068758" ], "abstract": [ "We present a novel approach for real-time facial reenactment of a monocular target video sequence (e.g., Youtube video). The source sequence is also a monocular video stream, captured live with a commodity webcam. Our goal is to animate the facial expressions of the target video by a source actor and re-render the manipulated output video in a photo-realistic fashion. To this end, we first address the under-constrained problem of facial identity recovery from monocular video by non-rigid model-based bundling. At run time, we track facial expressions of both source and target video using a dense photometric consistency measure. Reenactment is then achieved by fast and efficient deformation transfer between source and target. The mouth interior that best matches the re-targeted expression is retrieved from the target sequence and warped to produce an accurate fit. Finally, we convincingly re-render the synthesized target face on top of the corresponding video stream such that it seamlessly blends with the real-world illumination. We demonstrate our method in a live setup, where Youtube videos are reenacted in real time.", "We present a method for the real-time transfer of facial expressions from an actor in a source video to an actor in a target video, thus enabling the ad-hoc control of the facial expressions of the target actor. The novelty of our approach lies in the transfer and photorealistic re-rendering of facial deformations and detail into the target video in a way that the newly-synthesized expressions are virtually indistinguishable from a real video. To achieve this, we accurately capture the facial performances of the source and target subjects in real-time using a commodity RGB-D sensor. For each frame, we jointly fit a parametric model for identity, expression, and skin reflectance to the input color and depth data, and also reconstruct the scene lighting. For expression transfer, we compute the difference between the source and target expressions in parameter space, and modify the target parameters to match the source expressions. A major challenge is the convincing re-rendering of the synthesized target face into the corresponding video stream. This requires a careful consideration of the lighting and shading design, which both must correspond to the real-world environment. We demonstrate our method in a live setup, where we modify a video conference feed such that the facial expressions of a different person (e.g., translator) are matched in real-time." ] }
1905.10822	2945886414	Face performance capture and reenactment techniques use multiple cameras and sensors, positioned at a distance from the face or mounted on heavy wearable devices. This limits their applications in mobile and outdoor environments. We present EgoFace, a radically new lightweight setup for face performance capture and front-view videorealistic reenactment using a single egocentric RGB camera. Our lightweight setup allows operations in uncontrolled environments, and lends itself to telepresence applications such as video-conferencing from dynamic environments. The input image is projected into a low dimensional latent space of the facial expression parameters. Through careful adversarial training of the parameter-space synthetic rendering, a videorealistic animation is produced. Our problem is challenging as the human visual system is sensitive to the smallest face irregularities that could occur in the final results. This sensitivity is even stronger for video results. Our solution is trained in a pre-processing stage, through a supervised manner without manual annotations. EgoFace captures a wide variety of facial expressions, including mouth movements and asymmetrical expressions. It works under varying illuminations, background, movements, handles people from different ethnicities and can operate in real time.	While the aforementioned approaches produce compelling results, they only control facial expressions. presented HeadOn @cite_25 , a technique allowing control of the 3D pose and gaze, aswell as the expressions. Here, a personalized geometry proxy that embeds a parametric head, eye, and kinematic torso model is used. While HeadOn generates very interesting results, it requires a depth camera. @cite_18 presented Deep Video Portraits, the first approach for controlling the 3D pose, facial expressions and gaze from a single RGB camera. Their approach is based on neural networks where they show generative adversarial networks can produce visually pleasing facial animations, inspired by the success of GANs @cite_1 . A number of approaches produce video reenactment even given a single target image @cite_10 @cite_6 @cite_15 . While @cite_6 @cite_15 can only control the face expressions, @cite_10 can reenact both pose and expressions of the examined person. All of these approaches assume a close to frontal face view, while we deal with a highly distorted and occluded view of the face.	{ "cite_N": [ "@cite_18", "@cite_1", "@cite_6", "@cite_15", "@cite_10", "@cite_25" ], "mid": [ "2806833697", "2963073614", "", "2949194588", "2902836694", "2806609126" ], "abstract": [ "Our novel deep video portrait approach enables full control over a target actor by transferring head pose, facial expressions, and eye motion with a high level of photorealism.", "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. Moreover, since the release of the pix2pix software associated with this paper, hundreds of twitter users have posted their own artistic experiments using our system. As a community, we no longer hand-engineer our mapping functions, and this work suggests we can achieve reasonable results without handengineering our loss functions either.", "", "The objective of this paper is a neural network model that controls the pose and expression of a given face, using another face or modality (e.g. audio). This model can then be used for lightweight, sophisticated video and image editing. We make the following three contributions. First, we introduce a network, X2Face, that can control a source face (specified by one or more frames) using another face in a driving frame to produce a generated frame with the identity of the source frame but the pose and expression of the face in the driving frame. Second, we propose a method for training the network fully self-supervised using a large collection of video data. Third, we show that the generation process can be driven by other modalities, such as audio or pose codes, without any further training of the network. The generation results for driving a face with another face are compared to state-of-the-art self-supervised supervised methods. We show that our approach is more robust than other methods, as it makes fewer assumptions about the input data. We also show examples of using our framework for video face editing.", "With the rising interest in personalized VR and gaming experiences comes the need to create high quality 3D avatars that are both low-cost and variegated. Due to this, building dynamic avatars from a single unconstrained input image is becoming a popular application. While previous techniques that attempt this require multiple input images or rely on transferring dynamic facial appearance from a source actor, we are able to do so using only one 2D input image without any form of transfer from a source image. We achieve this using a new conditional Generative Adversarial Network design that allows fine-scale manipulation of any facial input image into a new expression while preserving its identity. Our photoreal avatar GAN (paGAN) can also synthesize the unseen mouth interior and control the eye-gaze direction of the output, as well as produce the final image from a novel viewpoint. The method is even capable of generating fully-controllable temporally stable video sequences, despite not using temporal information during training. After training, we can use our network to produce dynamic image-based avatars that are controllable on mobile devices in real time. To do this, we compute a fixed set of output images that correspond to key blendshapes, from which we extract textures in UV space. Using a subject's expression blendshapes at run-time, we can linearly blend these key textures together to achieve the desired appearance. Furthermore, we can use the mouth interior and eye textures produced by our network to synthesize on-the-fly avatar animations for those regions. Our work produces state-of-the-art quality image and video synthesis, and is the first to our knowledge that is able to generate a dynamically textured avatar with a mouth interior, all from a single image.", "HeadOn is the first real-time source-to-target reenactment approach for human portrait videos that enables the transfer of torso and head motion, facial expression, and eye gaze." ] }
1905.10822	2945886414	Face performance capture and reenactment techniques use multiple cameras and sensors, positioned at a distance from the face or mounted on heavy wearable devices. This limits their applications in mobile and outdoor environments. We present EgoFace, a radically new lightweight setup for face performance capture and front-view videorealistic reenactment using a single egocentric RGB camera. Our lightweight setup allows operations in uncontrolled environments, and lends itself to telepresence applications such as video-conferencing from dynamic environments. The input image is projected into a low dimensional latent space of the facial expression parameters. Through careful adversarial training of the parameter-space synthetic rendering, a videorealistic animation is produced. Our problem is challenging as the human visual system is sensitive to the smallest face irregularities that could occur in the final results. This sensitivity is even stronger for video results. Our solution is trained in a pre-processing stage, through a supervised manner without manual annotations. EgoFace captures a wide variety of facial expressions, including mouth movements and asymmetrical expressions. It works under varying illuminations, background, movements, handles people from different ethnicities and can operate in real time.	A number of facial reenactment techniques for wearable device have recently been proposed @cite_2 @cite_22 @cite_19 @cite_16 @cite_30 @cite_37 , most of which are designed for VR headsets @cite_2 @cite_22 @cite_19 @cite_30 . Such devices occlude the upper half of the face, and hence these techniques attempt to capture the full face through multiple sensors. Many approaches utilize an IR camera placed inside the headset to capture the eye movement @cite_2 @cite_22 @cite_19 . @cite_22 placed eight ultra-thin strain gauges (flexible met al foil sensors) on the foam liner of the headset. This measures surface strain caused by the upper facial expressions. An RGB-D camera is also used to capture the lower facial expressions. @cite_19 extended @cite_22 by exploiting the audio signal. Here, same utterance among different subjects is mapped to the same animation parameters and a single RGB camera is used to track the mouth movement.	{ "cite_N": [ "@cite_30", "@cite_37", "@cite_22", "@cite_19", "@cite_2", "@cite_16" ], "mid": [ "2811426698", "2886439287", "2039239232", "2556634673", "2810084975", "2890415768" ], "abstract": [ "We introduce Deep Appearance Models for capturing human facial appearance with a multi-view camera system, real-time realistic rendering in VR, and performance-driven animation from HMD-mounted cameras.", "Facial performance capture is used for animation production that projects a performer's facial expression to a computer graphics model. Retro-reflective markers and cameras are widely used for the performance capture. To capture expressions, we need to place markers on the performer's face and calibrate the intrinsic and extrinsic parameters of cameras in advance. However, the measurable space is limited to the calibrated area. In this study, we propose a system to capture facial performance using a smart eyewear with photo-reflective sensors and machine learning technique. Also, we show a result of principal components analysis of facial geometry to determine a good estimation parameter set.", "There are currently no solutions for enabling direct face-to-face interaction between virtual reality (VR) users wearing head-mounted displays (HMDs). The main challenge is that the headset obstructs a significant portion of a user's face, preventing effective facial capture with traditional techniques. To advance virtual reality as a next-generation communication platform, we develop a novel HMD that enables 3D facial performance-driven animation in real-time. Our wearable system uses ultra-thin flexible electronic materials that are mounted on the foam liner of the headset to measure surface strain signals corresponding to upper face expressions. These strain signals are combined with a head-mounted RGB-D camera to enhance the tracking in the mouth region and to account for inaccurate HMD placement. To map the input signals to a 3D face model, we perform a single-instance offline training session for each person. For reusable and accurate online operation, we propose a short calibration step to readjust the Gaussian mixture distribution of the mapping before each use. The resulting animations are visually on par with cutting-edge depth sensor-driven facial performance capture systems and hence, are suitable for social interactions in virtual worlds.", "Significant challenges currently prohibit expressive interaction in virtual reality (VR). Occlusions introduced by head-mounted displays (HMDs) make existing facial tracking techniques intractable, and even state-of-the-art techniques used for real-time facial tracking in unconstrained environments fail to capture subtle details of the user's facial expressions that are essential for compelling speech animation. We introduce a novel system for HMD users to control a digital avatar in real-time while producing plausible speech animation and emotional expressions. Using a monocular camera attached to an HMD, we record multiple subjects performing various facial expressions and speaking several phonetically-balanced sentences. These images are used with artist-generated animation data corresponding to these sequences to train a convolutional neural network (CNN) to regress images of a user's mouth region to the parameters that control a digital avatar. To make training this system more tractable, we use audio-based alignment techniques to map images of multiple users making the same utterance to the corresponding animation parameters. We demonstrate that this approach is also feasible for tracking the expressions around the user's eye region with an internal infrared (IR) camera, thereby enabling full facial tracking. This system requires no user-specific calibration, uses easily obtainable consumer hardware, and produces high-quality animations of speech and emotional expressions. Finally, we demonstrate the quality of our system on a variety of subjects and evaluate its performance against state-of-the-art real-time facial tracking techniques.", "FaceVR virtually removes VR-goggles using facial reenactment, which is of paramount importance in a VR-teleconferencing scenario. To this end, we transfer expressions as well as eye motions to a stereo target video.", "We propose a new approach for 3D reconstruction of dynamic indoor and outdoor scenes in everyday environments, leveraging only cameras worn by a user. This approach allows 3D reconstruction of experiences at any location and virtual tours from anywhere. The key innovation of the proposed ego-centric reconstruction system is to capture the wearer's body pose and facial expression from near-body views, e.g. cameras on the user's glasses, and to capture the surrounding environment using outward-facing views. The main challenge of the ego-centric reconstruction, however, is the poor coverage of the near-body views – that is, the user's body and face are observed from vantage points that are convenient for wear but inconvenient for capture. To overcome these challenges, we propose a parametric-model-based approach to user motion estimation. This approach utilizes convolutional neural networks (CNNs) for near-view body pose estimation, and we introduce a CNN-based approach for facial expression estimation that combines audio and video. For each time-point during capture, the intermediate model-based reconstructions from these systems are used to re-target a high-fidelity pre-scanned model of the user. We demonstrate that the proposed self-sufficient, head-worn capture system is capable of reconstructing the wearer's movements and their surrounding environment in both indoor and outdoor situations without any additional views. As a proof of concept, we show how the resulting 3D-plus-time reconstruction can be immersively experienced within a virtual reality system (e.g., the HTC Vive). We expect that the size of the proposed egocentric capture-and-reconstruction system will eventually be reduced to fit within future AR glasses, and will be widely useful for immersive 3D telepresence, virtual tours, and general use-anywhere 3D content creation." ] }
1905.10822	2945886414	Face performance capture and reenactment techniques use multiple cameras and sensors, positioned at a distance from the face or mounted on heavy wearable devices. This limits their applications in mobile and outdoor environments. We present EgoFace, a radically new lightweight setup for face performance capture and front-view videorealistic reenactment using a single egocentric RGB camera. Our lightweight setup allows operations in uncontrolled environments, and lends itself to telepresence applications such as video-conferencing from dynamic environments. The input image is projected into a low dimensional latent space of the facial expression parameters. Through careful adversarial training of the parameter-space synthetic rendering, a videorealistic animation is produced. Our problem is challenging as the human visual system is sensitive to the smallest face irregularities that could occur in the final results. This sensitivity is even stronger for video results. Our solution is trained in a pre-processing stage, through a supervised manner without manual annotations. EgoFace captures a wide variety of facial expressions, including mouth movements and asymmetrical expressions. It works under varying illuminations, background, movements, handles people from different ethnicities and can operate in real time.	@cite_2 presented FaceVR, a solution utilizing the optimization based infrastructure of @cite_36 @cite_29 . RGB-D cameras are used to build a personalized stereo avatar by solving for a face parametric model. During tracking, expressions, lighting and rigid pose are optimized. Eye-gaze tracking is performed through random ferns while data-driven solution is used for mouth retrieval, similar to Face2Face @cite_36 . Recently, @cite_30 presented a deep appearance model for human face rendering, and showed how to control it through a VR headset. The model is estimated using multi-view light stage data. The headset is equipped with 3 cameras, 1 for the mouth and 2 for the eyes. A learning-based approach controls the face model, trained with synthetic facial expression images. @cite_37 proposed an eye-wear for face capture. It uses 20 photo-reflective sensors, 16 placed on the glass frame and 4 in the lower face region. Mo-cap markers are placed on the face, and a relation is learned between their positions and the measurements of the photo-reflective sensors. At testing markers are removed and the learned model deforms a synthetic avatar. @cite_16 used two head-mounted cameras for face performance capture. Although interesting results are produced, the final output is a synthetic avatar.	{ "cite_N": [ "@cite_30", "@cite_37", "@cite_36", "@cite_29", "@cite_2", "@cite_16" ], "mid": [ "2811426698", "2886439287", "2301937176", "2004068758", "2810084975", "2890415768" ], "abstract": [ "We introduce Deep Appearance Models for capturing human facial appearance with a multi-view camera system, real-time realistic rendering in VR, and performance-driven animation from HMD-mounted cameras.", "Facial performance capture is used for animation production that projects a performer's facial expression to a computer graphics model. Retro-reflective markers and cameras are widely used for the performance capture. To capture expressions, we need to place markers on the performer's face and calibrate the intrinsic and extrinsic parameters of cameras in advance. However, the measurable space is limited to the calibrated area. In this study, we propose a system to capture facial performance using a smart eyewear with photo-reflective sensors and machine learning technique. Also, we show a result of principal components analysis of facial geometry to determine a good estimation parameter set.", "We present a novel approach for real-time facial reenactment of a monocular target video sequence (e.g., Youtube video). The source sequence is also a monocular video stream, captured live with a commodity webcam. Our goal is to animate the facial expressions of the target video by a source actor and re-render the manipulated output video in a photo-realistic fashion. To this end, we first address the under-constrained problem of facial identity recovery from monocular video by non-rigid model-based bundling. At run time, we track facial expressions of both source and target video using a dense photometric consistency measure. Reenactment is then achieved by fast and efficient deformation transfer between source and target. The mouth interior that best matches the re-targeted expression is retrieved from the target sequence and warped to produce an accurate fit. Finally, we convincingly re-render the synthesized target face on top of the corresponding video stream such that it seamlessly blends with the real-world illumination. We demonstrate our method in a live setup, where Youtube videos are reenacted in real time.", "We present a method for the real-time transfer of facial expressions from an actor in a source video to an actor in a target video, thus enabling the ad-hoc control of the facial expressions of the target actor. The novelty of our approach lies in the transfer and photorealistic re-rendering of facial deformations and detail into the target video in a way that the newly-synthesized expressions are virtually indistinguishable from a real video. To achieve this, we accurately capture the facial performances of the source and target subjects in real-time using a commodity RGB-D sensor. For each frame, we jointly fit a parametric model for identity, expression, and skin reflectance to the input color and depth data, and also reconstruct the scene lighting. For expression transfer, we compute the difference between the source and target expressions in parameter space, and modify the target parameters to match the source expressions. A major challenge is the convincing re-rendering of the synthesized target face into the corresponding video stream. This requires a careful consideration of the lighting and shading design, which both must correspond to the real-world environment. We demonstrate our method in a live setup, where we modify a video conference feed such that the facial expressions of a different person (e.g., translator) are matched in real-time.", "FaceVR virtually removes VR-goggles using facial reenactment, which is of paramount importance in a VR-teleconferencing scenario. To this end, we transfer expressions as well as eye motions to a stereo target video.", "We propose a new approach for 3D reconstruction of dynamic indoor and outdoor scenes in everyday environments, leveraging only cameras worn by a user. This approach allows 3D reconstruction of experiences at any location and virtual tours from anywhere. The key innovation of the proposed ego-centric reconstruction system is to capture the wearer's body pose and facial expression from near-body views, e.g. cameras on the user's glasses, and to capture the surrounding environment using outward-facing views. The main challenge of the ego-centric reconstruction, however, is the poor coverage of the near-body views – that is, the user's body and face are observed from vantage points that are convenient for wear but inconvenient for capture. To overcome these challenges, we propose a parametric-model-based approach to user motion estimation. This approach utilizes convolutional neural networks (CNNs) for near-view body pose estimation, and we introduce a CNN-based approach for facial expression estimation that combines audio and video. For each time-point during capture, the intermediate model-based reconstructions from these systems are used to re-target a high-fidelity pre-scanned model of the user. We demonstrate that the proposed self-sufficient, head-worn capture system is capable of reconstructing the wearer's movements and their surrounding environment in both indoor and outdoor situations without any additional views. As a proof of concept, we show how the resulting 3D-plus-time reconstruction can be immersively experienced within a virtual reality system (e.g., the HTC Vive). We expect that the size of the proposed egocentric capture-and-reconstruction system will eventually be reduced to fit within future AR glasses, and will be widely useful for immersive 3D telepresence, virtual tours, and general use-anywhere 3D content creation." ] }
1905.10628	2945456671	The ability of detecting out-of-distribution (OOD) samples is important to secure reliability of deep neural networks in real-world applications. Considering the nature of OOD samples, detection methods should not have hyperparameters whose optimal values vary sensitively depending on incoming OOD samples. This requirement is not met by many previous methods. In this paper, we propose a simple, hyperparameter-free method that is based on softmax of scaled cosine similarity. It resembles the approach employed by recent metric learning methods, but it differs in details; the differences are essential to achieve high detection performance. As compared with the current state-of-the-art methods, which needs hyperparameter tuning that could compromise real-world performance, the proposed method attains at least competitive detection accuracy even without (tuning of) a hyperparameter; furthermore, it is computationally more efficient, since it needs only a single forward pass unlike previous methods that need backpropagation for each input.	A more direct approach to the above issue is OOD detection. A baseline method that thresholds confidence score, i.e., the maximum softmax output, is evaluated in @cite_0 . Since then, many studies have been conducted. ODIN @cite_5 applies perturbation with a constant magnitude @math to an input @math in the direction of increasing the confidence score (i.e., the maximum softmax) and then uses the increased score in the same way as the baseline. An observation behind this procedure is that such perturbation tends to increase confidence score more for ID samples than for OOD samples. Rigorously, @math is perturbed to increase a temperature-scaled softmax value. In the experiments reported in @cite_5 , @math as well as the temperature are determined by using a portion of samples from a target OOD dataset; this is done for each pair of ID and OOD datasets.	{ "cite_N": [ "@cite_0", "@cite_5" ], "mid": [ "2531327146", "2963693742" ], "abstract": [ "We consider the two related problems of detecting if an example is misclassified or out-of-distribution. We present a simple baseline that utilizes probabilities from softmax distributions. Correctly classified examples tend to have greater maximum softmax probabilities than erroneously classified and out-of-distribution examples, allowing for their detection. We assess performance by defining several tasks in computer vision, natural language processing, and automatic speech recognition, showing the effectiveness of this baseline across all. We then show the baseline can sometimes be surpassed, demonstrating the room for future research on these underexplored detection tasks.", "We consider the problem of detecting out-of-distribution images in neural networks. We propose ODIN, a simple and effective method that does not require any change to a pre-trained neural network. Our method is based on the observation that using temperature scaling and adding small perturbations to the input can separate the softmax score distributions of in- and out-of-distribution images, allowing for more effective detection. We show in a series of experiments that ODIN is compatible with diverse network architectures and datasets. It consistently outperforms the baseline approach by a large margin, establishing a new state-of-the-art performance on this task. For example, ODIN reduces the false positive rate from the baseline 34.7 to 4.3 on the DenseNet (applied to CIFAR-10 and Tiny-ImageNet) when the true positive rate is 95 ." ] }
1905.10628	2945456671	The ability of detecting out-of-distribution (OOD) samples is important to secure reliability of deep neural networks in real-world applications. Considering the nature of OOD samples, detection methods should not have hyperparameters whose optimal values vary sensitively depending on incoming OOD samples. This requirement is not met by many previous methods. In this paper, we propose a simple, hyperparameter-free method that is based on softmax of scaled cosine similarity. It resembles the approach employed by recent metric learning methods, but it differs in details; the differences are essential to achieve high detection performance. As compared with the current state-of-the-art methods, which needs hyperparameter tuning that could compromise real-world performance, the proposed method attains at least competitive detection accuracy even without (tuning of) a hyperparameter; furthermore, it is computationally more efficient, since it needs only a single forward pass unlike previous methods that need backpropagation for each input.	The current state-of-the-art of OOD detection is achieved by the methods @cite_37 @cite_53 employing input perturbation similar to ODIN, although there are many other studies, such as generative models @cite_23 @cite_54 , a prior distribution @cite_46 , robustification by training networks to predict word embedding of class labels @cite_45 , and pretraining of networks @cite_19 @cite_56 .	{ "cite_N": [ "@cite_37", "@cite_54", "@cite_53", "@cite_56", "@cite_19", "@cite_45", "@cite_23", "@cite_46" ], "mid": [ "2889625178", "", "2867167548", "", "2904981516", "2888635935", "2963384319", "2963215553" ], "abstract": [ "As deep learning methods form a critical part in commercially important applications such as autonomous driving and medical diagnostics, it is important to reliably detect out-of-distribution (OOD) inputs while employing these algorithms. In this work, we propose an OOD detection algorithm which comprises of an ensemble of classifiers. We train each classifier in a self-supervised manner by leaving out a random subset of training data as OOD data and the rest as in-distribution (ID) data. We propose a novel margin-based loss over the softmax output which seeks to maintain at least a margin m between the average entropy of the OOD and in-distribution samples. In conjunction with the standard cross-entropy loss, we minimize the novel loss to train an ensemble of classifiers. We also propose a novel method to combine the outputs of the ensemble of classifiers to obtain OOD detection score and class prediction. Overall, our method convincingly outperforms [7] and the current state-of-the-art ODIN [13] on several OOD detection benchmarks.", "", "Detecting test samples drawn sufficiently far away from the training distribution statistically or adversarially is the fundamental requirement to deploy a good classifier in many real-world machine learning applications. However, deep neural networks with the softmax classifier are known to produce highly overconfident posterior distributions even for such abnormal samples. In this paper, we propose a simple yet effective method for detecting any abnormal samples, which is applicable to any pre-trained softmax neural classifier. We obtain the class conditional Gaussian distributions with respect to (low- and upper-level) features of the deep models under Gaussian discriminant analysis, which result in a confidence score based on the Mahalanobis distance. While most prior methods have been evaluated for detecting either out-of-distribution or adversarial samples, but not both, the proposed method achieves the state-of-art performances for both cases in our experiments. Moreover, we found that our proposed method is more robust in extreme cases, e.g., when the training dataset has noisy labels or small number of samples. Finally, we show that the proposed method enjoys broader usage by applying it to class incremental learning: whenever out-of-distribution samples are detected, our classification rule can incorporate new classes well without further training deep models.", "", "", "Deep Neural Networks are powerful models that attained remarkable results on a variety of tasks. These models are shown to be extremely efficient when training and test data are drawn from the same distribution. However, it is not clear how a network will act when it is fed with an out-of-distribution example. In this work, we consider the problem of out-of-distribution detection in neural networks. We propose to use multiple semantic dense representations instead of sparse representation as the target label. Specifically, we propose to use several word representations obtained from different corpora or architectures as target labels. We evaluated the proposed model on computer vision, and speech commands detection tasks and compared it to previous methods. Results suggest that our method compares favorably with previous work. Besides, we present the efficiency of our approach for detecting wrongly classified and adversarial examples.", "The problem of detecting whether a test sample is from in-distribution (i.e., training distribution by a classifier) or out-of-distribution sufficiently different from it arises in many real-world machine learning applications. However, the state-of-art deep neural networks are known to be highly overconfident in their predictions, i.e., do not distinguish in- and out-of-distributions. Recently, to handle this issue, several threshold-based detectors have been proposed given pre-trained neural classifiers. However, the performance of prior works highly depends on how to train the classifiers since they only focus on improving inference procedures. In this paper, we develop a novel training method for classifiers so that such inference algorithms can work better. In particular, we suggest two additional terms added to the original loss (e.g., cross entropy). The first one forces samples from out-of-distribution less confident by the classifier and the second one is for (implicitly) generating most effective training samples for the first one. In essence, our method jointly trains both classification and generative neural networks for out-of-distribution. We demonstrate its effectiveness using deep convolutional neural networks on various popular image datasets.", "Estimating how uncertain an AI system is in its predictions is important to improve the safety of such systems. Uncertainty in predictive can result from uncertainty in model parameters, irreducible data uncertainty and uncertainty due to distributional mismatch between the test and training data distributions. Different actions might be taken depending on the source of the uncertainty so it is important to be able to distinguish between them. Recently, baseline tasks and metrics have been defined and several practical methods to estimate uncertainty developed. These methods, however, attempt to model uncertainty due to distributional mismatch either implicitly through model uncertainty or as data uncertainty. This work proposes a new framework for modeling predictive uncertainty called Prior Networks (PNs) which explicitly models distributional uncertainty. PNs do this by parameterizing a prior distribution over predictive distributions. This work focuses on uncertainty for classification and evaluates PNs on the tasks of identifying out-of-distribution (OOD) samples and detecting misclassification on the MNIST dataset, where they are found to outperform previous methods. Experiments on synthetic and MNIST data show that unlike previous non-Bayesian methods PNs are able to distinguish between data and distributional uncertainty." ] }
1905.10628	2945456671	The ability of detecting out-of-distribution (OOD) samples is important to secure reliability of deep neural networks in real-world applications. Considering the nature of OOD samples, detection methods should not have hyperparameters whose optimal values vary sensitively depending on incoming OOD samples. This requirement is not met by many previous methods. In this paper, we propose a simple, hyperparameter-free method that is based on softmax of scaled cosine similarity. It resembles the approach employed by recent metric learning methods, but it differs in details; the differences are essential to achieve high detection performance. As compared with the current state-of-the-art methods, which needs hyperparameter tuning that could compromise real-world performance, the proposed method attains at least competitive detection accuracy even without (tuning of) a hyperparameter; furthermore, it is computationally more efficient, since it needs only a single forward pass unlike previous methods that need backpropagation for each input.	In @cite_37 , a method that employs an ensemble of networks and similar input perturbation is proposed, achieving the state-of-the-art performance. In the training step of this method, ID classes are split into two sets, one of which is virtually treated as ID classes and the other as OOD classes. A network is then trained so that the entropy for the former samples is minimized while that for the latter samples is maximized. Repeating this for different @math splits of classes yields @math leave-out classifiers (i.e., networks). At test time, an input @math is given to these @math networks, whose outputs are summed to calculate ID class scores and an OOD score, where @math is perturbed with magnitude @math in the direction of minimizing the entropy. In the experiments, @math and the temperature are determined by selecting a particular dataset (i.e., iSUN @cite_25 ) as the OOD dataset, and OOD detection performance on different OOD datasets is evaluated.	{ "cite_N": [ "@cite_37", "@cite_25" ], "mid": [ "2889625178", "1831674524" ], "abstract": [ "As deep learning methods form a critical part in commercially important applications such as autonomous driving and medical diagnostics, it is important to reliably detect out-of-distribution (OOD) inputs while employing these algorithms. In this work, we propose an OOD detection algorithm which comprises of an ensemble of classifiers. We train each classifier in a self-supervised manner by leaving out a random subset of training data as OOD data and the rest as in-distribution (ID) data. We propose a novel margin-based loss over the softmax output which seeks to maintain at least a margin m between the average entropy of the OOD and in-distribution samples. In conjunction with the standard cross-entropy loss, we minimize the novel loss to train an ensemble of classifiers. We also propose a novel method to combine the outputs of the ensemble of classifiers to obtain OOD detection score and class prediction. Overall, our method convincingly outperforms [7] and the current state-of-the-art ODIN [13] on several OOD detection benchmarks.", "Traditional eye tracking requires specialized hardware, which means collecting gaze data from many observers is expensive, tedious and slow. Therefore, existing saliency prediction datasets are order-of-magnitudes smaller than typical datasets for other vision recognition tasks. The small size of these datasets limits the potential for training data intensive algorithms, and causes overfitting in benchmark evaluation. To address this deficiency, this paper introduces a webcam-based gaze tracking system that supports large-scale, crowdsourced eye tracking deployed on Amazon Mechanical Turk (AMTurk). By a combination of careful algorithm and gaming protocol design, our system obtains eye tracking data for saliency prediction comparable to data gathered in a traditional lab setting, with relatively lower cost and less effort on the part of the researchers. Using this tool, we build a saliency dataset for a large number of natural images. We will open-source our tool and provide a web server where researchers can upload their images to get eye tracking results from AMTurk." ] }
1905.10628	2945456671	The ability of detecting out-of-distribution (OOD) samples is important to secure reliability of deep neural networks in real-world applications. Considering the nature of OOD samples, detection methods should not have hyperparameters whose optimal values vary sensitively depending on incoming OOD samples. This requirement is not met by many previous methods. In this paper, we propose a simple, hyperparameter-free method that is based on softmax of scaled cosine similarity. It resembles the approach employed by recent metric learning methods, but it differs in details; the differences are essential to achieve high detection performance. As compared with the current state-of-the-art methods, which needs hyperparameter tuning that could compromise real-world performance, the proposed method attains at least competitive detection accuracy even without (tuning of) a hyperparameter; furthermore, it is computationally more efficient, since it needs only a single forward pass unlike previous methods that need backpropagation for each input.	In @cite_53 , another method is proposed, which models layer activation over ID samples with class-wise Gaussian distributions. It uses the induced Mahalanobis distances to class centroids for conducting the classification as well as OOD detection. It employs logistic regression integrating information from multiple layers and input perturbation similar to ODIN, which possesses several hyperparameters. For their determination, in addition to the method using explicit OOD samples as in ODIN @cite_5 , another method is suggested to avoid this potentially unrealistic assumption, which is to create adversarial examples for ID samples and use them as OOD samples, determining the hyperparameters. However, even this method is not free of hyperparameters; the creation of adversarial examples needs one (i.e., perturbation magnitude). It is not discussed how to choose it in their paper.	{ "cite_N": [ "@cite_5", "@cite_53" ], "mid": [ "2963693742", "2867167548" ], "abstract": [ "We consider the problem of detecting out-of-distribution images in neural networks. We propose ODIN, a simple and effective method that does not require any change to a pre-trained neural network. Our method is based on the observation that using temperature scaling and adding small perturbations to the input can separate the softmax score distributions of in- and out-of-distribution images, allowing for more effective detection. We show in a series of experiments that ODIN is compatible with diverse network architectures and datasets. It consistently outperforms the baseline approach by a large margin, establishing a new state-of-the-art performance on this task. For example, ODIN reduces the false positive rate from the baseline 34.7 to 4.3 on the DenseNet (applied to CIFAR-10 and Tiny-ImageNet) when the true positive rate is 95 .", "Detecting test samples drawn sufficiently far away from the training distribution statistically or adversarially is the fundamental requirement to deploy a good classifier in many real-world machine learning applications. However, deep neural networks with the softmax classifier are known to produce highly overconfident posterior distributions even for such abnormal samples. In this paper, we propose a simple yet effective method for detecting any abnormal samples, which is applicable to any pre-trained softmax neural classifier. We obtain the class conditional Gaussian distributions with respect to (low- and upper-level) features of the deep models under Gaussian discriminant analysis, which result in a confidence score based on the Mahalanobis distance. While most prior methods have been evaluated for detecting either out-of-distribution or adversarial samples, but not both, the proposed method achieves the state-of-art performances for both cases in our experiments. Moreover, we found that our proposed method is more robust in extreme cases, e.g., when the training dataset has noisy labels or small number of samples. Finally, we show that the proposed method enjoys broader usage by applying it to class incremental learning: whenever out-of-distribution samples are detected, our classification rule can incorporate new classes well without further training deep models." ] }
1905.10628	2945456671	The ability of detecting out-of-distribution (OOD) samples is important to secure reliability of deep neural networks in real-world applications. Considering the nature of OOD samples, detection methods should not have hyperparameters whose optimal values vary sensitively depending on incoming OOD samples. This requirement is not met by many previous methods. In this paper, we propose a simple, hyperparameter-free method that is based on softmax of scaled cosine similarity. It resembles the approach employed by recent metric learning methods, but it differs in details; the differences are essential to achieve high detection performance. As compared with the current state-of-the-art methods, which needs hyperparameter tuning that could compromise real-world performance, the proposed method attains at least competitive detection accuracy even without (tuning of) a hyperparameter; furthermore, it is computationally more efficient, since it needs only a single forward pass unlike previous methods that need backpropagation for each input.	Unlike these methods, our method does not have any hyperparameter but those for the target classification task. It uses softmax of scaled cosine similarities instead of ordinary softmax of logits. A similar approach has already been employed in recent studies of metric learning, such as @math -constrained softmax @cite_7 , SphereFace @cite_49 , NormFace @cite_36 , ArcFace @cite_32 , CosFace @cite_29 , etc. Although it may seem rather straightforward to apply these methods to OOD detection, to the authors' knowledge, there is no study that has tried this before.	{ "cite_N": [ "@cite_7", "@cite_36", "@cite_29", "@cite_32", "@cite_49" ], "mid": [ "2600537992", "2609575245", "2962898354", "2784874046", "2963466847" ], "abstract": [ "In recent years, the performance of face verification systems has significantly improved using deep convolutional neural networks (DCNNs). A typical pipeline for face verification includes training a deep network for subject classification with softmax loss, using the penultimate layer output as the feature descriptor, and generating a cosine similarity score given a pair of face images. The softmax loss function does not optimize the features to have higher similarity score for positive pairs and lower similarity score for negative pairs, which leads to a performance gap. In this paper, we add an L2-constraint to the feature descriptors which restricts them to lie on a hypersphere of a fixed radius. This module can be easily implemented using existing deep learning frameworks. We show that integrating this simple step in the training pipeline significantly boosts the performance of face verification. Specifically, we achieve state-of-the-art results on the challenging IJB-A dataset, achieving True Accept Rate of 0.909 at False Accept Rate 0.0001 on the face verification protocol. Additionally, we achieve state-of-the-art performance on LFW dataset with an accuracy of 99.78 , and competing performance on YTF dataset with accuracy of 96.08 .", "Thanks to the recent developments of Convolutional Neural Networks, the performance of face verification methods has increased rapidly. In a typical face verification method, feature normalization is a critical step for boosting performance. This motivates us to introduce and study the effect of normalization during training. But we find this is non-trivial, despite normalization being differentiable. We identify and study four issues related to normalization through mathematical analysis, which yields understanding and helps with parameter settings. Based on this analysis we propose two strategies for training using normalized features. The first is a modification of softmax loss, which optimizes cosine similarity instead of inner-product. The second is a reformulation of metric learning by introducing an agent vector for each class. We show that both strategies, and small variants, consistently improve performance by between 0.2 to 0.4 on the LFW dataset based on two models. This is significant because the performance of the two models on LFW dataset is close to saturation at over 98 .", "Face recognition has made extraordinary progress owing to the advancement of deep convolutional neural networks (CNNs). The central task of face recognition, including face verification and identification, involves face feature discrimination. However, the traditional softmax loss of deep CNNs usually lacks the power of discrimination. To address this problem, recently several loss functions such as center loss, large margin softmax loss, and angular softmax loss have been proposed. All these improved losses share the same idea: maximizing inter-class variance and minimizing intra-class variance. In this paper, we propose a novel loss function, namely large margin cosine loss (LMCL), to realize this idea from a different perspective. More specifically, we reformulate the softmax loss as a cosine loss by L2 normalizing both features and weight vectors to remove radial variations, based on which a cosine margin term is introduced to further maximize the decision margin in the angular space. As a result, minimum intra-class variance and maximum inter-class variance are achieved by virtue of normalization and cosine decision margin maximization. We refer to our model trained with LMCL as CosFace. Extensive experimental evaluations are conducted on the most popular public-domain face recognition datasets such as MegaFace Challenge, Youtube Faces (YTF) and Labeled Face in the Wild (LFW). We achieve the state-of-the-art performance on these benchmarks, which confirms the effectiveness of our proposed approach.", "One of the main challenges in feature learning using Deep Convolutional Neural Networks (DCNNs) for large-scale face recognition is the design of appropriate loss functions that enhance discriminative power. Centre loss penalises the distance between the deep features and their corresponding class centres in the Euclidean space to achieve intra-class compactness. SphereFace assumes that the linear transformation matrix in the last fully connected layer can be used as a representation of the class centres in an angular space and penalises the angles between the deep features and their corresponding weights in a multiplicative way. Recently, a popular line of research is to incorporate margins in well-established loss functions in order to maximise face class separability. In this paper, we propose an Additive Angular Margin Loss (ArcFace) to obtain highly discriminative features for face recognition. The proposed ArcFace has a clear geometric interpretation due to the exact correspondence to the geodesic distance on the hypersphere. We present arguably the most extensive experimental evaluation of all the recent state-of-the-art face recognition methods on over 10 face recognition benchmarks including a new large-scale image database with trillion level of pairs and a large-scale video dataset. We show that ArcFace consistently outperforms the state-of-the-art and can be easily implemented with negligible computational overhead. We release all refined training data, training codes, pre-trained models and training logs, which will help reproduce the results in this paper.", "This paper addresses deep face recognition (FR) problem under open-set protocol, where ideal face features are expected to have smaller maximal intra-class distance than minimal inter-class distance under a suitably chosen metric space. However, few existing algorithms can effectively achieve this criterion. To this end, we propose the angular softmax (A-Softmax) loss that enables convolutional neural networks (CNNs) to learn angularly discriminative features. Geometrically, A-Softmax loss can be viewed as imposing discriminative constraints on a hypersphere manifold, which intrinsically matches the prior that faces also lie on a manifold. Moreover, the size of angular margin can be quantitatively adjusted by a parameter m. We further derive specific m to approximate the ideal feature criterion. Extensive analysis and experiments on Labeled Face in the Wild (LFW), Youtube Faces (YTF) and MegaFace Challenge 1 show the superiority of A-Softmax loss in FR tasks." ] }
1905.10628	2945456671	The ability of detecting out-of-distribution (OOD) samples is important to secure reliability of deep neural networks in real-world applications. Considering the nature of OOD samples, detection methods should not have hyperparameters whose optimal values vary sensitively depending on incoming OOD samples. This requirement is not met by many previous methods. In this paper, we propose a simple, hyperparameter-free method that is based on softmax of scaled cosine similarity. It resembles the approach employed by recent metric learning methods, but it differs in details; the differences are essential to achieve high detection performance. As compared with the current state-of-the-art methods, which needs hyperparameter tuning that could compromise real-world performance, the proposed method attains at least competitive detection accuracy even without (tuning of) a hyperparameter; furthermore, it is computationally more efficient, since it needs only a single forward pass unlike previous methods that need backpropagation for each input.	These metric learning methods are identical in that they use cosine similarity to improve a few issues with the ordinary softmax. They differ in i) if and how the weight @math or the feature @math of the last layer are normalized and ii) if and how margins are used with the cosine similarity that (further) encourage maximization of inter-class variance and minimization of intra-class variance. According to this categorization, our method is the most similar to NormFace @cite_36 , in which both @math and @math are normalized and no margin is utilized. However, our method still differs from NormFace that it uses only a single fully-connected layer to compute the cosine similarity and it predicts scale @math multiplied with the inputs to softmax, as will be described below.	{ "cite_N": [ "@cite_36" ], "mid": [ "2609575245" ], "abstract": [ "Thanks to the recent developments of Convolutional Neural Networks, the performance of face verification methods has increased rapidly. In a typical face verification method, feature normalization is a critical step for boosting performance. This motivates us to introduce and study the effect of normalization during training. But we find this is non-trivial, despite normalization being differentiable. We identify and study four issues related to normalization through mathematical analysis, which yields understanding and helps with parameter settings. Based on this analysis we propose two strategies for training using normalized features. The first is a modification of softmax loss, which optimizes cosine similarity instead of inner-product. The second is a reformulation of metric learning by introducing an agent vector for each class. We show that both strategies, and small variants, consistently improve performance by between 0.2 to 0.4 on the LFW dataset based on two models. This is significant because the performance of the two models on LFW dataset is close to saturation at over 98 ." ] }
1905.10604	2945162702	Voice profiling aims at inferring various human parameters from their speech, e.g. gender, age, etc. In this paper, we address the challenge posed by a subtask of voice profiling - reconstructing someone's face from their voice. The task is designed to answer the question: given an audio clip spoken by an unseen person, can we picture a face that has as many common elements, or associations as possible with the speaker, in terms of identity? To address this problem, we propose a simple but effective computational framework based on generative adversarial networks (GANs). The network learns to generate faces from voices by matching the identities of generated faces to those of the speakers, on a training set. We evaluate the performance of the network by leveraging a closely related task - cross-modal matching. The results show that our model is able to generate faces that match several biometric characteristics of the speaker, and results in matching accuracies that are much better than chance.	There currently exists a significant body of research on deriving personal profile parameters from a person's voice @cite_20 . Many useful characteristics about the speaker can be inferred from their speech, e.g. age @cite_11 @cite_30 , emotion @cite_37 , identity @cite_18 , anthropometric measurement @cite_32 , health status @cite_36 etc. These profile parameters may be viewed as providing linkages between a person's voice and their face.	{ "cite_N": [ "@cite_30", "@cite_37", "@cite_18", "@cite_36", "@cite_32", "@cite_20", "@cite_11" ], "mid": [ "2139749422", "1575712594", "2150769028", "2119728341", "", "", "2085999891" ], "abstract": [ "Motivated by the success of i-vectors in the field of speaker recognition, this paper proposes a new approach for age estimation from telephone speech patterns based on i-vectors. In this method, each utterance is modeled by its corresponding ivector. Then, Support Vector Regression (SVR) is applied to estimate the age of speakers. The proposed method is trained and tested on telephone conversations of the National Institute for Standard in Technology (NIST) 2010 and 2008 Speaker Recognition Evaluations databases. Evaluation results show that the proposed method outperforms different conventional methods in speaker age estimation.", "Automatic recognition of a speaker’s emotions is a natural objective for research, but is difficult to gauge the level of performance that is currently attainable. We describe a study that offers a rough benchmark. Speech data came from five passages of about 100 syllables each. They had been selected following pilot studies because they were effective at evoking specific emotion fear, anger, happiness, sadness, and neutrality. 40 subjects were recorded reading them. A battery of 32 potentially relevant features was extracted using our ASSESS system. They were broadly speaking prosodic, derived from contours tracing the movement of intensity and pitch. They were input to statistical decision mechanisms, of two types. Discriminant analysis uses linear combinations of variables to separate samples that belong to different categories. There are reasons to suspect that linear combination will not be appropriate, so neural net classifiers were also considered. An automatic relevance determination procedure was used to identify the most relevant parameters. Discriminant analysis outperformed the neural networks. Using 90 of the data for training, and testing on the remaining 10 , a classification rate of 55 (+ 0.08 ) was achieved. The most useful predictors covered a variety of properties – intensity (relative to the start of the passage) and its spread; pitch spread; durations of silences, rises in intensity, and syllables; and a property related to the shape of ‘tunes’, the number of inflections in the F0 contour per tune. Many more variables were less important, but nevertheless contributed.", "This paper presents an extension of our previous work which proposes a new speaker representation for speaker verification. In this modeling, a new low-dimensional speaker- and channel-dependent space is defined using a simple factor analysis. This space is named the total variability space because it models both speaker and channel variabilities. Two speaker verification systems are proposed which use this new representation. The first system is a support vector machine-based system that uses the cosine kernel to estimate the similarity between the input data. The second system directly uses the cosine similarity as the final decision score. We tested three channel compensation techniques in the total variability space, which are within-class covariance normalization (WCCN), linear discriminate analysis (LDA), and nuisance attribute projection (NAP). We found that the best results are obtained when LDA is followed by WCCN. We achieved an equal error rate (EER) of 1.12 and MinDCF of 0.0094 using the cosine distance scoring on the male English trials of the core condition of the NIST 2008 Speaker Recognition Evaluation dataset. We also obtained 4 absolute EER improvement for both-gender trials on the 10 s-10 s condition compared to the classical joint factor analysis scoring.", "Health information access by low-literate community health workers is a pressing need of community health programs across the developing world. We present results from a needs assessment we conducted to understand the health information access practices and needs of various types of health workers in Pakistan. We also present a prototype for speech-based health information access, as well as discuss our experiences from a pilot study involving its use by community health workers in a rural health center.", "", "", "The ability of 10 listeners to differentiate the voices of younger adults (under age 35) from older (over age 65) subjects on the basis of a prolonged vowel, a reading sample played backward, and a..." ] }
1905.10604	2945162702	Voice profiling aims at inferring various human parameters from their speech, e.g. gender, age, etc. In this paper, we address the challenge posed by a subtask of voice profiling - reconstructing someone's face from their voice. The task is designed to answer the question: given an audio clip spoken by an unseen person, can we picture a face that has as many common elements, or associations as possible with the speaker, in terms of identity? To address this problem, we propose a simple but effective computational framework based on generative adversarial networks (GANs). The network learns to generate faces from voices by matching the identities of generated faces to those of the speakers, on a training set. We evaluate the performance of the network by leveraging a closely related task - cross-modal matching. The results show that our model is able to generate faces that match several biometric characteristics of the speaker, and results in matching accuracies that are much better than chance.	There is a long line of research on generating faces using GANs. @cite_34 @cite_27 can be used to generate face images from noise, but it does not consider any conditioning information (like identity) as input. @cite_3 @cite_2 use a discrete label as the condition, but it only works on a closed-set scenario. @cite_26 @cite_35 achieve identity-preserving face generation in an open-set scenario. However, the problem they focus on is within-modal generation problem, where a reference face image of the target identity to be generated is shown to the model. Our task is more challenging, since the conditioning information provided is actually from a different modality, namely voice.	{ "cite_N": [ "@cite_35", "@cite_26", "@cite_3", "@cite_27", "@cite_2", "@cite_34" ], "mid": [ "2964339532", "2964337551", "2552611751", "2766527293", "2618574778", "2173520492" ], "abstract": [ "We propose a framework based on Generative Adversarial Networks to disentangle the identity and attributes of faces, such that we can conveniently recombine different identities and attributes for identity preserving face synthesis in open domains. Previous identity preserving face synthesis processes are largely confined to synthesizing faces with known identities that are already in the training dataset. To synthesize a face with identity outside the training dataset, our framework requires one input image of that subject to produce an identity vector, and any other input face image to extract an attribute vector capturing, e.g., pose, emotion, illumination, and even the background. We then recombine the identity vector and the attribute vector to synthesize a new face of the subject with the extracted attribute. Our proposed framework does not need to annotate the attributes of faces in any way. It is trained with an asymmetric loss function to better preserve the identity and stabilize the training process. It can also effectively leverage large amounts of unlabeled training face images to further improve the fidelity of the synthesized faces for subjects that are not presented in the labeled training face dataset. Our experiments demonstrate the efficacy of the proposed framework. We also present its usage in a much broader set of applications including face frontalization, face attribute morphing, and face adversarial example detection.", "Photorealistic frontal view synthesis from a single face image has a wide range of applications in the field of face recognition. Although data-driven deep learning methods have been proposed to address this problem by seeking solutions from ample face data, this problem is still challenging because it is intrinsically ill-posed. This paper proposes a Two-Pathway Generative Adversarial Network (TP-GAN) for photorealistic frontal view synthesis by simultaneously perceiving global structures and local details. Four landmark located patch networks are proposed to attend to local textures in addition to the commonly used global encoderdecoder network. Except for the novel architecture, we make this ill-posed problem well constrained by introducing a combination of adversarial loss, symmetry loss and identity preserving loss. The combined loss function leverages both frontal face distribution and pre-trained discriminative deep face models to guide an identity preserving inference of frontal views from profiles. Different from previous deep learning methods that mainly rely on intermediate features for recognition, our method directly leverages the synthesized identity preserving image for downstream tasks like face recognition and attribution estimation. Experimental results demonstrate that our method not only presents compelling perceptual results but also outperforms state-of-theart results on large pose face recognition.", "Generative Adversarial Networks (GANs) have recently demonstrated to successfully approximate complex data distributions. A relevant extension of this model is conditional GANs (cGANs), where the introduction of external information allows to determine specific representations of the generated images. In this work, we evaluate encoders to inverse the mapping of a cGAN, i.e., mapping a real image into a latent space and a conditional representation. This allows, for example, to reconstruct and modify real images of faces conditioning on arbitrary attributes. Additionally, we evaluate the design of cGANs. The combination of an encoder with a cGAN, which we call Invertible cGAN (IcGAN), enables to re-generate real images with deterministic complex modifications.", "We describe a new training methodology for generative adversarial networks. The key idea is to grow both the generator and discriminator progressively: starting from a low resolution, we add new layers that model increasingly fine details as training progresses. This both speeds the training up and greatly stabilizes it, allowing us to produce images of unprecedented quality, e.g., CelebA images at 1024^2. We also propose a simple way to increase the variation in generated images, and achieve a record inception score of 8.80 in unsupervised CIFAR10. Additionally, we describe several implementation details that are important for discouraging unhealthy competition between the generator and discriminator. Finally, we suggest a new metric for evaluating GAN results, both in terms of image quality and variation. As an additional contribution, we construct a higher-quality version of the CelebA dataset.", "", "In recent years, supervised learning with convolutional networks (CNNs) has seen huge adoption in computer vision applications. Comparatively, unsupervised learning with CNNs has received less attention. In this work we hope to help bridge the gap between the success of CNNs for supervised learning and unsupervised learning. We introduce a class of CNNs called deep convolutional generative adversarial networks (DCGANs), that have certain architectural constraints, and demonstrate that they are a strong candidate for unsupervised learning. Training on various image datasets, we show convincing evidence that our deep convolutional adversarial pair learns a hierarchy of representations from object parts to scenes in both the generator and discriminator. Additionally, we use the learned features for novel tasks - demonstrating their applicability as general image representations." ] }
1905.10604	2945162702	Voice profiling aims at inferring various human parameters from their speech, e.g. gender, age, etc. In this paper, we address the challenge posed by a subtask of voice profiling - reconstructing someone's face from their voice. The task is designed to answer the question: given an audio clip spoken by an unseen person, can we picture a face that has as many common elements, or associations as possible with the speaker, in terms of identity? To address this problem, we propose a simple but effective computational framework based on generative adversarial networks (GANs). The network learns to generate faces from voices by matching the identities of generated faces to those of the speakers, on a training set. We evaluate the performance of the network by leveraging a closely related task - cross-modal matching. The results show that our model is able to generate faces that match several biometric characteristics of the speaker, and results in matching accuracies that are much better than chance.	Cross modal matching between voice and face has become an increasingly popular problem in recent years @cite_0 . It is often posed as the problem of matching a probe input from one modality to a gallery of multiple inputs from the other modality. @cite_13 formulates this task as a @math -way classification problem. @cite_10 @cite_8 @cite_12 propose to learn common embeddings for the cross-modal inputs, such that the matching can be performed using the learned embeddings. All of these, however, are essentially selection problems. Our task requires actual generation is naturally more challenging than cross-modal matching. In part, this is also because many possible generated images could be the expected output for an audio input. There is no unique target output as a supervision signal to train our model.	{ "cite_N": [ "@cite_13", "@cite_8", "@cite_0", "@cite_10", "@cite_12" ], "mid": [ "2963887950", "2803113447", "2621109248", "2798513620", "2951237705" ], "abstract": [ "We introduce a seemingly impossible task: given only an audio clip of someone speaking, decide which of two face images is the speaker. In this paper we study this, and a number of related cross-modal tasks, aimed at answering the question: how much can we infer from the voice about the face and vice versa? We study this task \"in the wild\", employing the datasets that are now publicly available for face recognition from static images (VGGFace) and speaker identification from audio (VoxCeleb). These provide training and testing scenarios for both static and dynamic testing of cross-modal matching. We make the following contributions: (i) we introduce CNN architectures for both binary and multi-way cross-modal face and audio matching: (ii) we compare dynamic testing (where video information is available, but the audio is not from the same video) with static testing (where only a single still image is available): and (iii) we use human testing as a baseline to calibrate the difficulty of the task. We show that a CNN can indeed be trained to solve this task in both the static and dynamic scenarios, and is even well above chance on 10-way classification of the face given the voice. The CNN matches human performance on easy examples (e.g. different gender across faces) but exceeds human performance on more challenging examples (e.g. faces with the same gender, age and nationality).", "In this paper, we study the associations between human faces and voices. Audiovisual integration, specifically the integration of facial and vocal information is a well-researched area in neuroscience. It is shown that the overlapping information between the two modalities plays a significant role in perceptual tasks such as speaker identification. Through an online study on a new dataset we created, we confirm previous findings that people can associate unseen faces with corresponding voices and vice versa with greater than chance accuracy. We computationally model the overlapping information between faces and voices and show that the learned cross-modal representation contains enough information to identify matching faces and voices with performance similar to that of humans. Our representation exhibits correlations to certain demographic attributes and features obtained from either visual or aural modality alone. We release our dataset of audiovisual recordings and demographic annotations of people reading out short text used in our studies.", "In this paper, we present a system that associates faces with voices in a video by fusing information from the audio and visual signals. The thesis underlying our work is that an extremely simple approach to generating (weak) speech clusters can be combined with visual signals to effectively associate faces and voices by aggregating statistics across a video. This approach does not need any training data specific to this task and leverages the natural coherence of information in the audio and visual streams. It is particularly applicable to tracking speakers in videos on the web where a priori information about the environment (e.g., number of speakers, spatial signals for beamforming) is not available. We performed experiments on a real-world dataset using this analysis framework to determine the speaker in a video. Given a ground truth labeling determined by human rater consensus, our approach had 71 accuracy.", "We propose and investigate an identity sensitive joint embedding of face and voice. Such an embedding enables cross-modal retrieval from voice to face and from face to voice. We make the following four contributions: first, we show that the embedding can be learnt from videos of talking faces, without requiring any identity labels, using a form of cross-modal self-supervision; second, we develop a curriculum learning schedule for hard negative mining targeted to this task, that is essential for learning to proceed successfully; third, we demonstrate and evaluate cross-modal retrieval for identities unseen and unheard during training over a number of scenarios and establish a benchmark for this novel task; finally, we show an application of using the joint embedding for automatically retrieving and labelling characters in TV dramas.", "We propose a novel framework, called Disjoint Mapping Network (DIMNet), for cross-modal biometric matching, in particular of voices and faces. Different from the existing methods, DIMNet does not explicitly learn the joint relationship between the modalities. Instead, DIMNet learns a shared representation for different modalities by mapping them individually to their common covariates. These shared representations can then be used to find the correspondences between the modalities. We show empirically that DIMNet is able to achieve better performance than other current methods, with the additional benefits of being conceptually simpler and less data-intensive." ] }
1905.10604	2945162702	Voice profiling aims at inferring various human parameters from their speech, e.g. gender, age, etc. In this paper, we address the challenge posed by a subtask of voice profiling - reconstructing someone's face from their voice. The task is designed to answer the question: given an audio clip spoken by an unseen person, can we picture a face that has as many common elements, or associations as possible with the speaker, in terms of identity? To address this problem, we propose a simple but effective computational framework based on generative adversarial networks (GANs). The network learns to generate faces from voices by matching the identities of generated faces to those of the speakers, on a training set. We evaluate the performance of the network by leveraging a closely related task - cross-modal matching. The results show that our model is able to generate faces that match several biometric characteristics of the speaker, and results in matching accuracies that are much better than chance.	Talking face @cite_14 @cite_28 is a recently proposed task. Given a static face image and a speech clip, the goal is to generate a sequence of target face images where the lips are synchronized with the audio. The talking face task is significantly different from our task; in fact the two are actually somewhat orthogonal to each other because talking face focuses on the content of the speech and ignores the identity of the speaker, while Our task does the opposite, extracting the identity of speaker and discarding the content of speech.	{ "cite_N": [ "@cite_28", "@cite_14" ], "mid": [ "2914217321", "2883082281" ], "abstract": [ "We describe a method for generating a video of a talking face. The method takes still images of the target face and an audio speech segment as inputs, and generates a video of the target face lip synched with the audio. The method runs in real time and is applicable to faces and audio not seen at training time. To achieve this we develop an encoder–decoder convolutional neural network (CNN) model that uses a joint embedding of the face and audio to generate synthesised talking face video frames. The model is trained on unlabelled videos using cross-modal self-supervision. We also propose methods to re-dub videos by visually blending the generated face into the source video frame using a multi-stream CNN model.", "Talking face generation aims to synthesize a sequence of face images that correspond to a clip of speech. This is a challenging task because face appearance variation and semantics of speech are coupled together in the subtle movements of the talking face regions. Existing works either construct specific face appearance model on specific subjects or model the transformation between lip motion and speech. In this work, we integrate both aspects and enable arbitrary-subject talking face generation by learning disentangled audio-visual representation. We find that the talking face sequence is actually a composition of both subject-related information and speech-related information. These two spaces are then explicitly disentangled through a novel associative-and-adversarial training process. This disentangled representation has an advantage where both audio and video can serve as inputs for generation. Extensive experiments show that the proposed approach generates realistic talking face sequences on arbitrary subjects with much clearer lip motion patterns than previous work. We also demonstrate the learned audio-visual representation is extremely useful for the tasks of automatic lip reading and audio-video retrieval." ] }
1905.10346	2946299443	Portrait editing is a popular subject in photo manipulation. The Generative Adversarial Network (GAN) advances the generating of realistic faces and allows more face editing. In this paper, we argue about three issues in existing techniques: diversity, quality, and controllability for portrait synthesis and editing. To address these issues, we propose a novel end-to-end learning framework that leverages conditional GANs guided by provided face masks for generating faces. The framework learns feature embeddings for every face component (e.g., mouth, hair, eye), separately, contributing to better correspondences for image translation, and local face editing. With the mask, our network is available to many applications, like face synthesis driven by mask, face Swap+ (including hair in swapping), and local manipulation. It can also boost the performance of face parsing a bit as an option of data augmentation.	Generative adversarial networks (GANs) @cite_10 have achieved impressive results in many directions. It forces the generated samples to be indistinguishable from the target distribution by introducing an adversarial discriminator. The GAN family enables a wide variety of computer vision applications such as image synthesis @cite_32 @cite_15 , image translation @cite_20 @cite_33 @cite_21 @cite_3 , and representation disentangling @cite_5 @cite_31 @cite_22 , among others.	{ "cite_N": [ "@cite_31", "@cite_33", "@cite_22", "@cite_21", "@cite_32", "@cite_3", "@cite_5", "@cite_15", "@cite_10", "@cite_20" ], "mid": [ "", "2962793481", "2737047298", "2774625825", "", "", "2797650215", "2580360036", "", "" ], "abstract": [ "", "Image-to-image translation is a class of vision and graphics problems where the goal is to learn the mapping between an input image and an output image using a training set of aligned image pairs. However, for many tasks, paired training data will not be available. We present an approach for learning to translate an image from a source domain X to a target domain Y in the absence of paired examples. Our goal is to learn a mapping G : X → Y such that the distribution of images from G(X) is indistinguishable from the distribution Y using an adversarial loss. Because this mapping is highly under-constrained, we couple it with an inverse mapping F : Y → X and introduce a cycle consistency loss to push F(G(X)) ≈ X (and vice versa). Qualitative results are presented on several tasks where paired training data does not exist, including collection style transfer, object transfiguration, season transfer, photo enhancement, etc. Quantitative comparisons against several prior methods demonstrate the superiority of our approach.", "The large pose discrepancy between two face images is one of the key challenges in face recognition. Conventional approaches for pose-invariant face recognition either perform face frontalization on, or learn a pose-invariant representation from, a non-frontal face image. We argue that it is more desirable to perform both tasks jointly to allow them to leverage each other. To this end, this paper proposes Disentangled Representation learning-Generative Adversarial Network (DR-GAN) with three distinct novelties. First, the encoder-decoder structure of the generator allows DR-GAN to learn a generative and discriminative representation, in addition to image synthesis. Second, this representation is explicitly disentangled from other face variations such as pose, through the pose code provided to the decoder and pose estimation in the discriminator. Third, DR-GAN can take one or multiple images as the input, and generate one unified representation along with an arbitrary number of synthetic images. Quantitative and qualitative evaluation on both controlled and in-the-wild databases demonstrate the superiority of DR-GAN over the state of the art.", "We present a new method for synthesizing high-resolution photo-realistic images from semantic label maps using conditional generative adversarial networks (conditional GANs). Conditional GANs have enabled a variety of applications, but the results are often limited to low-resolution and still far from realistic. In this work, we generate 2048x1024 visually appealing results with a novel adversarial loss, as well as new multi-scale generator and discriminator architectures. Furthermore, we extend our framework to interactive visual manipulation with two additional features. First, we incorporate object instance segmentation information, which enables object manipulations such as removing adding objects and changing the object category. Second, we propose a method to generate diverse results given the same input, allowing users to edit the object appearance interactively. Human opinion studies demonstrate that our method significantly outperforms existing methods, advancing both the quality and the resolution of deep image synthesis and editing.", "", "", "Unsupervised image-to-image translation is an important and challenging problem in computer vision. Given an image in the source domain, the goal is to learn the conditional distribution of corresponding images in the target domain, without seeing any pairs of corresponding images. While this conditional distribution is inherently multimodal, existing approaches make an overly simplified assumption, modeling it as a deterministic one-to-one mapping. As a result, they fail to generate diverse outputs from a given source domain image. To address this limitation, we propose a Multimodal Unsupervised Image-to-image Translation (MUNIT) framework. We assume that the image representation can be decomposed into a content code that is domain-invariant, and a style code that captures domain-specific properties. To translate an image to another domain, we recombine its content code with a random style code sampled from the style space of the target domain. We analyze the proposed framework and establish several theoretical results. Extensive experiments with comparisons to the state-of-the-art approaches further demonstrates the advantage of the proposed framework. Moreover, our framework allows users to control the style of translation outputs by providing an example style image. Code and pretrained models are available at this https URL", "In this paper, we present the Lipschitz regularization theory and algorithms for a novel Loss-Sensitive Generative Adversarial Network (LS-GAN). Specifically, it trains a loss function to distinguish between real and fake samples by designated margins, while learning a generator alternately to produce realistic samples by minimizing their losses. The LS-GAN further regularizes its loss function with a Lipschitz regularity condition on the density of real data, yielding a regularized model that can better generalize to produce new data from a reasonable number of training examples than the classic GAN. We will further present a Generalized LS-GAN (GLS-GAN) and show it contains a large family of regularized GAN models, including both LS-GAN and Wasserstein GAN, as its special cases. Compared with the other GAN models, we will conduct experiments to show both LS-GAN and GLS-GAN exhibit competitive ability in generating new images in terms of the Minimum Reconstruction Error (MRE) assessed on a separate test set. We further extend the LS-GAN to a conditional form for supervised and semi-supervised learning problems, and demonstrate its outstanding performance on image classification tasks.", "", "" ] }
1905.10346	2946299443	Portrait editing is a popular subject in photo manipulation. The Generative Adversarial Network (GAN) advances the generating of realistic faces and allows more face editing. In this paper, we argue about three issues in existing techniques: diversity, quality, and controllability for portrait synthesis and editing. To address these issues, we propose a novel end-to-end learning framework that leverages conditional GANs guided by provided face masks for generating faces. The framework learns feature embeddings for every face component (e.g., mouth, hair, eye), separately, contributing to better correspondences for image translation, and local face editing. With the mask, our network is available to many applications, like face synthesis driven by mask, face Swap+ (including hair in swapping), and local manipulation. It can also boost the performance of face parsing a bit as an option of data augmentation.	Inspired by the conditional GAN models @cite_0 that generate images from masks, we propose a novel framework for mask-guided portrait editing. Our method leverages local embedding information for individual facial components, generating portrait images with higher diversity and controllability, than existing global-based approaches such as pix2pix @cite_20 or pix2pixHD @cite_21 .	{ "cite_N": [ "@cite_0", "@cite_21", "@cite_20" ], "mid": [ "2125389028", "2774625825", "" ], "abstract": [ "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 present a new method for synthesizing high-resolution photo-realistic images from semantic label maps using conditional generative adversarial networks (conditional GANs). Conditional GANs have enabled a variety of applications, but the results are often limited to low-resolution and still far from realistic. In this work, we generate 2048x1024 visually appealing results with a novel adversarial loss, as well as new multi-scale generator and discriminator architectures. Furthermore, we extend our framework to interactive visual manipulation with two additional features. First, we incorporate object instance segmentation information, which enables object manipulations such as removing adding objects and changing the object category. Second, we propose a method to generate diverse results given the same input, allowing users to edit the object appearance interactively. Human opinion studies demonstrate that our method significantly outperforms existing methods, advancing both the quality and the resolution of deep image synthesis and editing.", "" ] }
1905.10346	2946299443	Portrait editing is a popular subject in photo manipulation. The Generative Adversarial Network (GAN) advances the generating of realistic faces and allows more face editing. In this paper, we argue about three issues in existing techniques: diversity, quality, and controllability for portrait synthesis and editing. To address these issues, we propose a novel end-to-end learning framework that leverages conditional GANs guided by provided face masks for generating faces. The framework learns feature embeddings for every face component (e.g., mouth, hair, eye), separately, contributing to better correspondences for image translation, and local face editing. With the mask, our network is available to many applications, like face synthesis driven by mask, face Swap+ (including hair in swapping), and local manipulation. It can also boost the performance of face parsing a bit as an option of data augmentation.	Image editing has benefited a lot from the rapid growth of deep neural networks, including image completion @cite_43 , super-resolution @cite_26 , deep analogy @cite_34 , and sketch-based portrait editing @cite_18 @cite_47 , to name a few. Among them, the most related are mask-guided image editing methods, which train deep neural networks to translate masks into realistic images @cite_24 @cite_44 @cite_20 @cite_21 @cite_33 .	{ "cite_N": [ "@cite_18", "@cite_26", "@cite_33", "@cite_21", "@cite_24", "@cite_43", "@cite_44", "@cite_47", "@cite_34", "@cite_20" ], "mid": [ "2560481159", "2523714292", "2962793481", "2774625825", "2963522749", "2738588019", "2302243225", "2798386241", "2951924128", "" ], "abstract": [ "Several recent works have used deep convolutional networks to generate realistic imagery. These methods sidestep the traditional computer graphics rendering pipeline and instead generate imagery at the pixel level by learning from large collections of photos (e.g. faces or bedrooms). However, these methods are of limited utility because it is difficult for a user to control what the network produces. In this paper, we propose a deep adversarial image synthesis architecture that is conditioned on sketched boundaries and sparse color strokes to generate realistic cars, bedrooms, or faces. We demonstrate a sketch based image synthesis system which allows users to scribble over the sketch to indicate preferred color for objects. Our network can then generate convincing images that satisfy both the color and the sketch constraints of user. The network is feed-forward which allows users to see the effect of their edits in real time. We compare to recent work on sketch to image synthesis and show that our approach generates more realistic, diverse, and controllable outputs. The architecture is also effective at user-guided colorization of grayscale images.", "Despite the breakthroughs in accuracy and speed of single image super-resolution using faster and deeper convolutional neural networks, one central problem remains largely unsolved: how do we recover the finer texture details when we super-resolve at large upscaling factors? The behavior of optimization-based super-resolution methods is principally driven by the choice of the objective function. Recent work has largely focused on minimizing the mean squared reconstruction error. The resulting estimates have high peak signal-to-noise ratios, but they are often lacking high-frequency details and are perceptually unsatisfying in the sense that they fail to match the fidelity expected at the higher resolution. In this paper, we present SRGAN, a generative adversarial network (GAN) for image super-resolution (SR). To our knowledge, it is the first framework capable of inferring photo-realistic natural images for 4x upscaling factors. To achieve this, we propose a perceptual loss function which consists of an adversarial loss and a content loss. The adversarial loss pushes our solution to the natural image manifold using a discriminator network that is trained to differentiate between the super-resolved images and original photo-realistic images. In addition, we use a content loss motivated by perceptual similarity instead of similarity in pixel space. Our deep residual network is able to recover photo-realistic textures from heavily downsampled images on public benchmarks. An extensive mean-opinion-score (MOS) test shows hugely significant gains in perceptual quality using SRGAN. The MOS scores obtained with SRGAN are closer to those of the original high-resolution images than to those obtained with any state-of-the-art method.", "Image-to-image translation is a class of vision and graphics problems where the goal is to learn the mapping between an input image and an output image using a training set of aligned image pairs. However, for many tasks, paired training data will not be available. We present an approach for learning to translate an image from a source domain X to a target domain Y in the absence of paired examples. Our goal is to learn a mapping G : X → Y such that the distribution of images from G(X) is indistinguishable from the distribution Y using an adversarial loss. Because this mapping is highly under-constrained, we couple it with an inverse mapping F : Y → X and introduce a cycle consistency loss to push F(G(X)) ≈ X (and vice versa). Qualitative results are presented on several tasks where paired training data does not exist, including collection style transfer, object transfiguration, season transfer, photo enhancement, etc. Quantitative comparisons against several prior methods demonstrate the superiority of our approach.", "We present a new method for synthesizing high-resolution photo-realistic images from semantic label maps using conditional generative adversarial networks (conditional GANs). Conditional GANs have enabled a variety of applications, but the results are often limited to low-resolution and still far from realistic. In this work, we generate 2048x1024 visually appealing results with a novel adversarial loss, as well as new multi-scale generator and discriminator architectures. Furthermore, we extend our framework to interactive visual manipulation with two additional features. First, we incorporate object instance segmentation information, which enables object manipulations such as removing adding objects and changing the object category. Second, we propose a method to generate diverse results given the same input, allowing users to edit the object appearance interactively. Human opinion studies demonstrate that our method significantly outperforms existing methods, advancing both the quality and the resolution of deep image synthesis and editing.", "We present an approach to synthesizing photographic images conditioned on semantic layouts. Given a semantic label map, our approach produces an image with photographic appearance that conforms to the input layout. The approach thus functions as a rendering engine that takes a two-dimensional semantic specification of the scene and produces a corresponding photographic image. Unlike recent and contemporaneous work, our approach does not rely on adversarial training. We show that photographic images can be synthesized from semantic layouts by a single feedforward network with appropriate structure, trained end-to-end with a direct regression objective. The presented approach scales seamlessly to high resolutions; we demonstrate this by synthesizing photographic images at 2-megapixel resolution, the full resolution of our training data. Extensive perceptual experiments on datasets of outdoor and indoor scenes demonstrate that images synthesized by the presented approach are considerably more realistic than alternative approaches.", "We present a novel approach for image completion that results in images that are both locally and globally consistent. With a fully-convolutional neural network, we can complete images of arbitrary resolutions by filling-in missing regions of any shape. To train this image completion network to be consistent, we use global and local context discriminators that are trained to distinguish real images from completed ones. The global discriminator looks at the entire image to assess if it is coherent as a whole, while the local discriminator looks only at a small area centered at the completed region to ensure the local consistency of the generated patches. The image completion network is then trained to fool the both context discriminator networks, which requires it to generate images that are indistinguishable from real ones with regard to overall consistency as well as in details. We show that our approach can be used to complete a wide variety of scenes. Furthermore, in contrast with the patch-based approaches such as PatchMatch, our approach can generate fragments that do not appear elsewhere in the image, which allows us to naturally complete the images of objects with familiar and highly specific structures, such as faces.", "Convolutional neural networks (CNNs) have proven highly effective at image synthesis and style transfer. For most users, however, using them as tools can be a challenging task due to their unpredictable behavior that goes against common intuitions. This paper introduces a novel concept to augment such generative architectures with semantic annotations, either by manually authoring pixel labels or using existing solutions for semantic segmentation. The result is a content-aware generative algorithm that offers meaningful control over the outcome. Thus, we increase the quality of images generated by avoiding common glitches, make the results look significantly more plausible, and extend the functional range of these algorithms---whether for portraits or landscapes, etc. Applications include semantic style transfer and turning doodles with few colors into masterful paintings!", "We present a novel system for sketch-based face image editing, enabling users to edit images in an intuitive manner by sketching a few strokes on a region of interest. Our interface features tools to express a desired image manipulation by providing both geometry and color constraints as user-drawn strokes. As an alternative to the direct user input, our proposed system naturally supports a copy-paste mode, which allows users to edit a given image region by using parts of another exemplar image without the need of hand-drawn sketching at all. The proposed interface runs in real-time and facilitates an interactive and iterative workflow to quickly express the intended edits. Our system is based on a novel sketch domain and a convolutional neural network trained end-to-end to automatically learn to render image regions corresponding to the input strokes. To achieve high quality and semantically consistent results we train our neural network on two simultaneous tasks, namely image completion and image translation. To the best of our knowledge, we are the first to combine these two tasks in a unified framework for interactive image editing. Our results show that the proposed sketch domain, network architecture, and training procedure generalize well to real user input and enable high quality synthesis results without additional post-processing. Moreover, the proposed fully-convolutional model does not rely on any multi-scale or cascaded network architecture to synthesize high-resolution images and can be trained end-to-end to produce images of arbitrary size.", "We propose a new technique for visual attribute transfer across images that may have very different appearance but have perceptually similar semantic structure. By visual attribute transfer, we mean transfer of visual information (such as color, tone, texture, and style) from one image to another. For example, one image could be that of a painting or a sketch while the other is a photo of a real scene, and both depict the same type of scene. Our technique finds semantically-meaningful dense correspondences between two input images. To accomplish this, it adapts the notion of \"image analogy\" with features extracted from a Deep Convolutional Neutral Network for matching; we call our technique Deep Image Analogy. A coarse-to-fine strategy is used to compute the nearest-neighbor field for generating the results. We validate the effectiveness of our proposed method in a variety of cases, including style texture transfer, color style swap, sketch painting to photo, and time lapse.", "" ] }
1905.10346	2946299443	Portrait editing is a popular subject in photo manipulation. The Generative Adversarial Network (GAN) advances the generating of realistic faces and allows more face editing. In this paper, we argue about three issues in existing techniques: diversity, quality, and controllability for portrait synthesis and editing. To address these issues, we propose a novel end-to-end learning framework that leverages conditional GANs guided by provided face masks for generating faces. The framework learns feature embeddings for every face component (e.g., mouth, hair, eye), separately, contributing to better correspondences for image translation, and local face editing. With the mask, our network is available to many applications, like face synthesis driven by mask, face Swap+ (including hair in swapping), and local manipulation. It can also boost the performance of face parsing a bit as an option of data augmentation.	Our approach also relates to the visual attribute transfer methods, including style transfer @cite_29 @cite_45 and color transfer @cite_39 . Recently, the Paired-CycleGAN @cite_17 has been proposed for makeup transfer, in which a makeup transfer function and a makeup removal function are trained in pair. Though similar, the appearances of facial instances that our method disentangles differ from makeups. For example, the color and curly types of hairs which we can transfer are definitely not makeups. Furthermore, there are some works focusing on editing a specific component in faces ( , eyes @cite_7 @cite_28 ) or editing attributes of faces @cite_42 @cite_35 .	{ "cite_N": [ "@cite_35", "@cite_7", "@cite_28", "@cite_29", "@cite_42", "@cite_39", "@cite_45", "@cite_17" ], "mid": [ "", "2963521568", "2519156253", "2475287302", "2883861033", "2763915504", "2798520250", "2798600195" ], "abstract": [ "", "This paper introduces a novel approach to in-painting where the identity of the object to remove or change is preserved and accounted for at inference time: Exemplar GANs (ExGANs). ExGANs are a type of conditional GAN that utilize exemplar information to produce high-quality, personalized in-painting results. We propose using exemplar information in the form of a reference image of the region to in-paint, or a perceptual code describing that object. Unlike previous conditional GAN formulations, this extra information can be inserted at multiple points within the adversarial network, thus increasing its descriptive power. We show that ExGANs can produce photo-realistic personalized in-painting results that are both perceptually and semantically plausible by applying them to the task of closed-to-open eye in-painting in natural pictures. A new benchmark dataset is also introduced for the task of eye inpainting for future comparisons.", "Closed eyes and look-aways can ruin precious moments captured in photographs. In this article, we present a new framework for automatically editing eyes in photographs. We leverage a user’s personal photo collection to find a “good” set of reference eyes and transfer them onto a target image. Our example-based editing approach is robust and effective for realistic image editing. A fully automatic pipeline for realistic eye editing is challenging due to the unconstrained conditions under which the face appears in a typical photo collection. We use crowd-sourced human evaluations to understand the aspects of the target-reference image pair that will produce the most realistic results. We subsequently train a model that automatically selects the top-ranked reference candidate(s) by narrowing the gap in terms of pose, local contrast, lighting conditions, and even expressions. Finally, we develop a comprehensive pipeline of three-dimensional face estimation, image warping, relighting, image harmonization, automatic segmentation, and image compositing in order to achieve highly believable results. We evaluate the performance of our method via quantitative and crowd-sourced experiments.", "Rendering the semantic content of an image in different styles is a difficult image processing task. Arguably, a major limiting factor for previous approaches has been the lack of image representations that explicitly represent semantic information and, thus, allow to separate image content from style. Here we use image representations derived from Convolutional Neural Networks optimised for object recognition, which make high level image information explicit. We introduce A Neural Algorithm of Artistic Style that can separate and recombine the image content and style of natural images. The algorithm allows us to produce new images of high perceptual quality that combine the content of an arbitrary photograph with the appearance of numerous wellknown artworks. Our results provide new insights into the deep image representations learned by Convolutional Neural Networks and demonstrate their potential for high level image synthesis and manipulation.", "Recent advances in Generative Adversarial Networks (GANs) have shown impressive results for task of facial expression synthesis. The most successful architecture is StarGAN, that conditions GANs’ generation process with images of a specific domain, namely a set of images of persons sharing the same expression. While effective, this approach can only generate a discrete number of expressions, determined by the content of the dataset. To address this limitation, in this paper, we introduce a novel GAN conditioning scheme based on Action Units (AU) annotations, which describes in a continuous manifold the anatomical facial movements defining a human expression. Our approach allows controlling the magnitude of activation of each AU and combine several of them. Additionally, we propose a fully unsupervised strategy to train the model, that only requires images annotated with their activated AUs, and exploit attention mechanisms that make our network robust to changing backgrounds and lighting conditions. Extensive evaluation show that our approach goes beyond competing conditional generators both in the capability to synthesize a much wider range of expressions ruled by anatomically feasible muscle movements, as in the capacity of dealing with images in the wild.", "We propose a new algorithm for color transfer between images that have perceptually similar semantic structures. We aim to achieve a more accurate color transfer that leverages semantically-meaningful dense correspondence between images. To accomplish this, our algorithm uses neural representations for matching. Additionally, the color transfer should be spatially-variant and globally coherent. Therefore, our algorithm optimizes a local linear model for color transfer satisfying both local and global constraints. Our proposed approach jointly optimize matching and color transfer, adopting a coarse-to-fine strategy. The proposed method can be successfully extended from \"one-to-one\" to \"one-to-many\" color transfers. The latter further addresses the problem of mismatching elements of the input image. We validate our proposed method by testing it on a large variety of image content.", "This paper introduces a novel method by reshuffling deep features (i.e., permuting the spacial locations of a feature map) of the style image for arbitrary style transfer. We theoretically prove that our new style loss based on reshuffle connects both global and local style losses respectively used by most parametric and non-parametric neural style transfer methods. This simple idea can effectively address the challenging issues in existing style transfer methods. On one hand, it can avoid distortions in local style patterns, and allow semantic-level transfer, compared with neural parametric methods. On the other hand, it can preserve globally similar appearance to the style image, and avoid wash-out artifacts, compared with neural non-parametric methods. Based on the proposed loss, we also present a progressive feature-domain optimization approach. The experiments show that our method is widely applicable to various styles, and produces better quality than existing methods.", "This paper introduces an automatic method for editing a portrait photo so that the subject appears to be wearing makeup in the style of another person in a reference photo. Our unsupervised learning approach relies on a new framework of cycle-consistent generative adversarial networks. Different from the image domain transfer problem, our style transfer problem involves two asymmetric functions: a forward function encodes example-based style transfer, whereas a backward function removes the style. We construct two coupled networks to implement these functions - one that transfers makeup style and a second that can remove makeup - such that the output of their successive application to an input photo will match the input. The learned style network can then quickly apply an arbitrary makeup style to an arbitrary photo. We demonstrate the effectiveness on a broad range of portraits and styles." ] }
1905.10346	2946299443	Portrait editing is a popular subject in photo manipulation. The Generative Adversarial Network (GAN) advances the generating of realistic faces and allows more face editing. In this paper, we argue about three issues in existing techniques: diversity, quality, and controllability for portrait synthesis and editing. To address these issues, we propose a novel end-to-end learning framework that leverages conditional GANs guided by provided face masks for generating faces. The framework learns feature embeddings for every face component (e.g., mouth, hair, eye), separately, contributing to better correspondences for image translation, and local face editing. With the mask, our network is available to many applications, like face synthesis driven by mask, face Swap+ (including hair in swapping), and local manipulation. It can also boost the performance of face parsing a bit as an option of data augmentation.	Non-parametric image synthesis approaches @cite_14 @cite_37 @cite_19 usually generate new images by warping and stitching together existing patches from a database. The idea is extended by Qi al @cite_2 which combines neural networks to improve quality. Though similar at first glance, our method is intrinsically different from non-parametric image synthesis: our local embedding sub-network encodes facial instances as embeddings instead of image patches. New face images in our method are generated through the mask-guided generative sub-network, instead of warping and stitching image patches together. By jointly training all sub-networks, our model generates facial images that are higher quality than non-parametric methods that may also be difficult for them: synthesizing a face with an open mouth showing teeth from a source face with a closed mouth hiding all teeth.	{ "cite_N": [ "@cite_19", "@cite_37", "@cite_14", "@cite_2" ], "mid": [ "2134921974", "1563676183", "2171011251", "" ], "abstract": [ "We present a system for inserting new objects into existing photographs by querying a vast image-based object library, pre-computed using a publicly available Internet object database. The central goal is to shield the user from all of the arduous tasks typically involved in image compositing. The user is only asked to do two simple things: 1) pick a 3D location in the scene to place a new object; 2) select an object to insert using a hierarchical menu. We pose the problem of object insertion as a data-driven, 3D-based, context-sensitive object retrieval task. Instead of trying to manipulate the object to change its orientation, color distribution, etc. to fit the new image, we simply retrieve an object of a specified class that has all the required properties (camera pose, lighting, resolution, etc) from our large object library. We present new automatic algorithms for improving object segmentation and blending, estimating true 3D object size and orientation, and estimating scene lighting conditions. We also present an intuitive user interface that makes object insertion fast and simple even for the artistically challenged.", "Appearance Space Texture Synthesis (ASTS) provides fast texture synthesis by example. Unfortunately, speed is only achieved at the cost of a long pre-processing step. This is tedious for artists and limits the input resolution. In addition, the aggressively pruned search space results in strong artifacts when synthesizing multiple textures under user guidance (Texture-By-Numbers, TBN). In this thesis, we replace the k-coherence search step used by most modern synthesis algorithms with a new parallel, coherent random walk. We show that this drastically improves the synthesis quality with TBN, while maintaining the parallel nature, speed, and flexibility of the original ASTS runtime. Since it removes the expensive pre-computation of k-coherent candidates, we are able to use larger inputs, and start synthesis much faster. This is essential for artists designing high-quality exemplars.", "What can you do with a million images? In this paper we present a new image completion algorithm powered by a huge database of photographs gathered from the Web. The algorithm patches up holes in images by finding similar image regions in the database that are not only seamless but also semantically valid. Our chief insight is that while the space of images is effectively infinite, the space of semantically differentiable scenes is actually not that large. For many image completion tasks we are able to find similar scenes which contain image fragments that will convincingly complete the image. Our algorithm is entirely data-driven, requiring no annotations or labelling by the user. Unlike existing image completion methods, our algorithm can generate a diverse set of results for each input image and we allow users to select among them. We demonstrate the superiority of our algorithm over existing image completion approaches.", "" ] }
1905.10409	2946720822	We present a greedy-based approach to construct an efficient single hidden layer neural network with the ReLU activation that approximates a target function. In our approach we obtain a shallow network by utilizing a greedy algorithm with the prescribed dictionary provided by the available training data and a set of possible inner weights. To facilitate the greedy selection process we employ an integral representation of the network, based on the ridgelet transform, that significantly reduces the cardinality of the dictionary and hence promotes feasibility of the greedy selection. Our approach allows for the construction of efficient architectures which can be treated either as improved initializations to be used in place of random-based alternatives, or as fully-trained networks in certain cases, thus potentially nullifying the need for backpropagation training. Numerical experiments demonstrate the tenability of the proposed concept and its advantages compared to the conventional techniques for selecting architectures and initializations for neural networks.	The dictionary reduction step in our approach heavily relies on neural network integral representations, particularly on the so called ridgelet transform. Other similar approaches include those in @cite_0 and @cite_2 .	{ "cite_N": [ "@cite_0", "@cite_2" ], "mid": [ "1604316327", "1987618706" ], "abstract": [ "A new initialization method for hidden parameters in a neural network is proposed. Derived from the integral representation of the neural network, a nonparametric probability distribution of hidden parameters is introduced. In this proposal, hidden parameters are initialized by samples drawn from this distribution, and output parameters are fitted by ordinary linear regression. Numerical experiments show that backpropagation with proposed initialization converges faster than uniformly random initialization. Also it is shown that the proposed method achieves enough accuracy by itself without backpropagation in some cases.", "Abstract We present a strategy for learning optimal control. The approach uses functional-link neural network implementations which have several beneficial properties giving advantages over the more common generalized delta rule implementations. The learning task is decomposed into three parts: identification and monitoring, one-step-ahead control generation, and control path optimization. Each of these parts is accomplished with its own functional-link net and these are coordinated to provide the real-time learning of the optimal control path." ] }
1905.10409	2946720822	We present a greedy-based approach to construct an efficient single hidden layer neural network with the ReLU activation that approximates a target function. In our approach we obtain a shallow network by utilizing a greedy algorithm with the prescribed dictionary provided by the available training data and a set of possible inner weights. To facilitate the greedy selection process we employ an integral representation of the network, based on the ridgelet transform, that significantly reduces the cardinality of the dictionary and hence promotes feasibility of the greedy selection. Our approach allows for the construction of efficient architectures which can be treated either as improved initializations to be used in place of random-based alternatives, or as fully-trained networks in certain cases, thus potentially nullifying the need for backpropagation training. Numerical experiments demonstrate the tenability of the proposed concept and its advantages compared to the conventional techniques for selecting architectures and initializations for neural networks.	@cite_0 the authors suggest to use the values of the ridgelet transform to derive a probability distribution from which the inner weights are randomly selected. Then the resulted network is used as an initialization for training the network. In their approach the number of points on the sphere where the ridgelet transform needs to be computed, to derive the distribution can be potentially larger than what we require. And we replace the random selection step with a greedy selection step which, even though being more expensive as a selection method, results in a better approximation with a much fewer number of nodes.	{ "cite_N": [ "@cite_0" ], "mid": [ "1604316327" ], "abstract": [ "A new initialization method for hidden parameters in a neural network is proposed. Derived from the integral representation of the neural network, a nonparametric probability distribution of hidden parameters is introduced. In this proposal, hidden parameters are initialized by samples drawn from this distribution, and output parameters are fitted by ordinary linear regression. Numerical experiments show that backpropagation with proposed initialization converges faster than uniformly random initialization. Also it is shown that the proposed method achieves enough accuracy by itself without backpropagation in some cases." ] }
1905.10409	2946720822	We present a greedy-based approach to construct an efficient single hidden layer neural network with the ReLU activation that approximates a target function. In our approach we obtain a shallow network by utilizing a greedy algorithm with the prescribed dictionary provided by the available training data and a set of possible inner weights. To facilitate the greedy selection process we employ an integral representation of the network, based on the ridgelet transform, that significantly reduces the cardinality of the dictionary and hence promotes feasibility of the greedy selection. Our approach allows for the construction of efficient architectures which can be treated either as improved initializations to be used in place of random-based alternatives, or as fully-trained networks in certain cases, thus potentially nullifying the need for backpropagation training. Numerical experiments demonstrate the tenability of the proposed concept and its advantages compared to the conventional techniques for selecting architectures and initializations for neural networks.	Another related approach to ours is the Random Vector Functional-Link (or RVFL) network @cite_2 where the inner weights @math are generated randomly and the outer weights @math are found via a least square regularization. The random selection can be interpreted as a Monte-Carlo sampling for an integral representation. However the authors do not use any integral representation explicitly, only the fact of existence of such representation to justify their approach.	{ "cite_N": [ "@cite_2" ], "mid": [ "1987618706" ], "abstract": [ "Abstract We present a strategy for learning optimal control. The approach uses functional-link neural network implementations which have several beneficial properties giving advantages over the more common generalized delta rule implementations. The learning task is decomposed into three parts: identification and monitoring, one-step-ahead control generation, and control path optimization. Each of these parts is accomplished with its own functional-link net and these are coordinated to provide the real-time learning of the optimal control path." ] }
1905.10170	2945477881	Flow-based generative models have recently become one of the most efficient approaches to model the data generation. Indeed, they are constructed with a sequence of invertible and tractable transformations. Glow first introduced a simple type of generative flow using an invertible 1x1 convolution. However, the 1x1 convolution suffers from limited flexibility compared to the standard convolutions. In this paper, we propose a novel invertible nxn convolution approach that overcomes the limitations of the invertible 1x1 convolution. In addition, our proposed network is not only tractable and invertible but also uses fewer parameters than standard convolutions. The experiments on CIFAR-10, ImageNet, and Celeb-HQ datasets, have showed that our invertible nxn convolution helps to improve the performance of generative models significantly.	Germain et. al. @cite_14 introduced autoregressive autoencoders by constructing an extension of a non-variational autoencoder that can estimate distributions and is straightforward to compute their Jacobian determinant. Masked autoregressive flow @cite_8 is a type of normalizing flows, where the transformation layer is built as an autoregressive neural network. Inverse autoregressive flow @cite_12 formulates the conditional probability of the target variable as an autoregressive model.	{ "cite_N": [ "@cite_14", "@cite_12", "@cite_8" ], "mid": [ "1866230956", "2587284713", "2963047245" ], "abstract": [ "There has been a lot of recent interest in designing neural network models to estimate a distribution from a set of examples. We introduce a simple modification for autoencoder neural networks that yields powerful generative models. Our method masks the autoencoder's parameters to respect autoregressive constraints: each input is reconstructed only from previous inputs in a given ordering. Constrained this way, the autoencoder outputs can be interpreted as a set of conditional probabilities, and their product, the full joint probability. We can also train a single network that can decompose the joint probability in multiple different orderings. Our simple framework can be applied to multiple architectures, including deep ones. Vectorized implementations, such as on GPUs, are simple and fast. Experiments demonstrate that this approach is competitive with state-of-the-art tractable distribution estimators. At test time, the method is significantly faster and scales better than other autoregressive estimators.", "The framework of normalizing flows provides a general strategy for flexible variational inference of posteriors over latent variables. We propose a new type of normalizing flow, inverse autoregressive flow (IAF), that, in contrast to earlier published flows, scales well to high-dimensional latent spaces. The proposed flow consists of a chain of invertible transformations, where each transformation is based on an autoregressive neural network. In experiments, we show that IAF significantly improves upon diagonal Gaussian approximate posteriors. In addition, we demonstrate that a novel type of variational autoencoder, coupled with IAF, is competitive with neural autoregressive models in terms of attained log-likelihood on natural images, while allowing significantly faster synthesis.", "Autoregressive models are among the best performing neural density estimators. We describe an approach for increasing the flexibility of an autoregressive model, based on modelling the random numbers that the model uses internally when generating data. By constructing a stack of autoregressive models, each modelling the random numbers of the next model in the stack, we obtain a type of normalizing flow suitable for density estimation, which we call Masked Autoregressive Flow. This type of flow is closely related to Inverse Autoregressive Flow and is a generalization of Real NVP. Masked Autoregressive Flow achieves state-of-the-art performance in a range of general-purpose density estimation tasks." ] }
1905.10170	2945477881	Flow-based generative models have recently become one of the most efficient approaches to model the data generation. Indeed, they are constructed with a sequence of invertible and tractable transformations. Glow first introduced a simple type of generative flow using an invertible 1x1 convolution. However, the 1x1 convolution suffers from limited flexibility compared to the standard convolutions. In this paper, we propose a novel invertible nxn convolution approach that overcomes the limitations of the invertible 1x1 convolution. In addition, our proposed network is not only tractable and invertible but also uses fewer parameters than standard convolutions. The experiments on CIFAR-10, ImageNet, and Celeb-HQ datasets, have showed that our invertible nxn convolution helps to improve the performance of generative models significantly.	Kingma et. al. @cite_16 proposed to simplify the architecture via invertible @math convolutions. Learning a permutation matrix is a discrete optimization that is not amenable to gradient ascent. However, the permutation operation is simply a special case of a linear transformation with a square matrix. We can pursue this work with convolutional neural networks, as permuting the channels is equivalent to a @math convolution operation with an equal number of input and output channels. Therefore, the authors replace the fixed permutation with learned @math convolution operations.	{ "cite_N": [ "@cite_16" ], "mid": [ "2963139417" ], "abstract": [ "Flow-based generative models are conceptually attractive due to tractability of the exact log-likelihood, tractability of exact latent-variable inference, and parallelizability of both training and synthesis. In this paper we propose Glow, a simple type of generative flow using invertible 1x1 convolution. Using our method we demonstrate a significant improvement in log-likelihood and qualitative sample quality. Perhaps most strikingly, we demonstrate that a generative model optimized towards the plain log-likelihood objective is capable of efficient synthesis of large and subjectively realistic-looking images." ] }
1905.10170	2945477881	Flow-based generative models have recently become one of the most efficient approaches to model the data generation. Indeed, they are constructed with a sequence of invertible and tractable transformations. Glow first introduced a simple type of generative flow using an invertible 1x1 convolution. However, the 1x1 convolution suffers from limited flexibility compared to the standard convolutions. In this paper, we propose a novel invertible nxn convolution approach that overcomes the limitations of the invertible 1x1 convolution. In addition, our proposed network is not only tractable and invertible but also uses fewer parameters than standard convolutions. The experiments on CIFAR-10, ImageNet, and Celeb-HQ datasets, have showed that our invertible nxn convolution helps to improve the performance of generative models significantly.	@cite_16 performs an affine transformation using scale and bias parameters per channel. This layer simply shifts and scales the activations, with data-dependent initialization that normalizes the activations given an initial minibatch of data. This allows scaling down the minibatch size to 1 (for large images) and scaling up the size of the model.	{ "cite_N": [ "@cite_16" ], "mid": [ "2963139417" ], "abstract": [ "Flow-based generative models are conceptually attractive due to tractability of the exact log-likelihood, tractability of exact latent-variable inference, and parallelizability of both training and synthesis. In this paper we propose Glow, a simple type of generative flow using invertible 1x1 convolution. Using our method we demonstrate a significant improvement in log-likelihood and qualitative sample quality. Perhaps most strikingly, we demonstrate that a generative model optimized towards the plain log-likelihood objective is capable of efficient synthesis of large and subjectively realistic-looking images." ] }
1905.10452	2946521496	We present a theoretical and experimental investigation of the quantization problem for artificial neural networks. We provide a mathematical definition of quantized neural networks and analyze their approximation capabilities, showing in particular that any Lipschitz-continuous map defined on a hypercube can be uniformly approximated by a quantized neural network. We then focus on the regularization effect of additive noise on the arguments of multi-step functions inherent to the quantization of continuous variables. In particular, when the expectation operator is applied to a non-differentiable multi-step random function, and if the underlying probability density is differentiable (in either classical or weak sense), then a differentiable function is retrieved, with explicit bounds on its Lipschitz constant. Based on these results, we propose a novel gradient-based training algorithm for quantized neural networks that generalizes the straight-through estimator, acting on noise applied to the network's parameters. We evaluate our algorithm on the CIFAR-10 and ImageNet image classification benchmarks, showing state-of-the-art performance on AlexNet and MobileNetV2 for ternary networks.	The two most popular algorithms to train weights- @math -quantized DNNs are (BC) @cite_31 and (INQ) @cite_6 . BC is given a network architecture initialized with full-precision weights. Before inference is performed, a quantization function @math (the generalized Heaviside with codomain @math , also known as the sign function) is applied to these weights and returns binary values. Data is then propagated forward using these values as weights, the loss is evaluated, and gradients are computed. The updates directed to the binary weights are then applied to the weights instead, passing through the sign functions unaltered. More formally, STE replaces the exact distributional derivative @math of the sign function with the derivative of the hard hyperbolic tangent: ,. INQ is given a network architecture initialized with full-precision weights. The algorithm defines a quantization set @math , partitions the weights into a fixed number of subsets and defines a corresponding number of quantization time steps. INQ then starts training the full-precision model. When a quantization time step is reached, the corresponding weights subset is (i.e., its elements are projected onto the nearest quantization levels and never updated again).	{ "cite_N": [ "@cite_31", "@cite_6" ], "mid": [ "2963114950", "2964228333" ], "abstract": [ "Deep Neural Networks (DNN) have achieved state-of-the-art results in a wide range of tasks, with the best results obtained with large training sets and large models. In the past, GPUs enabled these breakthroughs because of their greater computational speed. In the future, faster computation at both training and test time is likely to be crucial for further progress and for consumer applications on low-power devices. As a result, there is much interest in research and development of dedicated hardware for Deep Learning (DL). Binary weights, i.e., weights which are constrained to only two possible values (e.g. -1 or 1), would bring great benefits to specialized DL hardware by replacing many multiply-accumulate operations by simple accumulations, as multipliers are the most space and power-hungry components of the digital implementation of neural networks. We introduce BinaryConnect, a method which consists in training a DNN with binary weights during the forward and backward propagations, while retaining precision of the stored weights in which gradients are accumulated. Like other dropout schemes, we show that BinaryConnect acts as regularizer and we obtain near state-of-the-art results with BinaryConnect on the permutation-invariant MNIST, CIFAR-10 and SVHN.", "This paper presents incremental network quantization (INQ), a novel method, targeting to efficiently convert any pre-trained full-precision convolutional neural network (CNN) model into a low-precision version whose weights are constrained to be either powers of two or zero. Unlike existing methods which are struggled in noticeable accuracy loss, our INQ has the potential to resolve this issue, as benefiting from two innovations. On one hand, we introduce three interdependent operations, namely weight partition, group-wise quantization and re-training. A well-proven measure is employed to divide the weights in each layer of a pre-trained CNN model into two disjoint groups. The weights in the first group are responsible to form a low-precision base, thus they are quantized by a variable-length encoding method. The weights in the other group are responsible to compensate for the accuracy loss from the quantization, thus they are the ones to be re-trained. On the other hand, these three operations are repeated on the latest re-trained group in an iterative manner until all the weights are converted into low-precision ones, acting as an incremental network quantization and accuracy enhancement procedure. Extensive experiments on the ImageNet classification task using almost all known deep CNN architectures including AlexNet, VGG-16, GoogleNet and ResNets well testify the efficacy of the proposed method. Specifically, at 5-bit quantization (a variable-length encoding: 1 bit for representing zero value, and the remaining 4 bits represent at most 16 different values for the powers of two), our models have improved accuracy than the 32-bit floating-point references. Taking ResNet-18 as an example, we further show that our quantized models with 4-bit, 3-bit and 2-bit ternary weights have improved or very similar accuracy against its 32-bit floating-point baseline. Besides, impressive results with the combination of network pruning and INQ are also reported. We believe that our method sheds new insights on how to make deep CNNs to be applicable on mobile or embedded devices. The code will be made publicly available." ] }
1905.10452	2946521496	We present a theoretical and experimental investigation of the quantization problem for artificial neural networks. We provide a mathematical definition of quantized neural networks and analyze their approximation capabilities, showing in particular that any Lipschitz-continuous map defined on a hypercube can be uniformly approximated by a quantized neural network. We then focus on the regularization effect of additive noise on the arguments of multi-step functions inherent to the quantization of continuous variables. In particular, when the expectation operator is applied to a non-differentiable multi-step random function, and if the underlying probability density is differentiable (in either classical or weak sense), then a differentiable function is retrieved, with explicit bounds on its Lipschitz constant. Based on these results, we propose a novel gradient-based training algorithm for quantized neural networks that generalizes the straight-through estimator, acting on noise applied to the network's parameters. We evaluate our algorithm on the CIFAR-10 and ImageNet image classification benchmarks, showing state-of-the-art performance on AlexNet and MobileNetV2 for ternary networks.	As an evolution of BC, activations- @math -quantized DNNs were first investigated by Hubara @cite_7 . They used the sign function also as the activation function in all the models they tested, and STE was effectively applied both to weights and activation functions. Building on STE, many algorithms have been developed. XNOR-Net @cite_13 tried to preserve inner products of full-precision filters and representation vectors by projecting them onto optimal binary vectors. Elaborating on this idea, ABC-Net @cite_26 decomposed filters onto a suitable binary basis and tried to learn multi-step activation functions that could be expressed as weighted sums of Heaviside functions. Taking the concept even further, @cite_11 decomposed entire residual branches into multiple branches modelled by binary weights and multi-step activation functions. Interestingly, GXNOR-Net @cite_1 presented a solution to update weights that does not use continuous weights, but uses the information carried by gradients to perform discrete state transitions instead.	{ "cite_N": [ "@cite_26", "@cite_7", "@cite_1", "@cite_13", "@cite_11" ], "mid": [ "2964008506", "2524428287", "2962939807", "2300242332", "2887204394" ], "abstract": [ "We introduce a novel scheme to train binary convolutional neural networks (CNNs) -- CNNs with weights and activations constrained to -1,+1 at run-time. It has been known that using binary weights and activations drastically reduce memory size and accesses, and can replace arithmetic operations with more efficient bitwise operations, leading to much faster test-time inference and lower power consumption. However, previous works on binarizing CNNs usually result in severe prediction accuracy degradation. In this paper, we address this issue with two major innovations: (1) approximating full-precision weights with the linear combination of multiple binary weight bases; (2) employing multiple binary activations to alleviate information loss. The implementation of the resulting binary CNN, denoted as ABC-Net, is shown to achieve much closer performance to its full-precision counterpart, and even reach the comparable prediction accuracy on ImageNet and forest trail datasets, given adequate binary weight bases and activations.", "We introduce a method to train Quantized Neural Networks (QNNs) -- neural networks with extremely low precision (e.g., 1-bit) weights and activations, at run-time. At traintime the quantized weights and activations are used for computing the parameter gradients. During the forward pass, QNNs drastically reduce memory size and accesses, and replace most arithmetic operations with bit-wise operations. As a result, power consumption is expected to be drastically reduced. We trained QNNs over the MNIST, CIFAR-10, SVHN and ImageNet datasets. The resulting QNNs achieve prediction accuracy comparable to their 32-bit counterparts. For example, our quantized version of AlexNet with 1-bit weights and 2-bit activations achieves 51 top-1 accuracy. Moreover, we quantize the parameter gradients to 6-bits as well which enables gradients computation using only bit-wise operation. Quantized recurrent neural networks were tested over the Penn Treebank dataset, and achieved comparable accuracy as their 32-bit counterparts using only 4-bits. Last but not least, we programmed a binary matrix multiplication GPU kernel with which it is possible to run our MNIST QNN 7 times faster than with an unoptimized GPU kernel, without suffering any loss in classification accuracy. The QNN code is available online.", "Abstract Although deep neural networks (DNNs) are being a revolutionary power to open up the AI era, the notoriously huge hardware overhead has challenged their applications. Recently, several binary and ternary networks, in which the costly multiply-accumulate operations can be replaced by accumulations or even binary logic operations, make the on-chip training of DNNs quite promising. Therefore there is a pressing need to build an architecture that could subsume these networks under a unified framework that achieves both higher performance and less overhead. To this end, two fundamental issues are yet to be addressed. The first one is how to implement the back propagation when neuronal activations are discrete. The second one is how to remove the full-precision hidden weights in the training phase to break the bottlenecks of memory computation consumption. To address the first issue, we present a multi-step neuronal activation discretization method and a derivative approximation technique that enable the implementing the back propagation algorithm on discrete DNNs. While for the second issue, we propose a discrete state transition (DST) methodology to constrain the weights in a discrete space without saving the hidden weights. Through this way, we build a unified framework that subsumes the binary or ternary networks as its special cases, and under which a heuristic algorithm is provided at the website https: github.com AcrossV Gated-XNOR . More particularly, we find that when both the weights and activations become ternary values, the DNNs can be reduced to sparse binary networks, termed as gated XNOR networks (GXNOR-Nets) since only the event of non-zero weight and non-zero activation enables the control gate to start the XNOR logic operations in the original binary networks. This promises the event-driven hardware design for efficient mobile intelligence. We achieve advanced performance compared with state-of-the-art algorithms. Furthermore, the computational sparsity and the number of states in the discrete space can be flexibly modified to make it suitable for various hardware platforms.", "We propose two efficient approximations to standard convolutional neural networks: Binary-Weight-Networks and XNOR-Networks. In Binary-Weight-Networks, the filters are approximated with binary values resulting in 32 ( ) memory saving. In XNOR-Networks, both the filters and the input to convolutional layers are binary. XNOR-Networks approximate convolutions using primarily binary operations. This results in 58 ( ) faster convolutional operations (in terms of number of the high precision operations) and 32 ( ) memory savings. XNOR-Nets offer the possibility of running state-of-the-art networks on CPUs (rather than GPUs) in real-time. Our binary networks are simple, accurate, efficient, and work on challenging visual tasks. We evaluate our approach on the ImageNet classification task. The classification accuracy with a Binary-Weight-Network version of AlexNet is the same as the full-precision AlexNet. We compare our method with recent network binarization methods, BinaryConnect and BinaryNets, and outperform these methods by large margins on ImageNet, more than (16 , ) in top-1 accuracy. Our code is available at: http: allenai.org plato xnornet.", "In this paper, we propose to train a network with binary weights and low-bitwidth activations, designed especially for mobile devices with limited power consumption. Most previous works on quantizing CNNs uncritically assume the same architecture, though with reduced precision. However, we take the view that for best performance it is possible (and even likely) that a different architecture may be better suited to dealing with low precision weights and activations. Specifically, we propose a \"network expansion\" strategy in which we aggregate a set of homogeneous low-precision branches to implicitly reconstruct the full-precision intermediate feature maps. Moreover, we also propose a group-wise feature approximation strategy which is very flexible and highly accurate. Experiments on ImageNet classification tasks demonstrate the superior performance of the proposed model, named Group-Net, over various popular architectures. In particular, with binary weights and activations, we outperform the previous best binary neural network in terms of accuracy as well as saving more than 5 times computational complexity on ImageNet with ResNet-18 and ResNet-50." ] }
1905.10452	2946521496	We present a theoretical and experimental investigation of the quantization problem for artificial neural networks. We provide a mathematical definition of quantized neural networks and analyze their approximation capabilities, showing in particular that any Lipschitz-continuous map defined on a hypercube can be uniformly approximated by a quantized neural network. We then focus on the regularization effect of additive noise on the arguments of multi-step functions inherent to the quantization of continuous variables. In particular, when the expectation operator is applied to a non-differentiable multi-step random function, and if the underlying probability density is differentiable (in either classical or weak sense), then a differentiable function is retrieved, with explicit bounds on its Lipschitz constant. Based on these results, we propose a novel gradient-based training algorithm for quantized neural networks that generalizes the straight-through estimator, acting on noise applied to the network's parameters. We evaluate our algorithm on the CIFAR-10 and ImageNet image classification benchmarks, showing state-of-the-art performance on AlexNet and MobileNetV2 for ternary networks.	The transition from continuous to finite spaces wipes out the differentiability property on which the backpropagation algorithm @cite_2 is based. In this context, the learning problem is converted from a numerical optimization problem into a discrete optimization one, where the state space is finite but very large. Gradient-free optimization algorithms, such as integer programming or Monte Carlo methods, quickly become unfeasible due to the combinatorial explosion of configurations. Although not formally correct in the context of discrete optimization, gradient-based optimization has been effectively brought back into the picture by STE. Nonetheless, the reason for its success is still mathematically unclear.	{ "cite_N": [ "@cite_2" ], "mid": [ "1498436455" ], "abstract": [ "We describe a new learning procedure, back-propagation, for networks of neurone-like units. The procedure repeatedly adjusts the weights of the connections in the network so as to minimize a measure of the difference between the actual output vector of the net and the desired output vector. As a result of the weight adjustments, internal ‘hidden’ units which are not part of the input or output come to represent important features of the task domain, and the regularities in the task are captured by the interactions of these units. The ability to create useful new features distinguishes back-propagation from earlier, simpler methods such as the perceptron-convergence procedure1." ] }
1905.10456	2946487406	The present paper considers leveraging network topology information to improve the convergence rate of ADMM for decentralized optimization, where networked nodes work collaboratively to minimize the objective. Such problems can be solved efficiently using ADMM via decomposing the objective into easier subproblems. Properly exploiting network topology can significantly improve the algorithm performance. Hybrid ADMM explores the direction of exploiting node information by taking into account node centrality but fails to utilize edge information. This paper fills the gap by incorporating both node and edge information and provides a novel convergence rate bound for decentralized ADMM that explicitly depends on network topology. Such a novel bound is attainable for certain class of problems, thus tight. The explicit dependence further suggests possible directions to optimal design of edge weights to achieve the best performance. Numerical experiments show that simple heuristic methods could achieve better performance, and also exhibits robustness to topology changes.	Our work builds on top of the ADMM algorithm. Since proposed in 1970s @cite_14 , ADMM has been successfully applied to many areas, especially distributed optimization and statistical learning, see @cite_25 and references therein. Decentralized ADMM has widely appreciated merits in applications that requires , see @cite_37 for applications in signal processing. To take advantage of topology information of nodes, hybrid ADMM @cite_28 adopt a local aggregate model where multiple local fusion centers work together each of which coordinates a subset of all nodes. A closely related work @cite_10 considers cluster-based ADMM that subsumes decentralized ADMM as special cases, but requires another subprocedure, e.g., gossip algorithm, to compute cluster related quantities.	{ "cite_N": [ "@cite_37", "@cite_14", "@cite_28", "@cite_10", "@cite_25" ], "mid": [ "1943968880", "2244987555", "2902441873", "1537500458", "2164278908" ], "abstract": [ "This chapter deals with decentralized learning algorithms for in-network processing of graph-valued data. A generic learning problem is formulated and recast into a separable form, which is iteratively minimized using the alternating-direction method of multipliers (ADMM) so as to gain the desired degree of parallelization. Without exchanging elements from the distributed training sets and keeping inter-node communications at affordable levels, the local (per-node) learners consent to the desired quantity inferred globally, meaning the one obtained if the entire training data set were centrally available. Impact of the decentralized learning framework to contemporary wireless communications and networking tasks is illustrated through case studies including target tracking using wireless sensor networks, unveiling Internet traffic anomalies, power system state estimation, as well as spectrum cartography for wireless cognitive radio networks.", "", "The present work introduces the hybrid consensus alternating direction method of multipliers (H-CADMM), a novel framework for optimization over networks which unifies existing distributed optimization approaches, including the centralized and the decentralized consensus ADMM. H-CADMM provides a flexible tool that leverages the underlying graph topology in order to achieve a desirable sweet spot between node-to-node communication overhead and rate of convergence—thereby alleviating known limitations of both C-CADMM and D-CADMM. A rigorous analysis of the novel method establishes linear convergence rate and also guides the choice of parameters to optimize this rate. The novel hybrid update rules of H-CADMM lend themselves to “in-network acceleration” that is shown to effect considerable—and essentially “free-of-charge”—performance boost over the fully decentralized ADMM. Comprehensive numerical tests validate the analysis and showcase the potential of the method in tackling efficiently, widely useful learning tasks.", "Consider a set of @math agents seeking to solve distributively the minimization problem @math where the convex functions @math are local to the agents. The popular Alternating Direction Method of Multipliers has the potential to handle distributed optimization problems of this kind. We provide a general reformulation of the problem and obtain a class of distributed algorithms which encompass various network architectures. The rate of convergence of our method is considered. It is assumed that the infimum of the problem is reached at a point @math , the functions @math are twice differentiable at this point and @math in the positive definite ordering of symmetric matrices. With these assumptions, it is shown that the convergence to the consensus @math is linear and the exact rate is provided. Application examples where this rate can be optimized with respect to the ADMM free parameter @math are also given.", "Many problems of recent interest in statistics and machine learning can be posed in the framework of convex optimization. Due to the explosion in size and complexity of modern datasets, it is increasingly important to be able to solve problems with a very large number of features or training examples. As a result, both the decentralized collection or storage of these datasets as well as accompanying distributed solution methods are either necessary or at least highly desirable. In this review, we argue that the alternating direction method of multipliers is well suited to distributed convex optimization, and in particular to large-scale problems arising in statistics, machine learning, and related areas. The method was developed in the 1970s, with roots in the 1950s, and is equivalent or closely related to many other algorithms, such as dual decomposition, the method of multipliers, Douglas–Rachford splitting, Spingarn's method of partial inverses, Dykstra's alternating projections, Bregman iterative algorithms for l1 problems, proximal methods, and others. After briefly surveying the theory and history of the algorithm, we discuss applications to a wide variety of statistical and machine learning problems of recent interest, including the lasso, sparse logistic regression, basis pursuit, covariance selection, support vector machines, and many others. We also discuss general distributed optimization, extensions to the nonconvex setting, and efficient implementation, including some details on distributed MPI and Hadoop MapReduce implementations." ] }
1905.10456	2946487406	The present paper considers leveraging network topology information to improve the convergence rate of ADMM for decentralized optimization, where networked nodes work collaboratively to minimize the objective. Such problems can be solved efficiently using ADMM via decomposing the objective into easier subproblems. Properly exploiting network topology can significantly improve the algorithm performance. Hybrid ADMM explores the direction of exploiting node information by taking into account node centrality but fails to utilize edge information. This paper fills the gap by incorporating both node and edge information and provides a novel convergence rate bound for decentralized ADMM that explicitly depends on network topology. Such a novel bound is attainable for certain class of problems, thus tight. The explicit dependence further suggests possible directions to optimal design of edge weights to achieve the best performance. Numerical experiments show that simple heuristic methods could achieve better performance, and also exhibits robustness to topology changes.	There are a lot works studying convergence of ADMM since it was proposed @cite_39 , see also @cite_25 @cite_8 . Convergence results for decentralized ADMM did not appear until recently @cite_24 @cite_35 @cite_7 @cite_33 , due to the difficulty of analyzing the convergence of multi-block ADMM @cite_31 . The relation between convergence rate of ADMM and the topology of underlying graph is explored using lifted Markov chain @cite_4 . Convergence of hybrid ADMM has also been shown @cite_28 .	{ "cite_N": [ "@cite_35", "@cite_4", "@cite_33", "@cite_7", "@cite_8", "@cite_28", "@cite_39", "@cite_24", "@cite_31", "@cite_25" ], "mid": [ "1549918636", "2763281404", "2962782955", "2123705108", "1603765807", "2902441873", "2045079045", "2105693192", "2111076441", "2164278908" ], "abstract": [ "The formulation @math minx,yf(x)+g(y),subjecttoAx+By=b,where f and g are extended-value convex functions, arises in many application areas such as signal processing, imaging and image processing, statistics, and machine learning either naturally or after variable splitting. In many common problems, one of the two objective functions is strictly convex and has Lipschitz continuous gradient. On this kind of problem, a very effective approach is the alternating direction method of multipliers (ADM or ADMM), which solves a sequence of f g-decoupled subproblems. However, its effectiveness has not been matched by a provably fast rate of convergence; only sublinear rates such as O(1 k) and @math O(1 k2) were recently established in the literature, though the O(1 k) rates do not require strong convexity. This paper shows that global linear convergence can be guaranteed under the assumptions of strong convexity and Lipschitz gradient on one of the two functions, along with certain rank assumptions on A and B. The result applies to various generalizations of ADM that allow the subproblems to be solved faster and less exactly in certain manners. The derived rate of convergence also provides some theoretical guidance for optimizing the ADM parameters. In addition, this paper makes meaningful extensions to the existing global convergence theory of ADM generalizations.", "When solving consensus optimization problems over a graph, there is often an explicit characterization of the convergence rate of Gradient Descent (GD) using the spectrum of the graph Laplacian. The same type of problems under the Alternating Direction Method of Multipliers (ADMM) are, however, poorly understood. For instance, simple but important non-strongly-convex consensus problems have not yet being analyzed, especially concerning the dependency of the convergence rate on the graph topology. Recently, for a non-strongly-convex consensus problem, a connection between distributed ADMM and lifted Markov chains was proposed, followed by a conjecture that ADMM is faster than GD by a square root factor in its convergence time, in close analogy to the mixing speedup achieved by lifting several Markov chains. Nevertheless, a proof of such a claim is is still lacking. Here we provide a full characterization of the convergence of distributed over-relaxed ADMM for the same type of consensus problem in terms of the topology of the underlying graph. Our results provide explicit formulas for optimal parameter selection in terms of the second largest eigenvalue of the transition matrix of the graph's random walk. Another consequence of our results is a proof of the aforementioned conjecture, which interestingly, we show it is valid for any graph, even the ones whose random walks cannot be accelerated via Markov chain lifting.", "We propose a new distributed algorithm based on alternating direction method of multipliers (ADMM) to minimize sum of locally known convex functions using communication over a network. This optimization problem emerges in many applications in distributed machine learning and statistical estimation. Our algorithm allows for a general choice of the communication weight matrix, which is used to combine the iterates at different nodes. We show that when functions are convex, both the objective function values and the feasibility violation converge with rate @math , where @math is the number of iterations. We then show that when functions are strongly convex and have Lipschitz continuous gradients, the sequence generated by our algorithm converges linearly to the optimal solution. In particular, an @math -optimal solution can be computed with @math iterations, where @math is the condition number of the problem. Our analysis highlights the effect of network and communication weights on the convergence rate through degrees of the nodes, the smallest nonzero eigenvalue, and operator norm of the communication matrix.", "In decentralized consensus optimization, a connected network of agents collaboratively minimize the sum of their local objective functions over a common decision variable, where their information exchange is restricted between the neighbors. To this end, one can first obtain a problem reformulation and then apply the alternating direction method of multipliers (ADMM). The method applies iterative computation at the individual agents and information exchange between the neighbors. This approach has been observed to converge quickly and deemed powerful. This paper establishes its linear convergence rate for the decentralized consensus optimization problem with strongly convex local objective functions. The theoretical convergence rate is explicitly given in terms of the network topology, the properties of local objective functions, and the algorithm parameter. This result is not only a performance guarantee but also a guideline toward accelerating the ADMM convergence.", "", "The present work introduces the hybrid consensus alternating direction method of multipliers (H-CADMM), a novel framework for optimization over networks which unifies existing distributed optimization approaches, including the centralized and the decentralized consensus ADMM. H-CADMM provides a flexible tool that leverages the underlying graph topology in order to achieve a desirable sweet spot between node-to-node communication overhead and rate of convergence—thereby alleviating known limitations of both C-CADMM and D-CADMM. A rigorous analysis of the novel method establishes linear convergence rate and also guides the choice of parameters to optimize this rate. The novel hybrid update rules of H-CADMM lend themselves to “in-network acceleration” that is shown to effect considerable—and essentially “free-of-charge”—performance boost over the fully decentralized ADMM. Comprehensive numerical tests validate the analysis and showcase the potential of the method in tackling efficiently, widely useful learning tasks.", "For variational problems of the form Infv∈V f(Av)+g(v) , we propose a dual method which decouples the difficulties relative to the functionals f and g from the possible ill-conditioning effects of the linear operator A. The approach is based on the use of an Augmented Lagrangian functional and leads to an efficient and simply implementable algorithm. We study also the finite element approximation of such problems, compatible with the use of our algorithm. The method is finally applied to solve several problems of continuum mechanics.", "We analyze the convergence rate of the alternating direction method of multipliers (ADMM) for minimizing the sum of two or more nonsmooth convex separable functions subject to linear constraints. Previous analysis of the ADMM typically assumes that the objective function is the sum of only two convex functions defined on two separable blocks of variables even though the algorithm works well in numerical experiments for three or more blocks. Moreover, there has been no rate of convergence analysis for the ADMM without strong convexity in the objective function. In this paper we establish the global R-linear convergence of the ADMM for minimizing the sum of any number of convex separable functions, assuming that a certain error bound condition holds true and the dual stepsize is sufficiently small. Such an error bound condition is satisfied for example when the feasible set is a compact polyhedron and the objective function consists of a smooth strictly convex function composed with a linear mapping, and a nonsmooth @math l1 regularizer. This result implies the linear convergence of the ADMM for contemporary applications such as LASSO without assuming strong convexity of the objective function.", "The alternating direction method of multipliers (ADMM) has been widely used for solving structured convex optimization problems. In particular, the ADMM can solve convex programs that minimize the sum of @math convex functions whose variables are linked by some linear constraints. While the convergence of the ADMM for @math was well established in the literature, it remained an open problem for a long time whether the ADMM for @math is still convergent. Recently, it was shown in [, Math. Program. (2014), DOI 10.1007 s10107-014-0826-5.] that without additional conditions the ADMM for @math generally fails to converge. In this paper, we show that under some easily verifiable and reasonable conditions the global linear convergence of the ADMM when @math can still be ensured, which is important since the ADMM is a popular method for solving large-scale multiblock optimization models and is known to perform very well in practice even when @math . Our study aims to offer an explanation for...", "Many problems of recent interest in statistics and machine learning can be posed in the framework of convex optimization. Due to the explosion in size and complexity of modern datasets, it is increasingly important to be able to solve problems with a very large number of features or training examples. As a result, both the decentralized collection or storage of these datasets as well as accompanying distributed solution methods are either necessary or at least highly desirable. In this review, we argue that the alternating direction method of multipliers is well suited to distributed convex optimization, and in particular to large-scale problems arising in statistics, machine learning, and related areas. The method was developed in the 1970s, with roots in the 1950s, and is equivalent or closely related to many other algorithms, such as dual decomposition, the method of multipliers, Douglas–Rachford splitting, Spingarn's method of partial inverses, Dykstra's alternating projections, Bregman iterative algorithms for l1 problems, proximal methods, and others. After briefly surveying the theory and history of the algorithm, we discuss applications to a wide variety of statistical and machine learning problems of recent interest, including the lasso, sparse logistic regression, basis pursuit, covariance selection, support vector machines, and many others. We also discuss general distributed optimization, extensions to the nonconvex setting, and efficient implementation, including some details on distributed MPI and Hadoop MapReduce implementations." ] }
1905.10456	2946487406	The present paper considers leveraging network topology information to improve the convergence rate of ADMM for decentralized optimization, where networked nodes work collaboratively to minimize the objective. Such problems can be solved efficiently using ADMM via decomposing the objective into easier subproblems. Properly exploiting network topology can significantly improve the algorithm performance. Hybrid ADMM explores the direction of exploiting node information by taking into account node centrality but fails to utilize edge information. This paper fills the gap by incorporating both node and edge information and provides a novel convergence rate bound for decentralized ADMM that explicitly depends on network topology. Such a novel bound is attainable for certain class of problems, thus tight. The explicit dependence further suggests possible directions to optimal design of edge weights to achieve the best performance. Numerical experiments show that simple heuristic methods could achieve better performance, and also exhibits robustness to topology changes.	In the setting of distributed averaging, the optimal weights and step size are considered in @cite_20 . Decentralized ADMM over weighted graphs was also considered @cite_36 , which offers the flexibility to optimize convergence rate by tuning weights. The importance of edges was incorporated into hybrid ADMM as well @cite_3 and its convergence rate bound was shown to be related to spectral properties of the communication graph.	{ "cite_N": [ "@cite_36", "@cite_3", "@cite_20" ], "mid": [ "2515624991", "2919745732", "2009531227" ], "abstract": [ "", "Distributed optimization has gained popularity in many areas because it can to cope with the increasing volume of data, and the corresponding demand for computational resources. A popular distributed solver is the alternating direction method of multipliers (ADMM). Fully decentralized (D) ADMM arises when node-to-node communications have to abide by the underlying network connectivity. DADMM’s convergence however, slows down as the network diameter grows large. To deal with this challenge, the recently proposed hybrid (H) ADMM provides considerable performance boost over DADMM by exploiting local topology information. But HADMM only applies to unweighted graphs. The present contribution develops a weighted hybrid (WH) consensus-based ADMM approach that can deal with weighted graphs. The resultant scheme further improves the performance of HADMM through graph-aware weight tuning. Theoretical analysis offers convergence guarantees and establishes linear convergence rate, while numerical tests on various graphs demonstrate the WHADMM merits.", "We derive the optimal step-size and over-relaxation parameter that minimizes the convergence time of two ADMM-based algorithms for distributed averaging. Our study shows that the convergence times for given step-size and over-relaxation parameters depend on the spectral properties of the normalized Laplacian of the underlying communication graph. Motivated by this, we optimize the edge-weights of the communication graph to improve the convergence speed even further. The performance of the ADMM algorithms with our parameter selection are compared with alternatives from the literature in extensive numerical simulations on random graphs." ] }
1905.10456	2946487406	The present paper considers leveraging network topology information to improve the convergence rate of ADMM for decentralized optimization, where networked nodes work collaboratively to minimize the objective. Such problems can be solved efficiently using ADMM via decomposing the objective into easier subproblems. Properly exploiting network topology can significantly improve the algorithm performance. Hybrid ADMM explores the direction of exploiting node information by taking into account node centrality but fails to utilize edge information. This paper fills the gap by incorporating both node and edge information and provides a novel convergence rate bound for decentralized ADMM that explicitly depends on network topology. Such a novel bound is attainable for certain class of problems, thus tight. The explicit dependence further suggests possible directions to optimal design of edge weights to achieve the best performance. Numerical experiments show that simple heuristic methods could achieve better performance, and also exhibits robustness to topology changes.	Many convex optimization problems can be analyzed from the perspective of @cite_6 @cite_9 . For example, the problem of minimizing @math can be cast as finding the zero of the subdifferntial @math , which happens to be a maximal monotone operator provided @math is convex, closed and proper. However, finding zeros of two operators can be difficult, whereas methods can be employed to split the problem into easier subproblems @cite_38 @cite_16 @cite_9 . One prominent example is @cite_22 @cite_13 .	{ "cite_N": [ "@cite_38", "@cite_22", "@cite_9", "@cite_6", "@cite_16", "@cite_13" ], "mid": [ "2019569173", "2058532290", "", "2038497950", "1525779601", "2003372231" ], "abstract": [ "Splitting algorithms for the sum of two monotone operators.We study two splitting algorithms for (stationary and evolution) problems involving the sum of two monotone operators. These algorithms ar...", "This paper shows, by means of an operator called asplitting operator, that the Douglas--Rachford splitting method for finding a zero of the sum of two monotone operators is a special case of the proximal point algorithm. Therefore, applications of Douglas--Rachford splitting, such as the alternating direction method of multipliers for convex programming decomposition, are also special cases of the proximal point algorithm. This observation allows the unification and generalization of a variety of convex programming algorithms. By introducing a modified version of the proximal point algorithm, we derive a new,generalized alternating direction method of multipliers for convex programming. Advances of this sort illustrate the power and generality gained by adopting monotone operator theory as a conceptual framework.", "", "For the problem of minimizing a lower semicontinuous proper convex function f on a Hilbert space, the proximal point algorithm in exact form generates a sequence @math by taking @math to be the minimizes of @math , where @math . This algorithm is of interest for several reasons, but especially because of its role in certain computational methods based on duality, such as the Hestenes-Powell method of multipliers in nonlinear programming. It is investigated here in a more general form where the requirement for exact minimization at each iteration is weakened, and the subdifferential @math is replaced by an arbitrary maximal monotone operator T. Convergence is established under several criteria amenable to implementation. The rate of convergence is shown to be “typically” linear with an arbitrarily good modulus if @math stays large enough, in fact superlinear if @math . The case of @math is treated in extra detail. Applicati...", "An electrically-operated shutter provided with a change-over device or arrangement for effecting daylight and flash photography automatically shifting from a daylight mode to a flash photography mode as a function of the detected brightness level of the field being photographed. One embodiment provides for manually presetting the camera for flash photography which is then automatically carried out with an automatic change-over from a daylight photography mode to a flash photography mode. When carrying out daylight photography the power source energy is used in control and in energizing an electromagnet that controls the exposure termination under control of a delay circuit. When taking flash exposure the delay circuit is not employed and a fixed exposure time is used. The power supply is then used to energize the flashbulb and the accuracy of the shutter timing is unaffected as might be the case if the power source were energizing both the flashbulb and driving or energizing the shutter. The same shutter-operating elements are used in both modes of operation and the change-over arrangement eliminates the use of the delay circuit when the brightness level is below a predetermined level so that flash photography becomes necessary.", "" ] }
1905.10456	2946487406	The present paper considers leveraging network topology information to improve the convergence rate of ADMM for decentralized optimization, where networked nodes work collaboratively to minimize the objective. Such problems can be solved efficiently using ADMM via decomposing the objective into easier subproblems. Properly exploiting network topology can significantly improve the algorithm performance. Hybrid ADMM explores the direction of exploiting node information by taking into account node centrality but fails to utilize edge information. This paper fills the gap by incorporating both node and edge information and provides a novel convergence rate bound for decentralized ADMM that explicitly depends on network topology. Such a novel bound is attainable for certain class of problems, thus tight. The explicit dependence further suggests possible directions to optimal design of edge weights to achieve the best performance. Numerical experiments show that simple heuristic methods could achieve better performance, and also exhibits robustness to topology changes.	ADMM was shown to be a special case of a method @cite_27 @cite_16 . Therefore, one can prove convergence of ADMM through Douglas-Rachford algorithm. Recently, a tight linear rate for Douglas-Rachford algorithm has been published @cite_18 . The tight bound for ADMM is deduced from that of Douglas-Rachford algorithm, due to their equivalence. This connection, however, entails assumptions of the constraints that does not always hold in decentralized case.	{ "cite_N": [ "@cite_27", "@cite_16", "@cite_18" ], "mid": [ "121840966", "1525779601", "2963000588" ], "abstract": [ "A zoom objective with a varifocal ratio close to 2:1, particularly for a projector, has a stationary positive front component, a single axially shiftable negative lens and a stationary multilens rear component. An object-side positive lens member of the rear component, adjoining the shiftable negative lens, has a focal length smaller than that of the group constituted by the remaining, image-side lens members of this component. The negatively refracting third member of the rear component may be a doublet with a positively refracting cemented surface, convex toward the image side, or an equivalent lens pair separated by a small air gap.", "An electrically-operated shutter provided with a change-over device or arrangement for effecting daylight and flash photography automatically shifting from a daylight mode to a flash photography mode as a function of the detected brightness level of the field being photographed. One embodiment provides for manually presetting the camera for flash photography which is then automatically carried out with an automatic change-over from a daylight photography mode to a flash photography mode. When carrying out daylight photography the power source energy is used in control and in energizing an electromagnet that controls the exposure termination under control of a delay circuit. When taking flash exposure the delay circuit is not employed and a fixed exposure time is used. The power supply is then used to energize the flashbulb and the accuracy of the shutter timing is unaffected as might be the case if the power source were energizing both the flashbulb and driving or energizing the shutter. The same shutter-operating elements are used in both modes of operation and the change-over arrangement eliminates the use of the delay circuit when the brightness level is below a predetermined level so that flash photography becomes necessary.", "Recently, several convergence rate results for Douglas-Rachford splitting and the alternating direction method of multipliers (ADMM) have been presented in the literature. In this paper, we show global linear convergence rate bounds for Douglas-Rachford splitting and ADMM under strong convexity and smoothness assumptions. We further show that the rate bounds are tight for the class of problems under consideration for all feasible algorithm parameters. For problems that satisfy the assumptions, we show how to select step-size and metric for the algorithm that optimize the derived convergence rate bounds. For problems with a similar structure that do not satisfy the assumptions, we present heuristic step-size and metric selection methods." ] }
1905.10157	2945236295	Convolutional neural networks (CNNs) have achieved remarkable performance in various fields, particularly in the domain of computer vision. However, why this architecture works well remains to be a mystery. In this work we move a small step toward understanding the success of CNNs by investigating the learning dynamics of a two-layer nonlinear convolutional neural network over some specific data distributions. Rather than the typical Gaussian assumption for input data distribution, we consider a more realistic setting that each data point (e.g. image) contains a specific pattern determining its class label. Within this setting, we both theoretically and empirically show that some convolutional filters will learn the key patterns in data and the norm of these filters will dominate during the training process with stochastic gradient descent. And with any high probability, when the number of iterations is sufficiently large, the CNN model could obtain 100 accuracy over the considered data distributions. Our experiments demonstrate that for practical image classification tasks our findings still hold to some extent.	Recently, there are many works analyze the convergence property of FCNs through a trajectory-based analysis @cite_21 @cite_13 @cite_4 @cite_11 @cite_16 @cite_19 . These works mainly focused on over-parameterized two-layer FCNs. According to @cite_18 , their proof techniques can be roughly classified into two categories. @cite_21 @cite_13 @cite_4 analyzed the trajectory of parameters and showed that on the trajectory, the objective function satisfies certain gradient dominance property, while @cite_11 @cite_16 analyzed the trajectory of network predictions on training samples and showed that it enjoys a strongly-convex-like property.	{ "cite_N": [ "@cite_18", "@cite_4", "@cite_21", "@cite_19", "@cite_16", "@cite_13", "@cite_11" ], "mid": [ "2911867426", "2900959181", "2886067286", "2765861987", "2899790086", "2899748887", "2894604724" ], "abstract": [ "Recent works have cast some light on the mystery of why deep nets fit any data and generalize despite being very overparametrized. This paper analyzes training and generalization for a simple 2-layer ReLU net with random initialization, and provides the following improvements over recent works: (i) Using a tighter characterization of training speed than recent papers, an explanation for why training a neural net with random labels leads to slower training, as originally observed in [ ICLR'17]. (ii) Generalization bound independent of network size, using a data-dependent complexity measure. Our measure distinguishes clearly between random labels and true labels on MNIST and CIFAR, as shown by experiments. Moreover, recent papers require sample complexity to increase (slowly) with the size, while our sample complexity is completely independent of the network size. (iii) Learnability of a broad class of smooth functions by 2-layer ReLU nets trained via gradient descent. The key idea is to track dynamics of training and generalization via properties of a related kernel.", "We study the problem of training deep neural networks with Rectified Linear Unit (ReLU) activation function using gradient descent and stochastic gradient descent. In particular, we study the binary classification problem and show that for a broad family of loss functions, with proper random weight initialization, both gradient descent and stochastic gradient descent can find the global minima of the training loss for an over-parameterized deep ReLU network, under mild assumption on the training data. The key idea of our proof is that Gaussian random initialization followed by (stochastic) gradient descent produces a sequence of iterates that stay inside a small perturbation region centering around the initial weights, in which the empirical loss function of deep ReLU networks enjoys nice local curvature properties that ensure the global convergence of (stochastic) gradient descent. Our theoretical results shed light on understanding the optimization for deep learning, and pave the way for studying the optimization dynamics of training modern deep neural networks.", "Neural networks have many successful applications, while much less theoretical understanding has been gained. Towards bridging this gap, we study the problem of learning a two-layer overparameterized ReLU neural network for multi-class classification via stochastic gradient descent (SGD) from random initialization. In the overparameterized setting, when the data comes from mixtures of well-separated distributions, we prove that SGD learns a network with a small generalization error, albeit the network has enough capacity to fit arbitrary labels. Furthermore, the analysis provides interesting insights into several aspects of learning neural networks and can be verified based on empirical studies on synthetic data and on the MNIST dataset.", "Neural networks exhibit good generalization behavior in the over-parameterized regime, where the number of network parameters exceeds the number of observations. Nonetheless, current generalization bounds for neural networks fail to explain this phenomenon. In an attempt to bridge this gap, we study the problem of learning a two-layer over-parameterized neural network, when the data is generated by a linearly separable function. In the case where the network has Leaky ReLU activations, we provide both optimization and generalization guarantees for over-parameterized networks. Specifically, we prove convergence rates of SGD to a global minimum and provide generalization guarantees for this global minimum that are independent of the network size. Therefore, our result clearly shows that the use of SGD for optimization both finds a global minimum, and avoids overfitting despite the high capacity of the model. This is the first theoretical demonstration that SGD can avoid overfitting, when learning over-specified neural network classifiers.", "Gradient descent finds a global minimum in training deep neural networks despite the objective function being non-convex. The current paper proves gradient descent achieves zero training loss in polynomial time for a deep over-parameterized neural network with residual connections (ResNet). Our analysis relies on the particular structure of the Gram matrix induced by the neural network architecture. This structure allows us to show the Gram matrix is stable throughout the training process and this stability implies the global optimality of the gradient descent algorithm. We further extend our analysis to deep residual convolutional neural networks and obtain a similar convergence result.", "Deep neural networks (DNNs) have demonstrated dominating performance in many fields; since AlexNet, networks used in practice are going wider and deeper. On the theoretical side, a long line of works has been focusing on training neural networks with one hidden layer. The theory of multi-layer networks remains largely unsettled. In this work, we prove why stochastic gradient descent (SGD) can find @math on the training objective of DNNs in @math . We only make two assumptions: the inputs are non-degenerate and the network is over-parameterized. The latter means the network width is sufficiently large: @math in @math , the number of layers and in @math , the number of samples. Our key technique is to derive that, in a sufficiently large neighborhood of the random initialization, the optimization landscape is almost-convex and semi-smooth even with ReLU activations. This implies an equivalence between over-parameterized neural networks and neural tangent kernel (NTK) in the finite (and polynomial) width setting. As concrete examples, starting from randomly initialized weights, we prove that SGD can attain 100 training accuracy in classification tasks, or minimize regression loss in linear convergence speed, with running time polynomial in @math . Our theory applies to the widely-used but non-smooth ReLU activation, and to any smooth and possibly non-convex loss functions. In terms of network architectures, our theory at least applies to fully-connected neural networks, convolutional neural networks (CNN), and residual neural networks (ResNet).", "One of the mysteries in the success of neural networks is randomly initialized first order methods like gradient descent can achieve zero training loss even though the objective function is non-convex and non-smooth. This paper demystifies this surprising phenomenon for two-layer fully connected ReLU activated neural networks. For an @math hidden node shallow neural network with ReLU activation and @math training data, we show as long as @math is large enough and no two inputs are parallel, randomly initialized gradient descent converges to a globally optimal solution at a linear convergence rate for the quadratic loss function. Our analysis relies on the following observation: over-parameterization and random initialization jointly restrict every weight vector to be close to its initialization for all iterations, which allows us to exploit a strong convexity-like property to show that gradient descent converges at a global linear rate to the global optimum. We believe these insights are also useful in analyzing deep models and other first order methods." ] }
1905.10157	2945236295	Convolutional neural networks (CNNs) have achieved remarkable performance in various fields, particularly in the domain of computer vision. However, why this architecture works well remains to be a mystery. In this work we move a small step toward understanding the success of CNNs by investigating the learning dynamics of a two-layer nonlinear convolutional neural network over some specific data distributions. Rather than the typical Gaussian assumption for input data distribution, we consider a more realistic setting that each data point (e.g. image) contains a specific pattern determining its class label. Within this setting, we both theoretically and empirically show that some convolutional filters will learn the key patterns in data and the norm of these filters will dominate during the training process with stochastic gradient descent. And with any high probability, when the number of iterations is sufficiently large, the CNN model could obtain 100 accuracy over the considered data distributions. Our experiments demonstrate that for practical image classification tasks our findings still hold to some extent.	The works @cite_12 @cite_15 @cite_7 @cite_20 focused on analyzing the learning a two-layer CNN for regression problem with Gaussian inputs, under the teacher-student network settings. They showed that the dynamics could converge to true parameter or zero loss. @cite_1 considered learning a two-layer CNN for some binary classification task assuming a restricted input distribution. Multi-layer linear CNNs were studied in @cite_14 for binary classification problem with linear separable data distribution. And they proved that the dynamics could converge to a linear predictor related to the @math bridge penalty in the frequency domain.	{ "cite_N": [ "@cite_14", "@cite_7", "@cite_1", "@cite_15", "@cite_12", "@cite_20" ], "mid": [ "2963826371", "2591714514", "2786689480", "2768041186", "2963672337", "2963744427" ], "abstract": [ "We show that gradient descent on full-width linear convolutional networks of depth L converges to a linear predictor related to the l2 L bridge penalty in the frequency domain. This is in contrast to linearly fully connected networks, where gradient descent converges to the hard margin linear SVM solution, regardless of depth.", "Deep learning models are often successfully trained using gradient descent, despite the worst case hardness of the underlying non-convex optimization problem. The key question is then under what conditions can one prove that optimization will succeed. Here we provide a strong result of this kind. We consider a neural net with one hidden layer and a convolutional structure with no overlap, and a ReLU activation function. For this architecture we show that learning is NP-complete in the general case, but that when the input distribution is Gaussian, gradient descent converges to the global optimum in polynomial time. To the best of our knowledge, this is the first global optimality guarantee of gradient descent on a convolutional neural network with ReLU activations.", "A major challenge in understanding the generalization of deep learning is to explain why (stochastic) gradient descent can exploit the network architecture to find solutions that have good generalization performance when using high capacity models. We find simple but realistic examples showing that this phenomenon exists even when learning linear classifiers --- between two linear networks with the same capacity, the one with a convolutional layer can generalize better than the other when the data distribution has some underlying spatial structure. We argue that this difference results from a combination of the convolution architecture, data distribution and gradient descent, all of which are necessary to be included in a meaningful analysis. We provide a general analysis of the generalization performance as a function of data distribution and convolutional filter size, given gradient descent as the optimization algorithm, then interpret the results using concrete examples. Experimental results show that our analysis is able to explain what happens in our introduced examples.", "In this paper, we consider parameter recovery for non-overlapping convolutional neural networks (CNNs) with multiple kernels. We show that when the inputs follow Gaussian distribution and the sample size is sufficiently large, the squared loss of such CNNs is @math in a basin of attraction near the global optima for most popular activation functions, like ReLU, Leaky ReLU, Squared ReLU, Sigmoid and Tanh. The required sample complexity is proportional to the dimension of the input and polynomial in the number of kernels and a condition number of the parameters. We also show that tensor methods are able to initialize the parameters to the local strong convex region. Hence, for most smooth activations, gradient descent following tensor initialization is guaranteed to converge to the global optimal with time that is linear in input dimension, logarithmic in precision and polynomial in other factors. To the best of our knowledge, this is the first work that provides recovery guarantees for CNNs with multiple kernels under polynomial sample and computational complexities.", "", "" ] }
1905.10238	2951046677	Linking pronominal expressions to the correct references requires, in many cases, better analysis of the contextual information and external knowledge. In this paper, we propose a two-layer model for pronoun coreference resolution that leverages both context and external knowledge, where a knowledge attention mechanism is designed to ensure the model leveraging the appropriate source of external knowledge based on different context. Experimental results demonstrate the validity and effectiveness of our model, where it outperforms state-of-the-art models by a large margin.	In previous work, external knowledge including manually defined rules @cite_25 @cite_21 , such as number gender requirement of different pronouns, and world knowledge @cite_9 , such as selectional preference @cite_1 @cite_26 and eventuality knowledge @cite_29 , have been proved to be helpful for pronoun coreference resolution. Recently, with the development of deep learning, proposed an end-to-end model that learns contextual information with an LSTM module and proved that such knowledge is helpful for coreference resolution when the context is properly encoded. The aforementioned two types of knowledge have their own advantages: the contextual information covers diverse text expressions that are difficult to be predefined while the external knowledge is usually more precisely constructed and able to provide extra information beyond the training data. Different from previous work, we explore the possibility of joining the two types of knowledge for pronoun coreference resolution rather than use only one of them. To the best of our knowledge, this is the first attempt that uses deep learning model to incorporate contextual information and external knowledge for pronoun coreference resolution.	{ "cite_N": [ "@cite_26", "@cite_9", "@cite_21", "@cite_1", "@cite_29", "@cite_25" ], "mid": [ "2806102503", "2103095311", "", "2042008249", "2942882688", "" ], "abstract": [ "This paper presents a new distributed representation based approach to capture commonsense knowledge. Currently, we focus on meanings of verbs and related dependency relations. Built upon such dependency relations, a set of vector space models are developed to capture the meaning of the verb and transfer it to related ambiguous pronouns. To demonstrate the capability of this data-driven approach, we do not incorporate any human knowledge. Among all the vector models without human knowledge, we can achieve the best performance on all of the verb similarity datasets. We can also achieve superior performance on a subset of Winograd Schema challenge dataset, compared with other existing pronoun co-reference and embedding models.", "While world knowledge has been shown to improve learning-based coreference resolvers, the improvements were typically obtained by incorporating world knowledge into a fairly weak baseline resolver. Hence, it is not clear whether these benefits can carry over to a stronger baseline. Moreover, since there has been no attempt to apply different sources of world knowledge in combination to coreference resolution, it is not clear whether they offer complementary benefits to a resolver. We systematically compare commonly-used and under-investigated sources of world knowledge for coreference resolution by applying them to two learning-based coreference models and evaluating them on documents annotated with two different annotation schemes.", "", "Abstract The paper describes the way in which a Preference Semantics system for natural language analysis and generation tackles a difficult class of anaphoric inference problems: those requiring either analytic (conceptual) knowledge of a complex sort, or requiring weak inductive knowledge of the course of events in the real world. The method employed converts all available knowledge to a canonical template form and endeavors to create chains of non-reductive inferences from the unknowns to the possible referents. Its method for this is consistent with the overall principle of “semantic preference” used to set up the original meaning representation.", "Understanding human's language requires complex world knowledge. However, existing large-scale knowledge graphs mainly focus on knowledge about entities while ignoring knowledge about activities, states, or events, which are used to describe how entities or things act in the real world. To fill this gap, we develop ASER (activities, states, events, and their relations), a large-scale eventuality knowledge graph extracted from more than 11-billion-token unstructured textual data. ASER contains 15 relation types belonging to five categories, 194-million unique eventualities, and 64-million unique edges among them. Both human and extrinsic evaluations demonstrate the quality and effectiveness of ASER.", "" ] }
1905.09891	2944839214	Many current methods to learn intuitive physics are based on interaction networks and similar approaches. However, they rely on information that has proven difficult to estimate directly from image data in the past. We aim to narrow this gap by inferring all the semantic information needed from raw pixel data in the form of a scene-graph. Our approach is based on neural-symbolic capsules, which identify which objects in the scene are static, dynamic, elastic or rigid, possible joints between them, as well as their collision information. By integrating all this with interaction networks, we demonstrate how our method is able to learn intuitive physics directly from image sequences and apply its knowledge to new scenes and objects, resulting in an inverse-simulation pipeline.	While there are end-to-end proposals for intuitive physics @cite_4 @cite_1 @cite_16 , we will focus on interaction networks @cite_5 , which rely on semantic knowledge of the scene to do physical predictions. Using semantics as part of the process of learning or predicting intuitive physics has proven to be successful and is used in many related methods @cite_14 @cite_8 @cite_0 @cite_7 . Such models for physical scene understanding often follow the idea of an inverse game engine @cite_11 @cite_6 .	{ "cite_N": [ "@cite_14", "@cite_4", "@cite_7", "@cite_8", "@cite_1", "@cite_6", "@cite_0", "@cite_5", "@cite_16", "@cite_11" ], "mid": [ "2559857813", "2293598046", "2963464736", "", "2592431710", "2671811403", "2963370555", "2952915411", "2884775584", "2059100041" ], "abstract": [ "We present the Neural Physics Engine (NPE), a framework for learning simulators of intuitive physics that naturally generalize across variable object count and different scene configurations. We propose a factorization of a physical scene into composable object-based representations and a neural network architecture whose compositional structure factorizes object dynamics into pairwise interactions. Like a symbolic physics engine, the NPE is endowed with generic notions of objects and their interactions; realized as a neural network, it can be trained via stochastic gradient descent to adapt to specific object properties and dynamics of different worlds. We evaluate the efficacy of our approach on simple rigid body dynamics in two-dimensional worlds. By comparing to less structured architectures, we show that the NPE's compositional representation of the structure in physical interactions improves its ability to predict movement, generalize across variable object count and different scene configurations, and infer latent properties of objects such as mass.", "Wooden blocks are a common toy for infants, allowing them to develop motor skills and gain intuition about the physical behavior of the world. In this paper, we explore the ability of deep feed-forward models to learn such intuitive physics. Using a 3D game engine, we create small towers of wooden blocks whose stability is randomized and render them collapsing (or remaining upright). This data allows us to train large convolutional network models which can accurately predict the outcome, as well as estimating the block trajectories. The models are also able to generalize in two important ways: (i) to new physical scenarios, e.g. towers with an additional block and (ii) to images of real wooden blocks, where it obtains a performance comparable to human subjects.", "", "", "Evolution has resulted in highly developed abilities in many natural intelligences to quickly and accurately predict mechanical phenomena. Humans have successfully developed laws of physics to abstract and model such mechanical phenomena. In the context of artificial intelligence, a recent line of work has focused on estimating physical parameters based on sensory data and use them in physical simulators to make long-term predictions. In contrast, we investigate the effectiveness of a single neural network for end-to-end long-term prediction of mechanical phenomena. Based on extensive evaluation, we demonstrate that such networks can outperform alternate approaches having even access to ground-truth physical simulators, especially when some physical parameters are unobserved or not known a-priori. Further, our network outputs a distribution of outcomes to capture the inherent uncertainty in the data. Our approach demonstrates for the first time the possibility of making actionable long-term predictions from sensor data without requiring to explicitly model the underlying physical laws.", "We explore the hypothesis that many intuitive physical inferences are based on a mental physics engine that is analogous in many ways to the machine physics engines used in building interactive video games. We describe the key features of game physics engines and their parallels in human mental representation, focusing especially on the intuitive physics of young infants where the hypothesis helps to unify many classic and otherwise puzzling phenomena, and may provide the basis for a computational account of how the physical knowledge of infants develops. This hypothesis also explains several ‘physics illusions', and helps to inform the development of artificial intelligence (AI) systems with more human-like common sense.", "Common-sense physical reasoning is an essential ingredient for any intelligent agent operating in the real-world. For example, it can be used to simulate the environment, or to infer the state of parts of the world that are currently unobserved. In order to match real-world conditions this causal knowledge must be learned without access to supervised data. To address this problem we present a novel method that learns to discover objects, and models their physical interactions from raw visual images in a purely unsupervised fashion. It incorporates prior knowledge about the compositional nature of human perception to factor interactions between object-pairs and learn efficiently. On videos of bouncing balls we show the superior modelling capabilities of our method compared to other unsupervised neural approaches that do not incorporate such prior knowledge. We demonstrate its ability to handle occlusion and show that it can extrapolate learned knowledge to environments with varying numbers of objects.", "Reasoning about objects, relations, and physics is central to human intelligence, and a key goal of artificial intelligence. Here we introduce the interaction network, a model which can reason about how objects in complex systems interact, supporting dynamical predictions, as well as inferences about the abstract properties of the system. Our model takes graphs as input, performs object- and relation-centric reasoning in a way that is analogous to a simulation, and is implemented using deep neural networks. We evaluate its ability to reason about several challenging physical domains: n-body problems, rigid-body collision, and non-rigid dynamics. Our results show it can be trained to accurately simulate the physical trajectories of dozens of objects over thousands of time steps, estimate abstract quantities such as energy, and generalize automatically to systems with different numbers and configurations of objects and relations. Our interaction network implementation is the first general-purpose, learnable physics engine, and a powerful general framework for reasoning about object and relations in a wide variety of complex real-world domains.", "We introduce a neural network architecture that models the physical reasoning process and that can be used to extract simple physical concepts from experimental data without being provided with additional prior knowledge. We apply the neural network to a variety of simple physical examples in classical and quantum mechanics, like damped pendulums, two-particle collisions, and qubits. The network finds the physically relevant parameters, exploits conservation laws to make predictions, and can be used to gain conceptual insights. For example, given a time series of the positions of the Sun and Mars as observed from Earth, the network discovers the heliocentric model of the solar system - that is, it encodes the data into the angles of the two planets as seen from the Sun. Our work provides a first step towards answering the question whether the traditional ways by which physicists model nature naturally arise from the experimental data without any mathematical and physical pre-knowledge, or if there are alternative elegant formalisms, which may solve some of the fundamental conceptual problems in modern physics, such as the measurement problem in quantum mechanics.", "In a glance, we can perceive whether a stack of dishes will topple, a branch will support a child’s weight, a grocery bag is poorly packed and liable to tear or crush its contents, or a tool is firmly attached to a table or free to be lifted. Such rapid physical inferences are central to how people interact with the world and with each other, yet their computational underpinnings are poorly understood. We propose a model based on an “intuitive physics engine,” a cognitive mechanism similar to computer engines that simulate rich physics in video games and graphics, but that uses approximate, probabilistic simulations to make robust and fast inferences in complex natural scenes where crucial information is unobserved. This single model fits data from five distinct psychophysical tasks, captures several illusions and biases, and explains core aspects of human mental models and common-sense reasoning that are instrumental to how humans understand their everyday world." ] }
1905.09891	2944839214	Many current methods to learn intuitive physics are based on interaction networks and similar approaches. However, they rely on information that has proven difficult to estimate directly from image data in the past. We aim to narrow this gap by inferring all the semantic information needed from raw pixel data in the form of a scene-graph. Our approach is based on neural-symbolic capsules, which identify which objects in the scene are static, dynamic, elastic or rigid, possible joints between them, as well as their collision information. By integrating all this with interaction networks, we demonstrate how our method is able to learn intuitive physics directly from image sequences and apply its knowledge to new scenes and objects, resulting in an inverse-simulation pipeline.	Since the introduction of interaction networks, there has been a lot of research into extending and refining these methods, as well as integrating them with other models. One such extension has been to automatically discover object relations in a scene @cite_12 . An alternative approach has been to connect interaction networks directly with multilayer perceptrons. Watters al @cite_3 showed that by adding a front-end in the form of a convolutional neural network, intuitive physics can be learned for rigid bodies directly from video data. There has also been research into integrating unsupervised methods @cite_18 , integrating graph networks @cite_10 , optimizing the flow of information to and from interaction networks @cite_13 and using interaction networks in planning systems @cite_2 .	{ "cite_N": [ "@cite_18", "@cite_10", "@cite_3", "@cite_2", "@cite_13", "@cite_12" ], "mid": [ "2809124645", "2804072623", "2770604561", "2738675347", "2963942401", "2590960023" ], "abstract": [ "We propose a framework for the completely unsupervised learning of latent object properties from their interactions: the perception-prediction network (PPN). Consisting of a perception module that extracts representations of latent object properties and a prediction module that uses those extracted properties to simulate system dynamics, the PPN can be trained in an end-to-end fashion purely from samples of object dynamics. The representations of latent object properties learned by PPNs not only are sufficient to accurately simulate the dynamics of systems comprised of previously unseen objects, but also can be translated directly into human-interpretable properties (e.g., mass, coefficient of restitution) in an entirely unsupervised manner. Crucially, PPNs also generalize to novel scenarios: their gradient-based training can be applied to many dynamical systems and their graph-based structure functions over systems comprised of different numbers of objects. Our results demonstrate the efficacy of graph-based neural architectures in object-centric inference and prediction tasks, and our model has the potential to discover relevant object properties in systems that are not yet well understood.", "Understanding and interacting with everyday physical scenes requires rich knowledge about the structure of the world, represented either implicitly in a value or policy function, or explicitly in a transition model. Here we introduce a new class of learnable models--based on graph networks--which implement an inductive bias for object- and relation-centric representations of complex, dynamical systems. Our results show that as a forward model, our approach supports accurate predictions from real and simulated data, and surprisingly strong and efficient generalization, across eight distinct physical systems which we varied parametrically and structurally. We also found that our inference model can perform system identification. Our models are also differentiable, and support online planning via gradient-based trajectory optimization, as well as offline policy optimization. Our framework offers new opportunities for harnessing and exploiting rich knowledge about the world, and takes a key step toward building machines with more human-like representations of the world.", "From just a glance, humans can make rich predictions about the future of a wide range of physical systems. On the other hand, modern approaches from engineering, robotics, and graphics are often restricted to narrow domains or require information about the underlying state. We introduce the Visual Interaction Network, a general-purpose model for learning the dynamics of a physical system from raw visual observations. Our model consists of a perceptual front-end based on convolutional neural networks and a dynamics predictor based on interaction networks. Through joint training, the perceptual front-end learns to parse a dynamic visual scene into a set of factored latent object representations. The dynamics predictor learns to roll these states forward in time by computing their interactions, producing a predicted physical trajectory of arbitrary length. We found that from just six input video frames the Visual Interaction Network can generate accurate future trajectories of hundreds of time steps on a wide range of physical systems. Our model can also be applied to scenes with invisible objects, inferring their future states from their effects on the visible objects, and can implicitly infer the unknown mass of objects. This work opens new opportunities for model-based decision-making and planning from raw sensory observations in complex physical environments.", "Conventional wisdom holds that model-based planning is a powerful approach to sequential decision-making. It is often very challenging in practice, however, because while a model can be used to evaluate a plan, it does not prescribe how to construct a plan. Here we introduce the \"Imagination-based Planner\", the first model-based, sequential decision-making agent that can learn to construct, evaluate, and execute plans. Before any action, it can perform a variable number of imagination steps, which involve proposing an imagined action and evaluating it with its model-based imagination. All imagined actions and outcomes are aggregated, iteratively, into a \"plan context\" which conditions future real and imagined actions. The agent can even decide how to imagine: testing out alternative imagined actions, chaining sequences of actions together, or building a more complex \"imagination tree\" by navigating flexibly among the previously imagined states using a learned policy. And our agent can learn to plan economically, jointly optimizing for external rewards and computational costs associated with using its imagination. We show that our architecture can learn to solve a challenging continuous control problem, and also learn elaborate planning strategies in a discrete maze-solving task. Our work opens a new direction toward learning the components of a model-based planning system and how to use them.", "Many machine learning systems are built to solve the hardest examples of a particular task, which often makes them large and expensive to run---especially with respect to the easier examples, which might require much less computation. For an agent with a limited computational budget, this \"one-size-fits-all\" approach may result in the agent wasting valuable computation on easy examples, while not spending enough on hard examples. Rather than learning a single, fixed policy for solving all instances of a task, we introduce a metacontroller which learns to optimize a sequence of \"imagined\" internal simulations over predictive models of the world in order to construct a more informed, and more economical, solution. The metacontroller component is a model-free reinforcement learning agent, which decides both how many iterations of the optimization procedure to run, as well as which model to consult on each iteration. The models (which we call \"experts\") can be state transition models, action-value functions, or any other mechanism that provides information useful for solving the task, and can be learned on-policy or off-policy in parallel with the metacontroller. When the metacontroller, controller, and experts were trained with \"interaction networks\" (, 2016) as expert models, our approach was able to solve a challenging decision-making problem under complex non-linear dynamics. The metacontroller learned to adapt the amount of computation it performed to the difficulty of the task, and learned how to choose which experts to consult by factoring in both their reliability and individual computational resource costs. This allowed the metacontroller to achieve a lower overall cost (task loss plus computational cost) than more traditional fixed policy approaches. These results demonstrate that our approach is a powerful framework for using rich forward models for efficient model-based reinforcement learning.", "Our world can be succinctly and compactly described as structured scenes of objects and relations. A typical room, for example, contains salient objects such as tables, chairs and books, and these objects typically relate to each other by virtue of their correlated features, such as position, function and shape. Humans exploit knowledge of objects and their relations for learning a wide spectrum of tasks, and more generally when learning the structure underlying observed data. In this work, we introduce relation networks (RNs) - a general purpose neural network architecture for object-relation reasoning. We show that RNs are capable of learning object relations from scene description data. Furthermore, we show that RNs can act as a bottleneck that induces the factorization of objects from entangled scene description inputs, and from distributed deep representations of scene images provided by a variational autoencoder. The model can also be used in conjunction with differentiable memory mechanisms for implicit relation discovery in one-shot learning tasks. Our results suggest that relation networks are a powerful architecture for solving a variety of problems that require object relation reasoning." ] }
1905.10009	2946763595	Self-explaining models are models that reveal decision making parameters in an interpretable manner so that the model reasoning process can be directly understood by human beings. General Linear Models (GLMs) are self-explaining because the model weights directly show how each feature contribute to the output value. However, deep neural networks (DNNs) are in general not self-explaining due to the non-linearity of the activation functions, complex architectures, obscure feature extraction and transformation process. In this work, we illustrate the fact that existing deep architectures are hard to interpret because each hidden layer carries a mix of high level features and low level features. As a solution, we propose a novel feature leveling architecture that isolates low level features from high level features on a per-layer basis to better utilize the GLM layer in the proposed architecture for interpretation. Experimental results show that our modified models are able to achieve competitive results comparing to main-stream architectures on standard datasets while being more self-explaining. Our implementations and configurations are publicly available for reproductions.	The ability to explain the reasoning process within a neural network is essential to validate the robustness of the model and to ensure that the network is secure against adversarial attacks @cite_29 @cite_3 @cite_16 . In recent years, Many works have been done to explain the reasoning process of an existing neural network either through extracting the decision boundary @cite_6 @cite_19 @cite_24 @cite_0 , or through a variety of visualization methods @cite_13 @cite_12 @cite_26 . Most of those methods are designed for validation purpose. However, their results cannot be easily used to improve the original models.	{ "cite_N": [ "@cite_26", "@cite_29", "@cite_6", "@cite_3", "@cite_24", "@cite_19", "@cite_0", "@cite_16", "@cite_13", "@cite_12" ], "mid": [ "1601924930", "2243397390", "2964231903", "", "2890660842", "2796284132", "1937344979", "2794609696", "1915485278", "1849277567" ], "abstract": [ "While neural networks have been successfully applied to many NLP tasks the resulting vector-based models are very difficult to interpret. For example it's not clear how they achieve compositionality , building sentence meaning from the meanings of words and phrases. In this paper we describe four strategies for visualizing compositionality in neural models for NLP, inspired by similar work in computer vision. We first plot unit values to visualize compositionality of negation, intensification, and concessive clauses, allow us to see well-known markedness asymmetries in negation. We then introduce three simple and straightforward methods for visualizing a unit's salience , the amount it contributes to the final composed meaning: (1) gradient back-propagation, (2) the variance of a token from the average word node, (3) LSTM-style gates that measure information flow. We test our methods on sentiment using simple recurrent nets and LSTMs. Our general-purpose methods may have wide applications for understanding compositionality and other semantic properties of deep networks , and also shed light on why LSTMs outperform simple recurrent nets,", "State-of-the-art deep neural networks have achieved impressive results on many image classification tasks. However, these same architectures have been shown to be unstable to small, well sought, perturbations of the images. Despite the importance of this phenomenon, no effective methods have been proposed to accurately compute the robustness of state-of-the-art deep classifiers to such perturbations on large-scale datasets. In this paper, we fill this gap and propose the DeepFool algorithm to efficiently compute perturbations that fool deep networks, and thus reliably quantify the robustness of these classifiers. Extensive experimental results show that our approach outperforms recent methods in the task of computing adversarial perturbations and making classifiers more robust.1", "While deep reinforcement learning has successfully solved many challenging control tasks, its real-world applicability has been limited by the inability to ensure the safety of learned policies. We propose an approach to verifiable reinforcement learning by training decision tree policies, which can represent complex policies (since they are nonparametric), yet can be efficiently verified using existing techniques (since they are highly structured). The challenge is that decision tree policies are difficult to train. We propose VIPER, an algorithm that combines ideas from model compression and imitation learning to learn decision tree policies guided by a DNN policy (called the oracle) and its Q-function, and show that it substantially outperforms two baselines. We use VIPER to (i) learn a provably robust decision tree policy for a variant of Atari Pong with a symbolic state space, (ii) learn a decision tree policy for a toy game based on Pong that provably never loses, and (iii) learn a provably stable decision tree policy for cart-pole. In each case, the decision tree policy achieves performance equal to that of the original DNN policy.", "", "Neural networks are increasingly deployed in real-world safety-critical domains such as autonomous driving, aircraft collision avoidance, and malware detection. However, these networks have been shown to often mispredict on inputs with minor adversarial or even accidental perturbations. Consequences of such errors can be disastrous and even potentially fatal as shown by the recent Tesla autopilot crash. Thus, there is an urgent need for formal analysis systems that can rigorously check neural networks for violations of different safety properties such as robustness against adversarial perturbations within a certain L-norm of a given image. An effective safety analysis system for a neural network must be able to either ensure that a safety property is satisfied by the network or find a counterexample, i.e., an input for which the network will violate the property. Unfortunately, most existing techniques for performing such analysis struggle to scale beyond very small networks and the ones that can scale to larger networks suffer from high false positives and cannot produce concrete counterexamples in case of a property violation. In this paper, we present a new efficient approach for rigorously checking different safety properties of neural networks that significantly outperforms existing approaches by multiple orders of magnitude. Our approach can check different safety properties and find concrete counterexamples for networks that are 10x larger than the ones supported by existing analysis techniques. We believe that our approach to estimating tight output bounds of a network for a given input range can also help improve the explainability of neural networks and guide the training process of more robust neural networks.", "We study the problem of generating interpretable and verifiable policies through reinforcement learning. Unlike the popular Deep Reinforcement Learning (DRL) paradigm, in which the policy is represented by a neural network, the aim in Programmatically Interpretable Reinforcement Learning is to find a policy that can be represented in a high-level programming language. Such programmatic policies have the benefits of being more easily interpreted than neural networks, and being amenable to verification by symbolic methods. We propose a new method, called Neurally Directed Program Search (NDPS), for solving the challenging nonsmooth optimization problem of finding a programmatic policy with maxima reward. NDPS works by first learning a neural policy network using DRL, and then performing a local search over programmatic policies that seeks to minimize a distance from this neural \"oracle\". We evaluate NDPS on the task of learning to drive a simulated car in the TORCS car-racing environment. We demonstrate that NDPS is able to discover human-readable policies that pass some significant performance bars. We also find that a well-designed policy language can serve as a regularizer, and result in the discovery of policies that lead to smoother trajectories and are more easily transferred to environments not encountered during training.", "In safety-critical applications it is necessary to verify if a neural net does not display any undesirable behavior due to overtraining. In particular in civil aviation, novel algorithms undergo the utmost scrutiny before the flight software is approved for use in aircraft. Neural nets' reputation for unpredictability has impeded their use in safety-critical applications. The paper presents a deterministic method for verification of a neural net trained to approximate another, difficult to implement mapping. First, approximation error is evaluated on a uniform grid of testing points. Then, maximal growth rates of both functions are used to bound the error anywhere between the testing points. The technique allows us to verify accuracy of nets that replace large multidimensional look-up tables. Practical ramifications of the method and further possible extensions are discussed.", "We present AI2, the first sound and scalable analyzer for deep neural networks. Based on overapproximation, AI2 can automatically prove safety properties (e.g., robustness) of realistic neural networks (e.g., convolutional neural networks). The key insight behind AI2 is to phrase reasoning about safety and robustness of neural networks in terms of classic abstract interpretation, enabling us to leverage decades of advances in that area. Concretely, we introduce abstract transformers that capture the behavior of fully connected and convolutional neural network layers with rectified linear unit activations (ReLU), as well as max pooling layers. This allows us to handle real-world neural networks, which are often built out of those types of layers. We present a complete implementation of AI2 together with an extensive evaluation on 20 neural networks. Our results demonstrate that: (i) AI2 is precise enough to prove useful specifications (e.g., robustness), (ii) AI2 can be used to certify the effectiveness of state-of-the-art defenses for neural networks, (iii) AI2 is significantly faster than existing analyzers based on symbolic analysis, which often take hours to verify simple fully connected networks, and (iv) AI2 can handle deep convolutional networks, which are beyond the reach of existing methods.", "Image representations, from SIFT and Bag of Visual Words to Convolutional Neural Networks (CNNs), are a crucial component of almost any image understanding system. Nevertheless, our understanding of them remains limited. In this paper we conduct a direct analysis of the visual information contained in representations by asking the following question: given an encoding of an image, to which extent is it possible to reconstruct the image itself? To answer this question we contribute a general framework to invert representations. We show that this method can invert representations such as HOG more accurately than recent alternatives while being applicable to CNNs too. We then use this technique to study the inverse of recent state-of-the-art CNN image representations for the first time. Among our findings, we show that several layers in CNNs retain photographically accurate information about the image, with different degrees of geometric and photometric invariance.", "Large Convolutional Network models have recently demonstrated impressive classification performance on the ImageNet benchmark [18]. However there is no clear understanding of why they perform so well, or how they might be improved. In this paper we explore both issues. We introduce a novel visualization technique that gives insight into the function of intermediate feature layers and the operation of the classifier. Used in a diagnostic role, these visualizations allow us to find model architectures that outperform on the ImageNet classification benchmark. We also perform an ablation study to discover the performance contribution from different model layers. We show our ImageNet model generalizes well to other datasets: when the softmax classifier is retrained, it convincingly beats the current state-of-the-art results on Caltech-101 and Caltech-256 datasets." ] }
1905.10009	2946763595	Self-explaining models are models that reveal decision making parameters in an interpretable manner so that the model reasoning process can be directly understood by human beings. General Linear Models (GLMs) are self-explaining because the model weights directly show how each feature contribute to the output value. However, deep neural networks (DNNs) are in general not self-explaining due to the non-linearity of the activation functions, complex architectures, obscure feature extraction and transformation process. In this work, we illustrate the fact that existing deep architectures are hard to interpret because each hidden layer carries a mix of high level features and low level features. As a solution, we propose a novel feature leveling architecture that isolates low level features from high level features on a per-layer basis to better utilize the GLM layer in the proposed architecture for interpretation. Experimental results show that our modified models are able to achieve competitive results comparing to main-stream architectures on standard datasets while being more self-explaining. Our implementations and configurations are publicly available for reproductions.	refers to various methods developed to reduce the number of parameters of a neural model. Many investigations have been done in using @math or @math @cite_9 @cite_27 @cite_21 @cite_5 regularization to prune neural network while maintaining differentiability for back propagation. Another choice for regularization and creating sparsity is the @math regularization. However, due to its discrete nature, it does not support parameter learning through backpropagation. A continuous approximation of @math is proposed in regard to resolve this problem and has shown effectiveness in pruning both FCNN and Convolutional Neural Networks (CNNs) in an end to end manner @cite_4 . This regularization technique is further applied not only to neural architecture pruning but to feature selections @cite_15 . Our work applies the @math regularization's feature selection ability in a novel context to select maximum amount of features as direct inputs for the GLM layer.	{ "cite_N": [ "@cite_4", "@cite_9", "@cite_21", "@cite_27", "@cite_5", "@cite_15" ], "mid": [ "2963828549", "2963674932", "", "", "1981025032", "2895929976" ], "abstract": [ "We propose a practical method for @math norm regularization for neural networks: pruning the network during training by encouraging weights to become exactly zero. Such regularization is interesting since (1) it can greatly speed up training and inference, and (2) it can improve generalization. AIC and BIC, well-known model selection criteria, are special cases of @math regularization. However, since the @math norm of weights is non-differentiable, we cannot incorporate it directly as a regularization term in the objective function. We propose a solution through the inclusion of a collection of non-negative stochastic gates, which collectively determine which weights to set to zero. We show that, somewhat surprisingly, for certain distributions over the gates, the expected @math regularized objective is differentiable with respect to the distribution parameters. We further propose the distribution for the gates, which is obtained by stretching'' a binary concrete distribution and then transforming its samples with a hard-sigmoid. The parameters of the distribution over the gates can then be jointly optimized with the original network parameters. As a result our method allows for straightforward and efficient learning of model structures with stochastic gradient descent and allows for conditional computation in a principled way. We perform various experiments to demonstrate the effectiveness of the resulting approach and regularizer.", "Neural networks are both computationally intensive and memory intensive, making them difficult to deploy on embedded systems. Also, conventional networks fix the architecture before training starts; as a result, training cannot improve the architecture. To address these limitations, we describe a method to reduce the storage and computation required by neural networks by an order of magnitude without affecting their accuracy by learning only the important connections. Our method prunes redundant connections using a three-step method. First, we train the network to learn which connections are important. Next, we prune the unimportant connections. Finally, we retrain the network to fine tune the weights of the remaining connections. On the ImageNet dataset, our method reduced the number of parameters of AlexNet by a factor of 9x, from 61 million to 6.7 million, without incurring accuracy loss. Similar experiments with VGG-16 found that the total number of parameters can be reduced by 13x, from 138 million to 10.3 million, again with no loss of accuracy.", "", "", "We had previously shown that regularization principles lead to approximation schemes that are equivalent to networks with one layer of hidden units, called regularization networks . In particular, standard smoothness functionals lead to a subclass of regularization networks, the well known radial basis functions approximation schemes. This paper shows that regularization networks encompass a much broader range of approximation schemes, including many of the popular general additive models and some of the neural networks. In particular, we introduce new classes of smoothness functionals that lead to different classes of basis functions. Additive splines as well as some tensor product splines can be obtained from appropriate classes of smoothness functionals. Furthermore, the same generalization that extends radial basis functions (RBF) to hyper basis functions (HBF) also leads from additive models to ridge approximation models, containing as special cases Breiman's hinge functions, some forms of projection pursuit regression, and several types of neural networks. We propose to use the term generalized regularization networks for this broad class of approximation schemes that follow from an extension of regularization. In the probabilistic interpretation of regularization, the different classes of basis functions correspond to different classes of prior probabilities on the approximating function spaces, and therefore to different types of smoothness assumptions. In summary, different multilayer networks with one hidden layer, which we collectively call generalized regularization networks, correspond to different classes of priors and associated smoothness functionals in a classical regularization principle. Three broad classes are (1) radial basis functions that can be generalized to hyper basis functions, (2) some tensor product splines, and (3) additive splines that can be generalized to schemes of the type of ridge approximation, hinge functions, and several perceptron-like neural networks with one hidden layer.", "In this study, we propose a novel non-parametric embedded feature selection method based on minimizing the @math norm of the vector of an indicator variable, whose point-wise product of an input selects a subset of features. Our approach relies on the continuous relaxation of Bernoulli distributions, which allows our model to learn the parameters of the approximate Bernoulli distributions via tractable methods. Using these tools we present a general neural network that simultaneously minimizes a loss function while selecting relevant features. We also provide an information-theoretic justification of incorporating Bernoulli distribution into our approach. Finally, we demonstrate the potential of the approach on synthetic and real-life applications." ] }
1905.09856	2946008212	Attention is an operation that selects some largest element from some set, where the notion of largest is defined elsewhere. Applying this operation to sequence to sequence mapping results in significant improvements to the task at hand. In this paper we provide the mathematical definition of attention and examine its application to sequence to sequence models. We highlight the exact correspondences between machine learning implementations of attention and our mathematical definition. We provide clear evidence of effectiveness of attention mechanisms evaluating models with varying degrees of attention on a very simple task: copying a sentence. We find that models that make greater use of attention perform much better on sequence to sequence mapping tasks, converge faster and are more stable.	The authors in @cite_11 compare three different variants of encoder-decoder models: two attention-based models and a character sequence to sequence model. They use data from Twitter Typo Curpose and Combilex data set which contains English graphemes to phonetic representation mappings. They compare these models against specialized monotone seq2seq models, and find them to be comparable in terms of performance. However, they did not look at the transformer or CNN architectures, nor did they use the same sentence input-output scheme that we propose here. In @cite_13 authors propose taxonomy of attention where they classify different models based on number of sequences, number of abstraction levels, number of positions and number of representations. They also point out the distinct network architectures that use attention. These include: encoder-decoder, memory networks, networks with RNNs. In @cite_3 authors group attention models based on tasks addressed'' classification. These include things like document classification, machine translation, morphology, information extraction, syntax etc. A unified attention model'' is introduced which is similar to Transformer model. The authors propose a taxonomy of attention models based on input representation, compatibility functions, distribution functions, multiplicity.	{ "cite_N": [ "@cite_13", "@cite_3", "@cite_11" ], "mid": [ "2941531368", "2912279724", "2542740526" ], "abstract": [ "Attention Model has now become an important concept in neural networks that has been researched within diverse application domains. This survey provides a structured and comprehensive overview of the developments in modeling attention. In particular, we propose a taxonomy which groups existing techniques into coherent categories. We review the different neural architectures in which attention has been incorporated, and also show how attention improves interpretability of neural models. Finally, we discuss some applications in which modeling attention has a significant impact. We hope this survey will provide a succinct introduction to attention models and guide practitioners while developing approaches for their applications.", "Attention is an increasingly popular mechanism used in a wide range of neural architectures. Because of the fast-paced advances in this domain, a systematic overview of attention is still missing. In this article, we define a unified model for attention architectures for natural language processing, with a focus on architectures designed to work with vector representation of the textual data. We discuss the dimensions along which proposals differ, the possible uses of attention, and chart the major research activities and open challenges in the area.", "We analyze the performance of encoder-decoder neural models and compare them with well-known established methods. The latter represent different classes of traditional approaches that are applied to the monotone sequence-to-sequence tasks OCR post-correction, spelling correction, grapheme-to-phoneme conversion, and lemmatization. Such tasks are of practical relevance for various higher-level research fields including digital humanities, automatic text correction, and speech recognition. We investigate how well generic deep-learning approaches adapt to these tasks, and how they perform in comparison with established and more specialized methods, including our own adaptation of pruned CRFs." ] }
1905.10029	2946234452	Graph convolutional networks (GCNs) are powerful tools for graph-structured data. However, they have been recently shown to be prone to topological attacks. Despite substantial efforts to search for new architectures, it still remains a challenge to improve performance in both benign and adversarial situations simultaneously. In this paper, we re-examine the fundamental building block of GCN---the Laplacian operator---and highlight some basic flaws in the spatial and spectral domains. As an alternative, we propose an operator based on graph powering, and prove that it enjoys a desirable property of "spectral separation." Based on the operator, we propose a robust learning paradigm, where the network is trained on a family of "'smoothed" graphs that span a spatial and spectral range for generalizability. We also use the new operator in replacement of the classical Laplacian to construct an architecture with improved spectral robustness, expressivity and interpretability. The enhanced performance and robustness are demonstrated in extensive experiments.	One limitation of using Laplacian in GCN is that the first layer only sees'' features of adjacent nodes. Several works have been proposed to address this issue @cite_11 @cite_5 @cite_15 . For instance, adjacency matrix and its powered versions are used in @cite_11 @cite_5 @cite_15 to construct GCN layers. However, simply powering the adjacency does not extract spectral gap and may even make the eigenspectrum more sensitive to perturbations. The architecture in @cite_15 also introduced additional weight matrices for neighbors at different distances. But this could substantially increase the risk of overfitting in the low-data regime and make the network vulnerable to adversarial attacks.	{ "cite_N": [ "@cite_5", "@cite_15", "@cite_11" ], "mid": [ "2789186812", "2942681667", "2891781872" ], "abstract": [ "Graph Convolutional Networks (GCNs) have shown significant improvements in semi-supervised learning on graph-structured data. Concurrently, unsupervised learning of graph embeddings has benefited from the information contained in random walks. In this paper, we propose a model: Network of GCNs (N-GCN), which marries these two lines of work. At its core, N-GCN trains multiple instances of GCNs over node pairs discovered at different distances in random walks, and learns a combination of the instance outputs which optimizes the classification objective. Our experiments show that our proposed N-GCN model improves state-of-the-art baselines on all of the challenging node classification tasks we consider: Cora, Citeseer, Pubmed, and PPI. In addition, our proposed method has other desirable properties, including generalization to recently proposed semi-supervised learning methods such as GraphSAGE, allowing us to propose N-SAGE, and resilience to adversarial input perturbations.", "Existing popular methods for semi-supervised learning with Graph Neural Networks (such as the Graph Convolutional Network) provably cannot learn a general class of neighborhood mixing relationships. To address this weakness, we propose a new model, MixHop, that can learn these relationships, including difference operators, by repeatedly mixing feature representations of neighbors at various distances. Mixhop requires no additional memory or computational complexity, and outperforms on challenging baselines. In addition, we propose sparsity regularization that allows us to visualize how the network prioritizes neighborhood information across different graph datasets. Our analysis of the learned architectures reveals that neighborhood mixing varies per datasets.", "Following the success of deep convolutional networks in various vision and speech related tasks, researchers have started investigating generalizations of the well-known technique for graph-structured data. A recently-proposed method called Graph Convolutional Networks has been able to achieve state-of-the-art results in the task of node classification. However, since the proposed method relies on localized first-order approximations of spectral graph convolutions, it is unable to capture higher-order interactions between nodes in the graph. In this work, we propose a motif-based graph attention model, called Motif Convolutional Networks (MCNs), which generalizes past approaches by using weighted multi-hop motif adjacency matrices to capture higher-order neighborhoods. A novel attention mechanism is used to allow each individual node to select the most relevant neighborhood to apply its filter. Experiments show that our proposed method is able to achieve state-of-the-art results on the semi-supervised node classification task." ] }
1905.10029	2946234452	Graph convolutional networks (GCNs) are powerful tools for graph-structured data. However, they have been recently shown to be prone to topological attacks. Despite substantial efforts to search for new architectures, it still remains a challenge to improve performance in both benign and adversarial situations simultaneously. In this paper, we re-examine the fundamental building block of GCN---the Laplacian operator---and highlight some basic flaws in the spatial and spectral domains. As an alternative, we propose an operator based on graph powering, and prove that it enjoys a desirable property of "spectral separation." Based on the operator, we propose a robust learning paradigm, where the network is trained on a family of "'smoothed" graphs that span a spatial and spectral range for generalizability. We also use the new operator in replacement of the classical Laplacian to construct an architecture with improved spectral robustness, expressivity and interpretability. The enhanced performance and robustness are demonstrated in extensive experiments.	The robustness of spectral methods has been extensively studied for graph partitioning clustering using adversarial perturbations or noise perturbations @cite_53 @cite_2 @cite_1 . A common approach is to use regularization on the Laplacian @cite_54 @cite_24 @cite_16 . Most recently, operators based on self-avoiding or nonbacktracking walks have become popular for SBM @cite_51 @cite_30 @cite_28 , which have been proved to achieve the theoretical detection threshold conjectured in @cite_27 . Our work is partly motivated by the graph powering approach @cite_32 , which leveraged the result of @cite_51 @cite_28 to show a spectral gap of the powering operator. The main difference with this line of work is that these operators are studied only for spectral clustering algorithms, which does not incorporate nodal features. Our proposed variable power operator can be viewed as a kernel version of the graph powering operator @cite_32 , thus substantially increasing the capability of learning complex nodal interactions while maintaining the nice spectral property.	{ "cite_N": [ "@cite_30", "@cite_28", "@cite_53", "@cite_54", "@cite_1", "@cite_32", "@cite_24", "@cite_27", "@cite_2", "@cite_16", "@cite_51" ], "mid": [ "1914749871", "2790609319", "2146670837", "857387926", "2942689850", "2889975307", "1973207991", "2004531067", "2103759851", "2464534118", "2023348178" ], "abstract": [ "We study a random graph model named the \"block model\" in statistics and the \"planted partition model\" in theoretical computer science. In its simplest form, this is a random graph with two equal-sized clusters, with a between-class edge probability of @math and a within-class edge probability of @math . A striking conjecture of Decelle, Krzkala, Moore and Zdeborov 'a based on deep, non-rigorous ideas from statistical physics, gave a precise prediction for the algorithmic threshold of clustering in the sparse planted partition model. In particular, if @math and @math , @math and @math then conjectured that it is possible to efficiently cluster in a way correlated with the true partition if @math and impossible if @math for some sufficiently large @math . In a previous work, we proved that indeed it is information theoretically impossible to to cluster if @math . A different independent proof of the same result was recently obtained by Laurent Massoulie.", "A nonbacktracking walk on a graph is a directed path such that no edge is the inverse of its preceding edge. The nonbacktracking matrix of a graph is indexed by its directed edges and can be used to count nonbacktracking walks of a given length. It has been used recently in the context of community detection and has appeared previously in connection with the Ihara zeta function and in some generalizations of Ramanujan graphs. In this work, we study the largest eigenvalues of the nonbacktracking matrix of the Erdős–Renyi random graph and of the stochastic block model in the regime where the number of edges is proportional to the number of vertices. Our results confirm the “spectral redemption conjecture” in the symmetric case and show that community detection can be made on the basis of the leading eigenvectors above the feasibility threshold.", "This paper aims to introduce the robustness against noise into the spectral clustering algorithm. First, we propose a warping model to map the data into a new space on the basis of regularization. During the warping, each point spreads smoothly its spatial information to other points. After the warping, empirical studies show that the clusters become relatively compact and well separated, including the noise cluster that is formed by the noise points. In this new space, the number of clusters can be estimated by eigenvalue analysis. We further apply the spectral mapping to the data to obtain a low-dimensional data representation. Finally, the K-means algorithm is used to perform clustering. The proposed method is superior to previous spectral clustering methods in that (i) it is robust against noise because the noise points are grouped into one new cluster; (ii) the number of clusters and the parameters of the algorithm are determined automatically. Experimental results on synthetic and real data have demonstrated this superiority.", "In this paper, we examine a spectral clustering algorithm for similarity graphs drawn from a simple random graph model, where nodes are allowed to have varying degrees, and we provide theoretical bounds on its performance. The random graph model we study is the Extended Planted Partition (EPP) model, a variant of the classical planted partition model. The standard approach to spectral clustering of graphs is to compute the bottom k singular vectors or eigenvectors of a suitable graph Laplacian, project the nodes of the graph onto these vectors, and then use an iterative clustering algorithm on the projected nodes. However a challenge with applying this approach to graphs generated from the EPP model is that unnormalized Laplacians do not work, and normalized Laplacians do not concentrate well when the graph has a number of low degree nodes. We resolve this issue by introducing the notion of a degree-corrected graph Laplacian. For graphs with many low degree nodes, degree correction has a regularizing eect on the Laplacian. Our spectral clustering algorithm projects the nodes in the graph onto the bottom k right singular vectors of the degree-corrected random-walk Laplacian, and clusters the nodes in this subspace. We show guarantees on the performance of this algorithm, demonstrating that it outputs the correct partition under a wide range of parameter values. Unlike some previous work, our algorithm does not require access to any generative parameters of the model.", "We study high-dimensional distribution learning in an agnostic setting where an adversary is allowed to arbitrarily corrupt an @math -fraction of the samples. Such questions have a rich history spanning statistics, machine learning, and theoretical computer science. Even in the most basic settings, the only known approaches are either computationally inefficient or lose dimension-dependent factors in their error guarantees. This raises the following question: Is high-dimensional agnostic distribution learning even possible, algorithmically? In this work, we obtain the first computationally efficient algorithms with dimension-independent error guarantees for agnostically learning several fundamental classes of high-dimensional distributions: (1) a single Gaussian, (2) a product distribution on the hypercube, (3) mixtures of two product distributions (under a natural balancedness condition), and (4) mixtures of spherical Gaussians. Our algorithms achieve error that is independent of the dimension, an...", "Spectral algorithms, such as principal component analysis and spectral clustering, typically require careful data transformations to be effective: upon observing a matrix @math , one may look at the spectrum of @math for a properly chosen @math . The issue is that the spectrum of @math might be contaminated by non-informational top eigenvalues, e.g., due to scale variations in the data, and the application of @math aims to remove these. Designing a good functional @math (and establishing what good means) is often challenging and model dependent. This paper proposes a simple and generic construction for sparse graphs, @math where @math denotes the adjacency matrix and @math is an integer (less than the graph diameter). This produces a graph connecting vertices from the original graph that are within distance @math , and is referred to as graph powering. It is shown that graph powering regularizes the graph and decontaminates its spectrum in the following sense: (i) If the graph is drawn from the sparse Erd o s-R 'enyi ensemble, which has no spectral gap, it is shown that graph powering produces a maximal' spectral gap, with the latter justified by establishing an Alon-Boppana result for powered graphs; (ii) If the graph is drawn from the sparse SBM, graph powering is shown to achieve the fundamental limit for weak recovery (the KS threshold) similarly to massoulie-STOC , settling an open problem therein. Further, graph powering is shown to be significantly more robust to tangles and cliques than previous spectral algorithms based on self-avoiding or nonbacktracking walk counts massoulie-STOC,Mossel_SBM2,bordenave,colin3 . This is illustrated on a geometric block model that is dense in cliques.", "Many algorithms have been proposed for fitting network models with communities, but most of them do not scale well to large networks, and often fail on sparse networks. Here we propose a new fast pseudo-likelihood method for fitting the stochastic block model for networks, as well as a variant that allows for an arbitrary degree distribution by conditioning on degrees. We show that the algorithms perform well under a range of settings, including on very sparse networks, and illustrate on the example of a network of political blogs. We also propose spectral clustering with perturbations, a method of independent interest, which works well on sparse networks where regular spectral clustering fails, and use it to provide an initial value for pseudo-likelihood. We prove that pseudo-likelihood provides consistent estimates of the communities under a mild condition on the starting value, for the case of a block model with two communities.", "In this paper we extend our previous work on the stochastic block model, a commonly used generative model for social and biological networks, and the problem of inferring functional groups or communities from the topology of the network. We use the cavity method of statistical physics to obtain an asymptotically exact analysis of the phase diagram. We describe in detail properties of the detectability undetectability phase transition and the easy hard phase transition for the community detection problem. Our analysis translates naturally into a belief propagation algorithm for inferring the group memberships of the nodes in an optimal way, i.e., that maximizes the overlap with the underlying group memberships, and learning the underlying parameters of the block model. Finally, we apply the algorithm to two examples of real-world networks and discuss its performance.", "Although spectral clustering has enjoyed considerable empirical success in machine learning, its theoretical properties are not yet fully developed. We analyze the performance of a spectral algorithm for hierarchical clustering and show that on a class of hierarchically structured similarity matrices, this algorithm can tolerate noise that grows with the number of data points while still perfectly recovering the hierarchical clusters with high probability. We additionally improve upon previous results for k-way spectral clustering to derive conditions under which spectral clustering makes no mistakes. Further, using minimax analysis, we derive tight upper and lower bounds for the clustering problem and compare the performance of spectral clustering to these information theoretic limits. We also present experiments on simulated and real world data illustrating our results.", "", "[1] conjectured the existence of a sharp threshold on model parameters for community detection in sparse random graphs drawn from the stochastic block model. Mossel, Neeman and Sly [2] established the negative part of the conjecture, proving impossibility of non-trivial reconstruction below the threshold. In this work we solve the positive part of the conjecture. To that end we introduce a modified adjacency matrix B which counts self-avoiding paths of a given length e between pairs of nodes. We then prove that for logarithmic length e, the leading eigenvectors of this modified matrix provide a non-trivial reconstruction of the underlying structure, thereby settling the conjecture. A key step in the proof consists in establishing a weak Ramanujan property of the constructed matrix B. Namely, the spectrum of B consists in two leading eigenvalues ρ(B), λ2 and n -- 2 eigenvalues of a lower order O(ne √ρ(B) for all e 0, ρ(B) denoting B's spectral radius." ] }
1905.10029	2946234452	Graph convolutional networks (GCNs) are powerful tools for graph-structured data. However, they have been recently shown to be prone to topological attacks. Despite substantial efforts to search for new architectures, it still remains a challenge to improve performance in both benign and adversarial situations simultaneously. In this paper, we re-examine the fundamental building block of GCN---the Laplacian operator---and highlight some basic flaws in the spatial and spectral domains. As an alternative, we propose an operator based on graph powering, and prove that it enjoys a desirable property of "spectral separation." Based on the operator, we propose a robust learning paradigm, where the network is trained on a family of "'smoothed" graphs that span a spatial and spectral range for generalizability. We also use the new operator in replacement of the classical Laplacian to construct an architecture with improved spectral robustness, expressivity and interpretability. The enhanced performance and robustness are demonstrated in extensive experiments.	While there is a surge of adversarial attacks on graph neural networks, very few methods have been proposed to defend against such attacks @cite_20 . Existing works experimented adversarial training, namely to add or drop edges during training @cite_21 @cite_23 . Soft labels'' have also been employed @cite_23 . However, the main issue is that these methods are direct applications of known methods from computer vision @cite_22 @cite_36 @cite_33 , and do not exploit the unique characteristics of graph-structured data. In particular, it has not been shown whether such approach will reduce performance in the usual testing with clean data, but it is likely to be a collateral cost of such training techniques. By comparison, our work shows that it is indeed possible to simultaneously improve both performance on clean data and robustness against adversarial attacks, as shown in Figure (details are presented in Section ).	{ "cite_N": [ "@cite_22", "@cite_33", "@cite_36", "@cite_21", "@cite_23", "@cite_20" ], "mid": [ "1673923490", "2183341477", "2963207607", "2964283260", "2921543411", "2908442265" ], "abstract": [ "Deep neural networks are highly expressive models that have recently achieved state of the art performance on speech and visual recognition tasks. While their expressiveness is the reason they succeed, it also causes them to learn uninterpretable solutions that could have counter-intuitive properties. In this paper we report two such properties. First, we find that there is no distinction between individual high level units and random linear combinations of high level units, according to various methods of unit analysis. It suggests that it is the space, rather than the individual units, that contains of the semantic information in the high layers of neural networks. Second, we find that deep neural networks learn input-output mappings that are fairly discontinuous to a significant extend. We can cause the network to misclassify an image by applying a certain imperceptible perturbation, which is found by maximizing the network's prediction error. In addition, the specific nature of these perturbations is not a random artifact of learning: the same perturbation can cause a different network, that was trained on a different subset of the dataset, to misclassify the same input.", "Convolutional networks are at the core of most state of-the-art computer vision solutions for a wide variety of tasks. Since 2014 very deep convolutional networks started to become mainstream, yielding substantial gains in various benchmarks. Although increased model size and computational cost tend to translate to immediate quality gains for most tasks (as long as enough labeled data is provided for training), computational efficiency and low parameter count are still enabling factors for various use cases such as mobile vision and big-data scenarios. Here we are exploring ways to scale up networks in ways that aim at utilizing the added computation as efficiently as possible by suitably factorized convolutions and aggressive regularization. We benchmark our methods on the ILSVRC 2012 classification challenge validation set demonstrate substantial gains over the state of the art: 21:2 top-1 and 5:6 top-5 error for single frame evaluation using a network with a computational cost of 5 billion multiply-adds per inference and with using less than 25 million parameters. With an ensemble of 4 models and multi-crop evaluation, we report 3:5 top-5 error and 17:3 top-1 error on the validation set and 3:6 top-5 error on the official test set.", "Abstract: Several machine learning models, including neural networks, consistently misclassify adversarial examples---inputs formed by applying small but intentionally worst-case perturbations to examples from the dataset, such that the perturbed input results in the model outputting an incorrect answer with high confidence. Early attempts at explaining this phenomenon focused on nonlinearity and overfitting. We argue instead that the primary cause of neural networks' vulnerability to adversarial perturbation is their linear nature. This explanation is supported by new quantitative results while giving the first explanation of the most intriguing fact about them: their generalization across architectures and training sets. Moreover, this view yields a simple and fast method of generating adversarial examples. Using this approach to provide examples for adversarial training, we reduce the test set error of a maxout network on the MNIST dataset.", "", "Machine learning has been successfully applied to complex network analysis in various areas, and graph neural networks (GNNs) based methods outperform others. Recently, adversarial attack on networks has attracted special attention since carefully crafted adversarial networks with slight perturbations on clean network may invalid lots of network applications, such as node classification, link prediction, and community detection etc. Such attacks are easily constructed with serious security threat to various analyze methods, including traditional methods and deep models. To the best of our knowledge, it is the first time that defense method against network adversarial attack is discussed. In this paper, we are interested in the possibility of defense against adversarial attack on network, and propose defense strategies for GNNs against attacks. First, we propose novel adversarial training strategies to improve GNNs' defensibility against attacks. Then, we analytically investigate the robustness properties for GNNs granted by the use of smooth defense, and propose two special smooth defense strategies: smoothing distillation and smoothing cross-entropy loss function. Both of them are capable of smoothing gradient of GNNs, and consequently reduce the amplitude of adversarial gradients, which benefits gradient masking from attackers. The comprehensive experiments show that our proposed strategies have great defensibility against different adversarial attacks on four real-world networks in different network analyze tasks.", "Deep neural networks (DNNs) have been widely applied in various applications involving image, text, audio, and graph data. However, recent studies have shown that DNNs are vulnerable to adversarial attack. Though there are several works studying adversarial attack and defense on domains such as images and text processing, it is difficult to directly transfer the learned knowledge to graph data due to its representation challenge. Given the importance of graph analysis, increasing number of works start to analyze the robustness of machine learning models on graph. Nevertheless, current studies considering adversarial behaviors on graph data usually focus on specific types of attacks with certain assumptions. In addition, each work proposes its own mathematical formulation which makes the comparison among different methods difficult. Therefore, in this paper, we aim to survey existing adversarial attack strategies on graph data and provide an unified problem formulation which can cover all current adversarial learning studies on graph. We also compare different attacks on graph data and discuss their corresponding contributions and limitations. Finally, we discuss several future research directions in this area." ] }
1905.10070	2945656525	Extreme multi-label text classification (XMTC) aims at tagging a document with most relevant labels from an extremely large-scale label set. It is a challenging problem especially for the tail labels because there are only few training documents to build classifier. This paper is motivated to better explore the semantic relationship between each document and extreme labels by taking advantage of both document content and label correlation. Our objective is to establish an explicit label-aware representation for each document with a hybrid attention deep neural network model(LAHA). LAHA consists of three parts. The first part adopts a multi-label self-attention mechanism to detect the contribution of each word to labels. The second part exploits the label structure and document content to determine the semantic connection between words and labels in a same latent space. An adaptive fusion strategy is designed in the third part to obtain the final label-aware document representation so that the essence of previous two parts can be sufficiently integrated. Extensive experiments have been conducted on six benchmark datasets by comparing with the state-of-the-art methods. The results show the superiority of our proposed LAHA method, especially on the tail labels.	Embedding-based methods aim at reducing the huge label space to a low dimensional space while preserving the label correlation as much as possible. Various approaches have been presented such as compressed sensing @cite_28 , output codes @cite_3 , Singular Value Decomposition @cite_7 , landmark labels @cite_0 , Bloom filters @cite_10 , etc. To efficiently handle large-scale label set, these embedding-based methods usually assume that the label matrix is low-rank. However, such methods have been proved unable to deliver high prediction accuracies as the low rank assumption is violated in most real world applications @cite_19 . @cite_19 can be taken as the most representative embedding-based method due to its significant accuracy and computationally efficiency. Its main idea is to learn a small ensemble of local distance preserving embeddings. Specifically, divides the training data set into several clusters, and in each cluster it detects embedding vectors by capturing non-linear label correlation and preserving the pairwise distance between labels. The @math -nearest neighbors search is used to do prediction only in the cluster into which the test document is fallen. Later, @cite_14 adopted deep neural network for non-linear modeling the label embedding.	{ "cite_N": [ "@cite_14", "@cite_7", "@cite_28", "@cite_3", "@cite_0", "@cite_19", "@cite_10" ], "mid": [ "2949249496", "2027869746", "2951238624", "2292376821", "2953000611", "2183087644", "2154109204" ], "abstract": [ "Extreme multi-label learning (XML) or classification has been a practical and important problem since the boom of big data. The main challenge lies in the exponential label space which involves @math possible label sets especially when the label dimension @math is huge, e.g., in millions for Wikipedia labels. This paper is motivated to better explore the label space by originally establishing an explicit label graph. In the meanwhile, deep learning has been widely studied and used in various classification problems including multi-label classification, however it has not been properly introduced to XML, where the label space can be as large as in millions. In this paper, we propose a practical deep embedding method for extreme multi-label classification, which harvests the ideas of non-linear embedding and graph priors-based label space modeling simultaneously. Extensive experiments on public datasets for XML show that our method performs competitive against state-of-the-art result.", "We consider a hypercube view to perceive the label space of multilabel classification problems geometrically. The view allows us not only to unify many existing multilabel classification approaches but also design a novel algorithm, principal label space transformation (PLST), that captures key correlations between labels before learning. The simple and efficient PLST relies on only singular value decomposition as the key step. We derive the theoretical guarantee of PLST and evaluate its empirical performance using real-world data sets. Experimental results demonstrate that PLST is faster than the traditional binary relevance approach and is superior to the modern compressive sensing approach in terms of both accuracy and efficiency.", "We consider multi-label prediction problems with large output spaces under the assumption of output sparsity -- that the target (label) vectors have small support. We develop a general theory for a variant of the popular error correcting output code scheme, using ideas from compressed sensing for exploiting this sparsity. The method can be regarded as a simple reduction from multi-label regression problems to binary regression problems. We show that the number of subproblems need only be logarithmic in the total number of possible labels, making this approach radically more efficient than others. We also state and prove robustness guarantees for this method in the form of regret transform bounds (in general), and also provide a more detailed analysis for the linear prediction setting.", "Traditional error-correcting output codes (ECOCs) decompose a multi-class classification problem into many binary problems. Although it seems natural to use ECOCs for multi-label problems as well, doing so naively creates issues related to: the validity of the encoding, the efficiency of the decoding, the predictability of the generated codeword, and the exploitation of the label dependency. Using canonical correlation analysis, we propose an error-correcting code for multi-label classification. Label dependency is characterized as the most predictable directions in the label space, which are extracted as canonical output variates and encoded into the codeword. Predictions for the codeword define a graphical model of labels with both Bernoulli potentials (from classifiers on the labels) and Gaussian potentials (from regression on the canonical output variates). Decoding is performed by mean-field approximation. We establish connections between the proposed code and research areas such as compressed sensing and ensemble learning. Some of these connections contribute to better understanding of the new code, and others lead to practical improvements in code design. In our empirical study, the proposed code leads to substantial improvements compared to various competitors in music emotion classification and outdoor scene recognition.", "Conditional modeling x y is a central problem in machine learning. A substantial research effort is devoted to such modeling when x is high dimensional. We consider, instead, the case of a high dimensional y, where x is either low dimensional or high dimensional. Our approach is based on selecting a small subset y_L of the dimensions of y, and proceed by modeling (i) x y_L and (ii) y_L y. Composing these two models, we obtain a conditional model x y that possesses convenient statistical properties. Multi-label classification and multivariate regression experiments on several datasets show that this model outperforms the one vs. all approach as well as several sophisticated multiple output prediction methods.", "The objective in extreme multi-label learning is to train a classifier that can automatically tag a novel data point with the most relevant subset of labels from an extremely large label set. Embedding based approaches attempt to make training and prediction tractable by assuming that the training label matrix is low-rank and reducing the effective number of labels by projecting the high dimensional label vectors onto a low dimensional linear subspace. Still, leading embedding approaches have been unable to deliver high prediction accuracies, or scale to large problems as the low rank assumption is violated in most real world applications. In this paper we develop the SLEEC classifier to address both limitations. The main technical contribution in SLEEC is a formulation for learning a small ensemble of local distance preserving embeddings which can accurately predict infrequently occurring (tail) labels. This allows SLEEC to break free of the traditional low-rank assumption and boost classification accuracy by learning embeddings which preserve pairwise distances between only the nearest label vectors. We conducted extensive experiments on several real-world, as well as benchmark data sets and compared our method against state-of-the-art methods for extreme multi-label classification. Experiments reveal that SLEEC can make significantly more accurate predictions then the state-of-the-art methods including both embedding-based (by as much as 35 ) as well as tree-based (by as much as 6 ) methods. SLEEC can also scale efficiently to data sets with a million labels which are beyond the pale of leading embedding methods.", "This paper presents an approach to multilabel classification (MLC) with a large number of labels. Our approach is a reduction to binary classification in which label sets are represented by low dimensional binary vectors. This representation follows the principle of Bloom filters, a space-efficient data structure originally designed for approximate membership testing. We show that a naive application of Bloom filters in MLC is not robust to individual binary classifiers' errors. We then present an approach that exploits a specific feature of real-world datasets when the number of labels is large: many labels (almost) never appear together. Our approach is provably robust, has sublinear training and inference complexity with respect to the number of labels, and compares favorably to state-of-the-art algorithms on two large scale multilabel datasets." ] }
1905.10070	2945656525	Extreme multi-label text classification (XMTC) aims at tagging a document with most relevant labels from an extremely large-scale label set. It is a challenging problem especially for the tail labels because there are only few training documents to build classifier. This paper is motivated to better explore the semantic relationship between each document and extreme labels by taking advantage of both document content and label correlation. Our objective is to establish an explicit label-aware representation for each document with a hybrid attention deep neural network model(LAHA). LAHA consists of three parts. The first part adopts a multi-label self-attention mechanism to detect the contribution of each word to labels. The second part exploits the label structure and document content to determine the semantic connection between words and labels in a same latent space. An adaptive fusion strategy is designed in the third part to obtain the final label-aware document representation so that the essence of previous two parts can be sufficiently integrated. Extensive experiments have been conducted on six benchmark datasets by comparing with the state-of-the-art methods. The results show the superiority of our proposed LAHA method, especially on the tail labels.	Tree-based methods, similar to the traditional Decision-Tree, introduce a tree structure to divide the documents recursively at each non-leaf node, so that documents in each leaf node share the similar label distribution. The most representative method @cite_25 implements this process by optimizing the normalized discounted cumulative gain (nDCG)-based ranking loss function. Then, a base classifier is trained at each leaf node which only focuses on a few active labels. To enhance the robustness of predictions, an ensemble of multiple induced trees are learned. The main advantage of tree-based methods is that the prediction time complexity is typically sublinear in the training-set size and would be logarithmic if the induced tree is balanced.	{ "cite_N": [ "@cite_25" ], "mid": [ "2068074736" ], "abstract": [ "The objective in extreme multi-label classification is to learn a classifier that can automatically tag a data point with the most relevant subset of labels from a large label set. Extreme multi-label classification is an important research problem since not only does it enable the tackling of applications with many labels but it also allows the reformulation of ranking problems with certain advantages over existing formulations. Our objective, in this paper, is to develop an extreme multi-label classifier that is faster to train and more accurate at prediction than the state-of-the-art Multi-label Random Forest (MLRF) algorithm [2] and the Label Partitioning for Sub-linear Ranking (LPSR) algorithm [35]. MLRF and LPSR learn a hierarchy to deal with the large number of labels but optimize task independent measures, such as the Gini index or clustering error, in order to learn the hierarchy. Our proposed FastXML algorithm achieves significantly higher accuracies by directly optimizing an nDCG based ranking loss function. We also develop an alternating minimization algorithm for efficiently optimizing the proposed formulation. Experiments reveal that FastXML can be trained on problems with more than a million labels on a standard desktop in eight hours using a single core and in an hour using multiple cores." ] }
1905.10070	2945656525	Extreme multi-label text classification (XMTC) aims at tagging a document with most relevant labels from an extremely large-scale label set. It is a challenging problem especially for the tail labels because there are only few training documents to build classifier. This paper is motivated to better explore the semantic relationship between each document and extreme labels by taking advantage of both document content and label correlation. Our objective is to establish an explicit label-aware representation for each document with a hybrid attention deep neural network model(LAHA). LAHA consists of three parts. The first part adopts a multi-label self-attention mechanism to detect the contribution of each word to labels. The second part exploits the label structure and document content to determine the semantic connection between words and labels in a same latent space. An adaptive fusion strategy is designed in the third part to obtain the final label-aware document representation so that the essence of previous two parts can be sufficiently integrated. Extensive experiments have been conducted on six benchmark datasets by comparing with the state-of-the-art methods. The results show the superiority of our proposed LAHA method, especially on the tail labels.	A recent extension work of is @cite_12 , which adopted a propensity scored objective function instead of nDCG-based loss. can prioritize predicting the few relevant labels over the large number of irrelevant ones and thus promote the accurate prediction of tail labels. @cite_11 is another tree-based method, which constructs balanced trees partitioning labels rather than instances. These tree-based methods sufficiently consider the label distribution when training the classifier, but they represent document via bag-of-words, where the words are treated as independent features, which will ignore the semantic dependency among words.	{ "cite_N": [ "@cite_12", "@cite_11" ], "mid": [ "2362855512", "2788125153" ], "abstract": [ "The choice of the loss function is critical in extreme multi-label learning where the objective is to annotate each data point with the most relevant subset of labels from an extremely large label set. Unfortunately, existing loss functions, such as the Hamming loss, are unsuitable for learning, model selection, hyperparameter tuning and performance evaluation. This paper addresses the issue by developing propensity scored losses which: (a) prioritize predicting the few relevant labels over the large number of irrelevant ones; (b) do not erroneously treat missing labels as irrelevant but instead provide unbiased estimates of the true loss function even when ground truth labels go missing under arbitrary probabilistic label noise models; and (c) promote the accurate prediction of infrequently occurring, hard to predict, but rewarding tail labels. Another contribution is the development of algorithms which efficiently scale to extremely large datasets with up to 9 million labels, 70 million points and 2 million dimensions and which give significant improvements over the state-of-the-art. This paper's results also apply to tagging, recommendation and ranking which are the motivating applications for extreme multi-label learning. They generalize previous attempts at deriving unbiased losses under the restrictive assumption that labels go missing uniformly at random from the ground truth. Furthermore, they provide a sound theoretical justification for popular label weighting heuristics used to recommend rare items. Finally, they demonstrate that the proposed contributions align with real world applications by achieving superior clickthrough rates on sponsored search advertising in Bing.", "This paper develops the Parabel algorithm for extreme multi-label learning where the objective is to learn classifiers that can annotate a data point with the most relevant subset of labels from an extremely large label set. The state-of-the-art DiSMEC and PPDSparse algorithms are the most accurate but take weeks for training and prediction as they learn and apply an independent linear classifier per label. Consequently, they do not scale to large datasets with millions of labels. Parabel addresses these limitations by: (a) cutting down the training time to a few hours on a single core of a standard desktop by learning a hierarchy of coarse to fine label classifiers, each trained on a small subset of datapoints and (b) by cutting down the prediction time to a few milliseconds per test point by leveraging the classifier hierarchy for logarithmic time prediction in number of labels. This allows Parabel to scale to tasks considered infeasible for DiSMEC and PPDSparse such as predicting the subset of search engine queries that might lead to a click on a given ad-landing page for dynamic search advertising. Experimental results demonstrated that Parabel could train in 80 hours on a proprietary dataset with 7 million labels which is beyond the scale of both DiSMEC and PPDSparse. Results on some of the largest publically available datasets revealed that Parabel could be 1,000x faster at training than both DiSMEC and PPDSparse, as well as 10,000x and 40x faster at prediction than DiSMEC and PPDSparse without any significant loss in prediction accuracy. Moreover, Parabel was also found to be much more accurate than other tree based extreme classifiers and could be more than 10x faster at training with a 10x smaller model. Finally, Parabel was demonstrated to significantly improve dynamic search advertising on Bing by more than doubling the ad coverage, as well as improving the click-through rate by 20 ." ] }
1905.10070	2945656525	Extreme multi-label text classification (XMTC) aims at tagging a document with most relevant labels from an extremely large-scale label set. It is a challenging problem especially for the tail labels because there are only few training documents to build classifier. This paper is motivated to better explore the semantic relationship between each document and extreme labels by taking advantage of both document content and label correlation. Our objective is to establish an explicit label-aware representation for each document with a hybrid attention deep neural network model(LAHA). LAHA consists of three parts. The first part adopts a multi-label self-attention mechanism to detect the contribution of each word to labels. The second part exploits the label structure and document content to determine the semantic connection between words and labels in a same latent space. An adaptive fusion strategy is designed in the third part to obtain the final label-aware document representation so that the essence of previous two parts can be sufficiently integrated. Extensive experiments have been conducted on six benchmark datasets by comparing with the state-of-the-art methods. The results show the superiority of our proposed LAHA method, especially on the tail labels.	To solve the fundamental limitation of the aforementioned methods, ignoring the semantic dependency among words, researchers adopted deep learning models in text classification task due to its strong ability of representation. Its main idea is to compress the input word-level representation into a document-level representation, i.e., they encode one document in a latent space @cite_26 by considering the contextual information. The popular deep models include convolutional neural network (CNN) @cite_17 , gated recurrent units (GRU) @cite_5 , recurrent neural network (RNN) @cite_13 , long short-term memory (LSTM) @cite_27 , bidirectional LSTM (Bi-LSTM) @cite_9 , bidirectional encoder representation from transformer (BERT) @cite_22 and several combination networks @cite_23 @cite_29 . Even though they have achieved great success in traditional NLP tasks, few work is designed for XMTC.	{ "cite_N": [ "@cite_26", "@cite_22", "@cite_9", "@cite_29", "@cite_27", "@cite_23", "@cite_5", "@cite_13", "@cite_17" ], "mid": [ "2963973721", "2896457183", "2950141408", "2571563030", "2221711388", "2284289336", "2950635152", "2130942839", "2963012544" ], "abstract": [ "", "We introduce a new language representation model called BERT, which stands for Bidirectional Encoder Representations from Transformers. Unlike recent language representation models, BERT is designed to pre-train deep bidirectional representations from unlabeled text by jointly conditioning on both left and right context in all layers. As a result, the pre-trained BERT model can be fine-tuned with just one additional output layer to create state-of-the-art models for a wide range of tasks, such as question answering and language inference, without substantial task-specific architecture modifications. BERT is conceptually simple and empirically powerful. It obtains new state-of-the-art results on eleven natural language processing tasks, including pushing the GLUE score to 80.5 (7.7 point absolute improvement), MultiNLI accuracy to 86.7 (4.6 absolute improvement), SQuAD v1.1 question answering Test F1 to 93.2 (1.5 point absolute improvement) and SQuAD v2.0 Test F1 to 83.1 (5.1 point absolute improvement).", "Recurrent Neural Network (RNN) is one of the most popular architectures used in Natural Language Processsing (NLP) tasks because its recurrent structure is very suitable to process variable-length text. RNN can utilize distributed representations of words by first converting the tokens comprising each text into vectors, which form a matrix. And this matrix includes two dimensions: the time-step dimension and the feature vector dimension. Then most existing models usually utilize one-dimensional (1D) max pooling operation or attention-based operation only on the time-step dimension to obtain a fixed-length vector. However, the features on the feature vector dimension are not mutually independent, and simply applying 1D pooling operation over the time-step dimension independently may destroy the structure of the feature representation. On the other hand, applying two-dimensional (2D) pooling operation over the two dimensions may sample more meaningful features for sequence modeling tasks. To integrate the features on both dimensions of the matrix, this paper explores applying 2D max pooling operation to obtain a fixed-length representation of the text. This paper also utilizes 2D convolution to sample more meaningful information of the matrix. Experiments are conducted on six text classification tasks, including sentiment analysis, question classification, subjectivity classification and newsgroup classification. Compared with the state-of-the-art models, the proposed models achieve excellent performance on 4 out of 6 tasks. Specifically, one of the proposed models achieves highest accuracy on Stanford Sentiment Treebank binary classification and fine-grained classification tasks.", "", "Natural language inference (NLI) is a fundamentally important task in natural language processing that has many applications. The recently released Stanford Natural Language Inference (SNLI) corpus has made it possible to develop and evaluate learning-centered methods such as deep neural networks for natural language inference (NLI). In this paper, we propose a special long short-term memory (LSTM) architecture for NLI. Our model builds on top of a recently proposed neural attention model for NLI but is based on a significantly different idea. Instead of deriving sentence embeddings for the premise and the hypothesis to be used for classification, our solution uses a match-LSTM to perform word-by-word matching of the hypothesis with the premise. This LSTM is able to place more emphasis on important word-level matching results. In particular, we observe that this LSTM remembers important mismatches that are critical for predicting the contradiction or the neutral relationship label. On the SNLI corpus, our model achieves an accuracy of 86.1 , outperforming the state of the art.", "Neural network models have been demonstrated to be capable of achieving remarkable performance in sentence and document modeling. Convolutional neural network (CNN) and recurrent neural network (RNN) are two mainstream architectures for such modeling tasks, which adopt totally different ways of understanding natural languages. In this work, we combine the strengths of both architectures and propose a novel and unified model called C-LSTM for sentence representation and text classification. C-LSTM utilizes CNN to extract a sequence of higher-level phrase representations, and are fed into a long short-term memory recurrent neural network (LSTM) to obtain the sentence representation. C-LSTM is able to capture both local features of phrases as well as global and temporal sentence semantics. We evaluate the proposed architecture on sentiment classification and question classification tasks. The experimental results show that the C-LSTM outperforms both CNN and LSTM and can achieve excellent performance on these tasks.", "In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.", "Deep Neural Networks (DNNs) are powerful models that have achieved excellent performance on difficult learning tasks. Although DNNs work well whenever large labeled training sets are available, they cannot be used to map sequences to sequences. In this paper, we present a general end-to-end approach to sequence learning that makes minimal assumptions on the sequence structure. Our method uses a multilayered Long Short-Term Memory (LSTM) to map the input sequence to a vector of a fixed dimensionality, and then another deep LSTM to decode the target sequence from the vector. Our main result is that on an English to French translation task from the WMT-14 dataset, the translations produced by the LSTM achieve a BLEU score of 34.8 on the entire test set, where the LSTM's BLEU score was penalized on out-of-vocabulary words. Additionally, the LSTM did not have difficulty on long sentences. For comparison, a phrase-based SMT system achieves a BLEU score of 33.3 on the same dataset. When we used the LSTM to rerank the 1000 hypotheses produced by the aforementioned SMT system, its BLEU score increases to 36.5, which is close to the previous state of the art. The LSTM also learned sensible phrase and sentence representations that are sensitive to word order and are relatively invariant to the active and the passive voice. Finally, we found that reversing the order of the words in all source sentences (but not target sentences) improved the LSTM's performance markedly, because doing so introduced many short term dependencies between the source and the target sentence which made the optimization problem easier.", "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." ] }
1905.10070	2945656525	Extreme multi-label text classification (XMTC) aims at tagging a document with most relevant labels from an extremely large-scale label set. It is a challenging problem especially for the tail labels because there are only few training documents to build classifier. This paper is motivated to better explore the semantic relationship between each document and extreme labels by taking advantage of both document content and label correlation. Our objective is to establish an explicit label-aware representation for each document with a hybrid attention deep neural network model(LAHA). LAHA consists of three parts. The first part adopts a multi-label self-attention mechanism to detect the contribution of each word to labels. The second part exploits the label structure and document content to determine the semantic connection between words and labels in a same latent space. An adaptive fusion strategy is designed in the third part to obtain the final label-aware document representation so that the essence of previous two parts can be sufficiently integrated. Extensive experiments have been conducted on six benchmark datasets by comparing with the state-of-the-art methods. The results show the superiority of our proposed LAHA method, especially on the tail labels.	@cite_18 can be taken as the first and most representative work using deep learning model in XMTC. It takes advantage of CNN, dynamic max-pooling and bottle-neck layer to build the deep model. Due to the limited window size, can not capture the long-distance dependency among text. Later, GRU and Bi-LSTM language models are adopted in @cite_8 and @cite_21 to effectively represent document for XMTC. Meanwhile, these two methods consider the difference of one document represesntation along different labels. @cite_8 adopts self-attention mechanism @cite_15 , while @cite_21 exploits the label content information to calculate the relations between words and classes. Although obtains promising performance, it ignores the label structure which has been proved very important in embedding-based and tree-based multi-label learning methods.	{ "cite_N": [ "@cite_18", "@cite_21", "@cite_15", "@cite_8" ], "mid": [ "2739996966", "2901115292", "2597655663", "2899313146" ], "abstract": [ "Extreme multi-label text classification (XMTC) refers to the problem of assigning to each document its most relevant subset of class labels from an extremely large label collection, where the number of labels could reach hundreds of thousands or millions. The huge label space raises research challenges such as data sparsity and scalability. Significant progress has been made in recent years by the development of new machine learning methods, such as tree induction with large-margin partitions of the instance spaces and label-vector embedding in the target space. However, deep learning has not been explored for XMTC, despite its big successes in other related areas. This paper presents the first attempt at applying deep learning to XMTC, with a family of new Convolutional Neural Network (CNN) models which are tailored for multi-label classification in particular. With a comparative evaluation of 7 state-of-the-art methods on 6 benchmark datasets where the number of labels is up to 670,000, we show that the proposed CNN approach successfully scaled to the largest datasets, and consistently produced the best or the second best results on all the datasets. On the Wikipedia dataset with over 2 million documents and 500,000 labels in particular, it outperformed the second best method by 11.7 15.3 in precision@K and by 11.5 11.7 in NDCG@K for K = 1,3,5.", "Text classification is one of the fundamental tasks in natural language processing. Recently, deep neural networks have achieved promising performance in the text classification task compared to shallow models. Despite of the significance of deep models, they ignore the fine-grained (matching signals between words and classes) classification clues since their classifications mainly rely on the text-level representations. To address this problem, we introduce the interaction mechanism to incorporate word-level matching signals into the text classification task. In particular, we design a novel framework, EXplicit interAction Model (dubbed as EXAM), equipped with the interaction mechanism. We justified the proposed approach on several benchmark datasets including both multi-label and multi-class text classification tasks. Extensive experimental results demonstrate the superiority of the proposed method. As a byproduct, we have released the codes and parameter settings to facilitate other researches.", "This paper proposes a new model for extracting an interpretable sentence embedding by introducing self-attention. Instead of using a vector, we use a 2-D matrix to represent the embedding, with each row of the matrix attending on a different part of the sentence. We also propose a self-attention mechanism and a special regularization term for the model. As a side effect, the embedding comes with an easy way of visualizing what specific parts of the sentence are encoded into the embedding. We evaluate our model on 3 different tasks: author profiling, sentiment classification, and textual entailment. Results show that our model yields a significant performance gain compared to other sentence embedding methods in all of the 3 tasks.", "Extreme multi-label text classification (XMTC) is a task for tagging each given text with the most relevant multiple labels from an extremely large-scale label set. This task can be found in many applications, such as product categorization,web page tagging, news annotation and so on. Many methods have been proposed so far for solving XMTC, while most of the existing methods use traditional bag-of-words (BOW) representation, ignoring word context as well as deep semantic information. XML-CNN, a state-of-the-art deep learning-based method, uses convolutional neural network (CNN) with dynamic pooling to process the text, going beyond the BOW-based appraoches but failing to capture 1) the long-distance dependency among words and 2) different levels of importance of a word for each label. We propose a new deep learning-based method, AttentionXML, which uses bidirectional long short-term memory (LSTM) and a multi-label attention mechanism for solving the above 1st and 2nd problems, respectively. We empirically compared AttentionXML with other six state-of-the-art methods over five benchmark datasets. AttentionXML outperformed all competing methods under all experimental settings except only a couple of cases. In addition, a consensus ensemble of AttentionXML with the second best method, Parabel, could further improve the performance over all five benchmark datasets." ] }
1905.09964	2946364891	We introduce the Skipping Sampler, a novel algorithm to efficiently sample from the restriction of an arbitrary probability density to an arbitrary measurable set. Such conditional densities can arise in the study of risk and reliability and are often of complex nature, for example having multiple isolated modes and non-convex or disconnected support. The sampler can be seen as an instance of the Metropolis-Hastings algorithm with a particular proposal structure, and we establish sufficient conditions under which the Strong Law of Large Numbers and the Central Limit Theorem hold. We give theoretical and numerical evidence of improved performance relative to the Random Walk Metropolis algorithm.	The Skipping Sampler is an adaptation of the MH algorithm designed to sample from targets which have areas of zero density. It skips' across such areas, much as a flat stone can skip or skim repeatedly across the surface of water. Our original motivation is the study of rare events in stochastic systems, such as that in @cite_36 and @cite_27 . Suppose the system has been designed to satisfy some constraint with high probability, under a random exogenous input vector @math which is known only at the level of its distribution. For example, a power system should remain secure despite the randomness in renewable generation. Although the subset @math where the constraint is violated has a small probability mass @math (hence corresponding to a rare event), nevertheless violations may be extremely costly, so that simulating their occurrence is important to improve system design.	{ "cite_N": [ "@cite_36", "@cite_27" ], "mid": [ "2963473861", "2756907874" ], "abstract": [ "The frequency stability of power systems is increasingly challenged by various types of disturbance. In particular, the increasing penetration of renewable energy sources is increasing the variability of power generation while reducing system inertia against disturbances. In this paper we explore how this could give rise to rate of change of frequency (RoCoF) violations. Correlated and non -Gaussian power disturbances, such as may arise from renewable generation, have been shown to be significant in power system security analysis. We therefore introduce ghost sampling which, given any unconditional distribution of disturbances, efficiently produces samples conditional on a violation occurring. Our goal is to address questions such as “which generator is most likely to be disconnected due to a RoCoF violation?” or “what is the probability of having simultaneous RoCoF violations, given that a violation occurs?”", "We model power grids transporting electricity generated by intermittent renewable sources as complex networks, where line failures can emerge indirectly by noisy power input at the nodes. By combining concepts from statistical physics and the physics of power flows and taking weather correlations into account, we rank line failures according to their likelihood and establish the most likely way such failures occur and propagate. Our insights are mathematically rigorous in a small-noise limit and are validated with data from the German transmission grid." ] }
1905.09964	2946364891	We introduce the Skipping Sampler, a novel algorithm to efficiently sample from the restriction of an arbitrary probability density to an arbitrary measurable set. Such conditional densities can arise in the study of risk and reliability and are often of complex nature, for example having multiple isolated modes and non-convex or disconnected support. The sampler can be seen as an instance of the Metropolis-Hastings algorithm with a particular proposal structure, and we establish sufficient conditions under which the Strong Law of Large Numbers and the Central Limit Theorem hold. We give theoretical and numerical evidence of improved performance relative to the Random Walk Metropolis algorithm.	In many stochastic systems, however, the relationship between exogenous inputs and system behaviour is implicit and non-linear. Examples include markets or distributed systems, where complexity arises from the actions of multiple players, or systems involving complex optimisation procedures such as security constrained optimal power flow @cite_21 . In the present work (and unlike @cite_36 ) we therefore do not assume that the set @math is known but, rather, that any sampling procedure may only test pointwise whether @math . As illustrated in Figure , the areas of zero density induced by conditioning on @math can create multiple isolated modes. We therefore now briefly recall the literature on sampling from rare events and from multimodal targets.	{ "cite_N": [ "@cite_36", "@cite_21" ], "mid": [ "2963473861", "349704142" ], "abstract": [ "The frequency stability of power systems is increasingly challenged by various types of disturbance. In particular, the increasing penetration of renewable energy sources is increasing the variability of power generation while reducing system inertia against disturbances. In this paper we explore how this could give rise to rate of change of frequency (RoCoF) violations. Correlated and non -Gaussian power disturbances, such as may arise from renewable generation, have been shown to be significant in power system security analysis. We therefore introduce ghost sampling which, given any unconditional distribution of disturbances, efficiently produces samples conditional on a violation occurring. Our goal is to address questions such as “which generator is most likely to be disconnected due to a RoCoF violation?” or “what is the probability of having simultaneous RoCoF violations, given that a violation occurs?”", "Preface. Chapter 1: Market Overview in Electric Power Systems. Chapter 2: Short-Term Load Forecasting. Chapter 3: Electricity Price Forecasting. Chapter 4: Price-Based Unit Commitment. Chapter 5: Arbitrage in Electricity Markets. Chapter 6: Market Power Analysis Based on Game Theory. Chapter 7: Generation Asset Valuation and Risk Analysis. Chapter 8: Security-Constrained Unit Commitment. Chapter 9: Ancillary Services Auction Market Design. Chapter 10: Transmission Congestion Management and Pricing. Appendix A: List of Symbols. Appendix B: Mathematical Derivation. Appendix C: RTS Load Data. Appendix D: Example Systems Data. Appendix E: Game Theory Concepts. Appendix F: Congestion Charges Calculation. References. Index." ] }
1905.09964	2946364891	We introduce the Skipping Sampler, a novel algorithm to efficiently sample from the restriction of an arbitrary probability density to an arbitrary measurable set. Such conditional densities can arise in the study of risk and reliability and are often of complex nature, for example having multiple isolated modes and non-convex or disconnected support. The sampler can be seen as an instance of the Metropolis-Hastings algorithm with a particular proposal structure, and we establish sufficient conditions under which the Strong Law of Large Numbers and the Central Limit Theorem hold. We give theoretical and numerical evidence of improved performance relative to the Random Walk Metropolis algorithm.	Many methods for sampling and integrating over a measurable subset @math use prior knowledge of the subset. In Splitting (see for example @cite_32 and references therein) and related approaches such as Sequential Monte Carlo @cite_34 , it is necessary to define a sequence of nested subsets which decrease to @math . If satisfaction of the constraint is determined implicitly (for example by a black box' function) then it may be unclear how to construct such a sequence. In methods using Importance Sampling (see for example @cite_29 @cite_32 ) a biased sampling distribution should be constructed based on knowledge of the set @math , to sample from @math more frequently, and integrals are corrected using certain weights.	{ "cite_N": [ "@cite_29", "@cite_34", "@cite_32" ], "mid": [ "1979969656", "2147357149", "" ], "abstract": [ "Importance sampling methods can be iterated like MCMC algorithms, while being more robust against dependence and starting values. The population Monte Carlo principle consists of iterated generations of importance samples, with importance functions depending on the previously generated importance samples. The advantage over MCMC algorithms is that the scheme is unbiased at any iteration and can thus be stopped at any time, while iterations improve the performances of the importance function, thus leading to an adaptive importance sampling. We illustrate this method on a mixture example with multiscale importance functions. A second example reanalyzes the ion channel model using an importance sampling scheme based on a hidden Markov representation, and compares population Monte Carlo with a corresponding MCMC algorithm.", "We propose a methodology to sample sequentially from a sequence of probability distributions that are defined on a common space, each distribution being known up to a normalizing constant. These probability distributions are approximated by a cloud of weighted random samples which are propagated over time by using sequential Monte Carlo methods. This methodology allows us to derive simple algorithms to make parallel Markov chain Monte Carlo algorithms interact to perform global optimization and sequential Bayesian estimation and to compute ratios of normalizing constants. We illustrate these algorithms for various integration tasks arising in the context of Bayesian inference. Copyright 2006 Royal Statistical Society.", "" ] }
1905.09964	2946364891	We introduce the Skipping Sampler, a novel algorithm to efficiently sample from the restriction of an arbitrary probability density to an arbitrary measurable set. Such conditional densities can arise in the study of risk and reliability and are often of complex nature, for example having multiple isolated modes and non-convex or disconnected support. The sampler can be seen as an instance of the Metropolis-Hastings algorithm with a particular proposal structure, and we establish sufficient conditions under which the Strong Law of Large Numbers and the Central Limit Theorem hold. We give theoretical and numerical evidence of improved performance relative to the Random Walk Metropolis algorithm.	Sampling schemes developed for multimodal targets typically require information about the target density @math . For example, known mode locations may be used to design global moves for the sampler @cite_11 @cite_2 @cite_41 @cite_39 @cite_6 @cite_0 , or local moves may be guided by the known derivatives of a differentiable target density @cite_41 @cite_33 . Alternatively, multimodality may be addressed by running a population of Markov chains in parallel @cite_20 @cite_44 @cite_15 @cite_7 . Among these, methods using a temperature-like parameter (see, for example, @cite_35 @cite_44 @cite_22 @cite_15 ) do not necessarily overcome the structure of @math , whose support may be disconnected by regions of zero density which are therefore insensitive to the temperature parameter. Population methods also typically use additional information on @math (which depends on @math ) in order to select appropriate parameters or initial distributions.	{ "cite_N": [ "@cite_35", "@cite_33", "@cite_7", "@cite_22", "@cite_41", "@cite_6", "@cite_39", "@cite_0", "@cite_44", "@cite_2", "@cite_15", "@cite_20", "@cite_11" ], "mid": [ "2170171286", "2339321571", "316347119", "2157326321", "2149959333", "116703061", "1594106604", "2165497092", "", "2136796925", "2013145865", "2162340617", "2085767032" ], "abstract": [ "We review the history of the parallel tempering simulation method. From its origins in data analysis, the parallel tempering method has become a standard workhorse of physicochemical simulations. We discuss the theory behind the method and its various generalizations. We mention a selected set of the many applications that have become possible with the introduction of parallel tempering, and we suggest several promising avenues for future research.", "ABSTRACTAlthough the Metropolis algorithm is simple to implement, it often has difficulties exploring multimodal distributions. We propose the repelling–attracting Metropolis (RAM) algorithm that maintains the simple-to-implement nature of the Metropolis algorithm, but is more likely to jump between modes. The RAM algorithm is a Metropolis-Hastings algorithm with a proposal that consists of a downhill move in density that aims to make local modes repelling, followed by an uphill move in density that aims to make local modes attracting. The downhill move is achieved via a reciprocal Metropolis ratio so that the algorithm prefers downward movement. The uphill move does the opposite using the standard Metropolis ratio which prefers upward movement. This down-up movement in density increases the probability of a proposed move to a different mode. Because the acceptance probability of the proposal involves a ratio of intractable integrals, we introduce an auxiliary variable which creates a term in the acceptan...", "Abstract : The Normal Kernel Coupler (NKC) is an adaptive Markov Chain Monte Carlo (MCMC) method which maintains a set of current state vectors. At each iteration one state vector is updated using a density estimate formed by applying a normal kernel to the full set of states. This sampler is ergodic (irreducible, Harris recurrent and aperiodic) for any continuous distribution on d-dimensioual Euclidean space. The NKC outperforms standard MCMC methods on a variety of unimodal and bimodal problems in low to moderate dimension. We illustrate the utility of the NKC by fitting a mixture model for genetic instability in cancer cells. This model which has two distinct and dissimilar modes is not well handled by standard MCMC methods. In contrast, the NKC efficiently samples from this model and yields results that are consistent with current scientific understanding.", "S U M M A R Y Non-linear inverse problems in the geosciences often involve probabilistic sampling of multimodal density functions or global optimization and sometimes both. Efficient algorithmic tools for carrying out sampling or optimization in challenging cases are of major interest. Here results are presented of some numerical experiments with a technique, known as Parallel Tempering, which originated in the field of computational statistics but is finding increasing numbers of applications in fields ranging from Chemical Physics to Astronomy. To date, experience in use of Parallel Tempering within earth sciences problems is very limited. In this paper, we describe Parallel Tempering and compare it to related methods of Simulated Annealing and Simulated Tempering for optimization and sampling, respectively. A key feature of Parallel Tempering is that it satisfies the detailed balance condition required for convergence of Markov chain Monte Carlo (McMC) algorithms while improving the efficiency of probabilistic sampling. Numerical results are presented on use of Parallel Tempering for trans-dimensional inversion of synthetic seismic receiver functions and also the simultaneous fitting of multiple receiver functions using global optimization. These suggest that its use can significantly accelerate sampling algorithms and improve exploration of parameter space in optimization. Parallel Tempering is a meta-algorithm which may be used together with many existing McMC sampling and direct search optimization techniques. It’s generality and demonstrated performance suggests that there is significant potential for applications to both sampling and optimization problems in the geosciences.", "In machine learning and statistics, probabilistic inference involving multimodal distributions is quite difficult. This is especially true in high dimensional problems, where most existing algorithms cannot easily move from one mode to another. To address this issue, we propose a novel Bayesian inference approach based on Markov Chain Monte Carlo. Our method can effectively sample from multimodal distributions, especially when the dimension is high and the modes are isolated. To this end, it exploits and modifies the Riemannian geometric properties of the target distribution to create wormholes connecting modes in order to facilitate moving between them. Further, our proposed method uses the regeneration technique in order to adapt the algorithm by identifying new modes and updating the network of wormholes without affecting the stationary distribution. To find new modes, as opposed to rediscovering those previously identified, we employ a novel mode searching algorithm that explores a residual energy function obtained by subtracting an approximate Gaussian mixture density (based on previously discovered modes) from the target density function.", "One of the main shortcomings of Markov chain Monte Carlo samplers is their inability to mix between modes of the target distribution. In this paper we show that advance knowledge of the location of these modes can be incorporated into the MCMC sampler by introducing mode-hopping moves that satisfy detailed balance. The proposed sampling algorithm explores local mode structure through local MCMC moves (e.g. diffusion or Hybrid Monte Carlo) but in addition also represents the relative strengths of the different modes correctly using a set of global moves. This “mode-hopping” MCMC sampler can be viewed as a generalization of the darting method [1].", "", "Markov chain Monte Carlo algorithms generate samples from a target distribution by simulating a Markov chain. Large flexibility exists in specification of transition matrix of the chain. In practice, however, most algorithms used only allow small changes in the state vector in each iteration. This choice typically causes problems for multimodal distributions as moves between modes become rare and, in turn, results in slow convergence to the target distribution. In this paper we consider continuous distributions on R n and specify how optimization for local maxima of the target distribution can be incorporated in the specification of the Markov chain. Thereby, we obtain a chain with frequent jumps between modes. We demonstrate the effectiveness of the approach in three examples. The first considers a simple mixture of bivariate normal distributions, whereas the two last examples consider sampling from posterior distributions based on previously analysed data sets.", "", "Several Markov chain methods are available for sampling from a posterior distribution. Two important examples are the Gibbs sampler and the Metropolis algorithm. In addition, several strategies are available for constructing hybrid algorithms. This paper outlines some of the basic methods and strategies and discusses some related theoretical and practical issues. On the theoretical side, results from the theory of general state space Markov chains can be used to obtain convergence rates, laws of large numbers and central limit theorems for estimates obtained from Markov chain methods. These theoretical results can be used to guide the construction of more efficient algorithms. For the practical use of Markov chain methods, standard simulation methodology provides several variance reduction techniques and also gives guidance on the choice of sample size and allocation", "A new Monte Carlo method is presented for simulations of systems with quenched random interactions. The approach greatly reduces the long correlation times characteristic of standard methods, allowing the investigation of lower temperatures with less computer time than previously necessary.", "Starting with the seminal paper of Haario, Saksman and Tamminen ( (2001)), a substantial amount of work has been done to validate adaptive Markov chain Monte Carlo algorithms. In this paper we focus on two practical aspects of adaptive Metropolis samplers. First, we draw attention to the deficient performance of standard adaptation when the target distribution is multi-modal. We propose a parallel chain adaptation strategy that incorporates multiple Markov chains which are run in parallel. Second, we note that", "The “Smart Walking” Monte Carlo algorithm is examined. In general, due to a bias imposed by the interbasin trial move, the algorithm does not satisfy detailed balance. While it has been shown that it can provide good estimates of equilibrium averages for certain potentials, for other potentials the estimates are poor. A modified version of the algorithm, Smart Darting Monte Carlo, which obeys the detailed balance condition, is proposed. Calculations on a one-dimensional model potential, on a Lennard-Jones cluster and on the alanine dipeptide demonstrate the accuracy and promise of the method for deeply quenched systems." ] }
1905.09964	2946364891	We introduce the Skipping Sampler, a novel algorithm to efficiently sample from the restriction of an arbitrary probability density to an arbitrary measurable set. Such conditional densities can arise in the study of risk and reliability and are often of complex nature, for example having multiple isolated modes and non-convex or disconnected support. The sampler can be seen as an instance of the Metropolis-Hastings algorithm with a particular proposal structure, and we establish sufficient conditions under which the Strong Law of Large Numbers and the Central Limit Theorem hold. We give theoretical and numerical evidence of improved performance relative to the Random Walk Metropolis algorithm.	Multipoint MCMC @cite_14 , based on Multiple-try Metropolis @cite_13 , does not require additional information about @math or @math . A fixed number of draws are made from a correlated proposal density and one is selected at random. The weight function used in this selection may be chosen to encourage exploration of @math , thus addressing multimodality. Its random-grid implementation, in particular, has similarities with the Skipping Sampler presented below. However our method does not fix the number of draws a priori but instead allows this to be guided by the unknown set @math , and it also lies in the MH class. The latter feature both simplifies the implementation and makes it easier to prove theoretical results such as the main results of this paper.	{ "cite_N": [ "@cite_14", "@cite_13" ], "mid": [ "1975941150", "2073412813" ], "abstract": [ "Abstract We propose the multipoint Metropolis algorithm as an extension of the orientational-bias Monte Carlo of Frenkel and Smit. A ratio statistics similar to that in the Metropolis algorithm is introduced to maintain the detailed balance. The multipoint idea can be applied to improve the efficiency of a general Markov chain-based Monte Carlo algorithm. To illustrate, we describe two variations of the idea—the random-grid Metropolis and the multipoint Hybrid Monte Carlo—and apply them to a number of examples.", "Abstract This article describes a new Metropolis-like transition rule, the multiple-try Metropolis, for Markov chain Monte Carlo (MCMC) simulations. By using this transition rule together with adaptive direction sampling, we propose a novel method for incorporating local optimization steps into a MCMC sampler in continuous state-space. Numerical studies show that the new method performs significantly better than the traditional Metropolis-Hastings (M-H) sampler. With minor tailoring in using the rule, the multiple-try method can also be exploited to achieve the effect of a griddy Gibbs sampler without having to bear with griddy approximations, and the effect of a hit-and-run algorithm without having to figure out the required conditional distribution in a random direction." ] }
1905.09964	2946364891	We introduce the Skipping Sampler, a novel algorithm to efficiently sample from the restriction of an arbitrary probability density to an arbitrary measurable set. Such conditional densities can arise in the study of risk and reliability and are often of complex nature, for example having multiple isolated modes and non-convex or disconnected support. The sampler can be seen as an instance of the Metropolis-Hastings algorithm with a particular proposal structure, and we establish sufficient conditions under which the Strong Law of Large Numbers and the Central Limit Theorem hold. We give theoretical and numerical evidence of improved performance relative to the Random Walk Metropolis algorithm.	Although designed for target densities of the form , the sampler introduced by the present authors in @cite_36 is different. It uses knowledge of the set C to add a single skip' of sufficient length to cross regions of zero density. It therefore does not address the challenge faced in the present work.	{ "cite_N": [ "@cite_36" ], "mid": [ "2963473861" ], "abstract": [ "The frequency stability of power systems is increasingly challenged by various types of disturbance. In particular, the increasing penetration of renewable energy sources is increasing the variability of power generation while reducing system inertia against disturbances. In this paper we explore how this could give rise to rate of change of frequency (RoCoF) violations. Correlated and non -Gaussian power disturbances, such as may arise from renewable generation, have been shown to be significant in power system security analysis. We therefore introduce ghost sampling which, given any unconditional distribution of disturbances, efficiently produces samples conditional on a violation occurring. Our goal is to address questions such as “which generator is most likely to be disconnected due to a RoCoF violation?” or “what is the probability of having simultaneous RoCoF violations, given that a violation occurs?”" ] }
1905.09964	2946364891	We introduce the Skipping Sampler, a novel algorithm to efficiently sample from the restriction of an arbitrary probability density to an arbitrary measurable set. Such conditional densities can arise in the study of risk and reliability and are often of complex nature, for example having multiple isolated modes and non-convex or disconnected support. The sampler can be seen as an instance of the Metropolis-Hastings algorithm with a particular proposal structure, and we establish sufficient conditions under which the Strong Law of Large Numbers and the Central Limit Theorem hold. We give theoretical and numerical evidence of improved performance relative to the Random Walk Metropolis algorithm.	Adaptive MCMC methods, in which parameters governing the evolution of the Markov chain can be learned during sampling @cite_24 @cite_45 , are an active current field of research addressing multimodality @cite_20 @cite_2 @cite_28 . Although adaptive algorithms can be more problematic to analyse, nevertheless adaptation could be added to the Skipping Sampler, for example in the underlying RWM proposal and the number of draws, and we reserve this for future work. Conversely, as an MH sampler, the Skipping Sampler can be plugged into existing sampling methods as appropriate, for instance the Gibbs Sampler @cite_18 @cite_9 and the Sequential Monte Carlo or Slice Samplers. An application to Slice Sampling is explored in Section .	{ "cite_N": [ "@cite_18", "@cite_28", "@cite_9", "@cite_24", "@cite_45", "@cite_2", "@cite_20" ], "mid": [ "2083875149", "2343666936", "1997063559", "1995780830", "2047978125", "2136796925", "2162340617" ], "abstract": [ "Abstract Stochastic substitution, the Gibbs sampler, and the sampling-importance-resampling algorithm can be viewed as three alternative sampling- (or Monte Carlo-) based approaches to the calculation of numerical estimates of marginal probability distributions. The three approaches will be reviewed, compared, and contrasted in relation to various joint probability structures frequently encountered in applications. In particular, the relevance of the approaches to calculating Bayesian posterior densities for a variety of structured models will be discussed and illustrated.", "We propose Adaptive Incremental Mixture Markov chain Monte Carlo (AIMM), a novel approach to sample from challenging probability distributions defined on a general state-space. While adaptive MCMC methods usually update a parametric proposal kernel with a global rule, AIMM locally adapts a semiparametric kernel. AIMM is based on an independent Metropolis-Hastings proposal distribution which takes the form of a finite mixture of Gaussian distributions. Central to this approach is the idea that the proposal distribution adapts to the target by locally adding a mixture component when the discrepancy between the proposal mixture and the target is deemed to be too large. As a result, the number of components in the mixture proposal is not fixed in advance. Theoretically, we prove that there exists a process that can be made arbitrarily close to AIMM and that converges to the correct target distribution. We also illustrate that it performs well in practice in a variety of challenging situations, including high-dimensional and multimodal target distributions.", "We make an analogy between images and statistical mechanics systems. Pixel gray levels and the presence and orientation of edges are viewed as states of atoms or molecules in a lattice-like physical system. The assignment of an energy function in the physical system determines its Gibbs distribution. Because of the Gibbs distribution, Markov random field (MRF) equivalence, this assignment also determines an MRF image model. The energy function is a more convenient and natural mechanism for embodying picture attributes than are the local characteristics of the MRF. For a range of degradation mechanisms, including blurring, non-linear deformations, and multiplicative or additive noise, the posterior distribution is an MRF with a structure akin to the image model. By the analogy, the posterior distribution defines another (imaginary) physical system. Gradual temperature reduction in the physical system isolates low-energy states (‘annealing’), or what is the same thing, the most probable states under the Gib...", "A proper choice of a proposal distribution for Markov chain Monte Carlo methods, for example for the Metropolis-Hastings algorithm, is well known to be a crucial factor for the convergence of the algorithm. In this paper we introduce an adaptive Metropolis (AM) algorithm, where the Gaussian proposal distribution is updated along the process using the full information cumulated so far. Due to the adaptive nature of the process, the AM algorithm is non-Markovian, but we establish here that it has the correct ergodic properties. We also include the results of our numerical tests, which indicate that the AM algorithm competes well with traditional Metropolis-Hastings algorithms, and demonstrate that the AM algorithm is easy to use in practical computation.", "We investigate the use of adaptive MCMC algorithms to automatically tune the Markov chain parameters during a run. Examples include the Adaptive Metropolis (AM) multivariate algorithm of Haario, Saksman, and Tamminen (2001), Metropolis-within-Gibbs algorithms for nonconjugate hierarchical models, regionally adjusted Metropolis algorithms, and logarithmic scalings. Computer simulations indicate that the algorithms perform very well compared to nonadaptive algorithms, even in high dimension.", "Several Markov chain methods are available for sampling from a posterior distribution. Two important examples are the Gibbs sampler and the Metropolis algorithm. In addition, several strategies are available for constructing hybrid algorithms. This paper outlines some of the basic methods and strategies and discusses some related theoretical and practical issues. On the theoretical side, results from the theory of general state space Markov chains can be used to obtain convergence rates, laws of large numbers and central limit theorems for estimates obtained from Markov chain methods. These theoretical results can be used to guide the construction of more efficient algorithms. For the practical use of Markov chain methods, standard simulation methodology provides several variance reduction techniques and also gives guidance on the choice of sample size and allocation", "Starting with the seminal paper of Haario, Saksman and Tamminen ( (2001)), a substantial amount of work has been done to validate adaptive Markov chain Monte Carlo algorithms. In this paper we focus on two practical aspects of adaptive Metropolis samplers. First, we draw attention to the deficient performance of standard adaptation when the target distribution is multi-modal. We propose a parallel chain adaptation strategy that incorporates multiple Markov chains which are run in parallel. Second, we note that" ] }
1905.10085	2951955299	Crowd counting from a single image is a challenging task due to high appearance similarity, perspective changes and severe congestion. Many methods only focus on the local appearance features and they cannot handle the aforementioned challenges. In order to tackle them, we propose a Perspective Crowd Counting Network (PCC Net), which consists of three parts: 1) Density Map Estimation (DME) focuses on learning very local features for density map estimation; 2) Random High-level Density Classification (R-HDC) extracts global features to predict the coarse density labels of random patches in images; 3) Fore- Background Segmentation (FBS) encodes mid-level features to segments the foreground and background. Besides, the DULR module is embedded in PCC Net to encode the perspective changes on four directions (Down, Up, Left and Right). The proposed PCC Net is verified on five mainstream datasets, which achieves the state-of-the-art performance on the one and attains the competitive results on the other four datasets. The source code is available at this https URL.	Since the patch-based methods cannot encode the global contextual information, some works @cite_0 @cite_37 @cite_31 @cite_3 focus on the whole image-based scheme. Zhang @cite_0 propose a Multi-column CNN to predict density map, which processes the input image with arbitrary size or resolution. Shang @cite_37 propose an end-to-end network that consists of CNN model and Long-short time memory (LSTM) decoder to predict the number of people. Sheng @cite_26 present a novel image representation which takes into consideration semantic attributes and spatial cues. Sindagi @cite_31 propose a cascaded CNNs framework to incorporate learning of a high-level prior to prompt the quality of the density map. Sindagi and Vishal @cite_3 propose a Contextual Pyramid CNN to generate the high-quality density map, which encodes the image at the patch and whole image level. To be specific, it contains three estimators: Global Context Estimator, Local Context Estimator and Density Map Estimator. The first two estimators generate the contextual information on the patches, and the last estimator outputs the density map of the whole image.	{ "cite_N": [ "@cite_37", "@cite_26", "@cite_3", "@cite_0", "@cite_31" ], "mid": [ "2514654788", "2343818649", "2963035940", "2463631526", "2962720716" ], "abstract": [ "Crowd counting is a very challenging task in crowded scenes due to heavy occlusions, appearance variations and perspective distortions. Current crowd counting methods typically operate on an image patch level with overlaps, then sum over the patches to get the final count. In this paper, we propose an end-to-end convolutional neural network (CNN) architecture that takes a whole image as its input and directly outputs the counting result. While making use of sharing computations over overlapping regions, our method takes advantages of contextual information when predicting both local and global count. In particular, we first feed the image to a pre-trained CNN to get a set of high level features. Then the features are mapped to local counting numbers using recurrent network layers with memory cells. We perform the experiments on several challenging crowd counting datasets, which achieve the state-of-the-art results and demonstrate the effectiveness of our method.", "Crowd counting is an important task in computer vision, which has many applications in video surveillance. Although the regression-based framework has achieved great improvements for crowd counting, how to improve the discriminative power of image representation is still an open problem. Conventional holistic features used in crowd counting often fail to capture semantic attributes and spatial cues of the image. In this paper, we propose integrating semantic information into learning locality-aware feature (LAF) sets for accurate crowd counting. First, with the help of a convolutional neural network, the original pixel space is mapped onto a dense attribute feature map, where each dimension of the pixelwise feature indicates the probabilistic strength of a certain semantic class. Then, LAF built on the idea of spatial pyramids on neighboring patches is proposed to explore more spatial context and local information. Finally, the traditional vector of locally aggregated descriptor (VLAD) encoding method is extended to a more generalized form weighted-VLAD (W-VLAD) in which diverse coefficient weights are taken into consideration. Experimental results validate the effectiveness of our presented method.", "We present a novel method called Contextual Pyramid CNN (CP-CNN) for generating high-quality crowd density and count estimation by explicitly incorporating global and local contextual information of crowd images. The proposed CP-CNN consists of four modules: Global Context Estimator (GCE), Local Context Estimator (LCE), Density Map Estimator (DME) and a Fusion-CNN (F-CNN). GCE is a VGG-16 based CNN that encodes global context and it is trained to classify input images into different density classes, whereas LCE is another CNN that encodes local context information and it is trained to perform patch-wise classification of input images into different density classes. DME is a multi-column architecture-based CNN that aims to generate high-dimensional feature maps from the input image which are fused with the contextual information estimated by GCE and LCE using F-CNN. To generate high resolution and high-quality density maps, F-CNN uses a set of convolutional and fractionally-strided convolutional layers and it is trained along with the DME in an end-to-end fashion using a combination of adversarial loss and pixellevel Euclidean loss. Extensive experiments on highly challenging datasets show that the proposed method achieves significant improvements over the state-of-the-art methods.", "This paper aims to develop a method than can accurately estimate the crowd count from an individual image with arbitrary crowd density and arbitrary perspective. To this end, we have proposed a simple but effective Multi-column Convolutional Neural Network (MCNN) architecture to map the image to its crowd density map. The proposed MCNN allows the input image to be of arbitrary size or resolution. By utilizing filters with receptive fields of different sizes, the features learned by each column CNN are adaptive to variations in people head size due to perspective effect or image resolution. Furthermore, the true density map is computed accurately based on geometry-adaptive kernels which do not need knowing the perspective map of the input image. Since exiting crowd counting datasets do not adequately cover all the challenging situations considered in our work, we have collected and labelled a large new dataset that includes 1198 images with about 330,000 heads annotated. On this challenging new dataset, as well as all existing datasets, we conduct extensive experiments to verify the effectiveness of the proposed model and method. In particular, with the proposed simple MCNN model, our method outperforms all existing methods. In addition, experiments show that our model, once trained on one dataset, can be readily transferred to a new dataset.", "Estimating crowd count in densely crowded scenes is an extremely challenging task due to non-uniform scale variations. In this paper, we propose a novel end-to-end cascaded network of CNNs to jointly learn crowd count classification and density map estimation. Classifying crowd count into various groups is tantamount to coarsely estimating the total count in the image thereby incorporating a high-level prior into the density estimation network. This enables the layers in the network to learn globally relevant discriminative features which aid in estimating highly refined density maps with lower count error. The joint training is performed in an end-to-end fashion. Extensive experiments on highly challenging publicly available datasets show that the proposed method achieves lower count error and better quality density maps as compared to the recent state-of-the-art methods." ] }
1905.10085	2951955299	Crowd counting from a single image is a challenging task due to high appearance similarity, perspective changes and severe congestion. Many methods only focus on the local appearance features and they cannot handle the aforementioned challenges. In order to tackle them, we propose a Perspective Crowd Counting Network (PCC Net), which consists of three parts: 1) Density Map Estimation (DME) focuses on learning very local features for density map estimation; 2) Random High-level Density Classification (R-HDC) extracts global features to predict the coarse density labels of random patches in images; 3) Fore- Background Segmentation (FBS) encodes mid-level features to segments the foreground and background. Besides, the DULR module is embedded in PCC Net to encode the perspective changes on four directions (Down, Up, Left and Right). The proposed PCC Net is verified on five mainstream datasets, which achieves the state-of-the-art performance on the one and attains the competitive results on the other four datasets. The source code is available at this https URL.	In 2015, Long @cite_33 propose a Fully Convolutional Networks (FCN) for semantic segmentation, which is a variant of traditional CNN. Gao @cite_4 propose a Siamesed FCN (s-FCN) to combine RGB and contour information for object segmentation. In order to encode the contextual and spatial information, some novel CNN architectures @cite_20 @cite_42 @cite_5 @cite_16 design the spatial propagation operation. Yu and Vladlen @cite_20 develop a dilated convolution to systematically aggregate multi-scale contextual information without losing resolution. Liu @cite_42 present a Spatial Propagation Networks (SPN) for learning the affinity matrix for vision tasks, which is a three-way connection for the linear propagation model. Pan @cite_5 design a slice-by-slice convolutional operation within feature maps, which propagates spatial information across rows and columns in a layer. Wang @cite_16 present a soft restricted Markov Random Field (MRF) to encode the spatial information in the urban scenes, which effectively remedy the over-smoothness phenomenon in second-order term of traditional MRF.	{ "cite_N": [ "@cite_4", "@cite_33", "@cite_42", "@cite_5", "@cite_16", "@cite_20" ], "mid": [ "2736675202", "1903029394", "2964242696", "2780740184", "2749439257", "2286929393" ], "abstract": [ "Road detection from the perspective of moving vehicles is a challenging issue in autonomous driving. Recently, many deep learning methods spring up for this task because they can extract high-level local features to find road regions from raw RGB data, such as Convolutional Neural Networks (CNN) and Fully Convolutional Networks (FCN). However, how to detect the boundary of road accurately is still an intractable problem. In this paper, we propose a siamesed fully convolutional network (named as “s-FCN-loc”) based on VGG-net architecture, which is able to consider RGB-channel, semantic contour and location prior simultaneously to segment road region elaborately. To be specific, the s-FCN-loc has two streams to process original RGB images and contour maps respectively. At the same time, the location prior is directly appended to the last feature map to promote the final detection performance. Experiments demonstrate that the proposed s-FCN-loc can learn more discriminative features of road boundaries and converge 30 faster than the original FCN during the training stage. Finally, the proposed approach is evaluated on KITTI road detection benchmark, and achieves a competitive result.", "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 [20], the VGG net [31], and GoogLeNet [32]) into fully convolutional networks and transfer their learned representations by fine-tuning [3] to the segmentation task. We then define a skip architecture that combines semantic information from a deep, coarse layer with appearance information from a shallow, fine layer to produce accurate and detailed segmentations. Our fully convolutional 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 less than one fifth of a second for a typical image.", "In this paper, we propose a spatial propagation networks for learning affinity matrix. We show that by constructing a row column linear propagation model, the spatially variant transformation matrix constitutes an affinity matrix that models dense, global pairwise similarities of an image. Specifically, we develop a three-way connection for the linear propagation model, which (a) formulates a sparse transformation matrix where all elements can be the output from a deep CNN, but (b) results in a dense affinity matrix that is effective to model any task-specific pairwise similarity. Instead of designing the similarity kernels according to image features of two points, we can directly output all similarities in a pure data-driven manner. The spatial propagation network is a generic framework that can be applied to numerous tasks, which traditionally benefit from designed affinity, e.g., image matting, colorization, and guided filtering, to name a few. Furthermore, the model can also learn semantic-aware affinity for high-level vision tasks due to the learning capability of the deep model. We validate the proposed framework by refinement of object segmentation. Experiments on the HELEN face parsing and PASCAL VOC-2012 semantic segmentation tasks show that the spatial propagation network provides general, effective and efficient solutions for generating high-quality segmentation results.", "Convolutional neural networks (CNNs) are usually built by stacking convolutional operations layer-by-layer. Although CNN has shown strong capability to extract semantics from raw pixels, its capacity to capture spatial relationships of pixels across rows and columns of an image is not fully explored. These relationships are important to learn semantic objects with strong shape priors but weak appearance coherences, such as traffic lanes, which are often occluded or not even painted on the road surface as shown in Fig. 1 (a). In this paper, we propose Spatial CNN (SCNN), which generalizes traditional deep layer-by-layer convolutions to slice-byslice convolutions within feature maps, thus enabling message passings between pixels across rows and columns in a layer. Such SCNN is particular suitable for long continuous shape structure or large objects, with strong spatial relationship but less appearance clues, such as traffic lanes, poles, and wall. We apply SCNN on a newly released very challenging traffic lane detection dataset and Cityscapse dataset. The results show that SCNN could learn the spatial relationship for structure output and significantly improves the performance. We show that SCNN outperforms the recurrent neural network (RNN) based ReNet and MRF+CNN (MRFNet) in the lane detection dataset by 8.7 and 4.6 respectively. Moreover, our SCNN won the 1st place on the TuSimple Benchmark Lane Detection Challenge, with an accuracy of 96.53 .", "Street scene understanding is an essential task for autonomous driving. One important step toward this direction is scene labeling, which annotates each pixel in the images with a correct class label. Although many approaches have been developed, there are still some weak points. First, many methods are based on the hand-crafted features whose image representation ability is limited. Second, they cannot label foreground objects accurately due to the data set bias. Third, in the refinement stage, the traditional Markov random filed inference is prone to over smoothness. For improving the above problems, this paper proposes a joint method of priori convolutional neural networks at superpixel level (called as “priori s-CNNs”) and soft restricted context transfer. Our contributions are threefold: 1) a priori s-CNNs model that learns priori location information at superpixel level is proposed to describe various objects discriminatingly; 2) a hierarchical data augmentation method is presented to alleviate data set bias in the priori s-CNNs training stage, which improves foreground objects labeling significantly; and 3) a soft restricted MRF energy function is defined to improve the priori s-CNNs model’s labeling performance and reduce the over smoothness at the same time. The proposed approach is verified on CamVid data set (11 classes) and SIFT Flow Street data set (16 classes) and achieves a competitive performance.", "State-of-the-art models for semantic segmentation are based on adaptations of convolutional networks that had originally been designed for image classification. However, dense prediction and image classification are structurally different. In this work, we develop a new convolutional network module that is specifically designed for dense prediction. The presented module uses dilated convolutions to systematically aggregate multi-scale contextual information without losing resolution. The architecture is based on the fact that dilated convolutions support exponential expansion of the receptive field without loss of resolution or coverage. We show that the presented context module increases the accuracy of state-of-the-art semantic segmentation systems. In addition, we examine the adaptation of image classification networks to dense prediction and show that simplifying the adapted network can increase accuracy." ] }
1905.09899	2944926927	Stochastic methods with coordinate-wise adaptive stepsize (such as RMSprop and Adam) have been widely used in training deep neural networks. Despite their fast convergence, they can generalize worse than stochastic gradient descent. In this paper, by revisiting the design of Adagrad, we propose to split the network parameters into blocks, and use a blockwise adaptive stepsize. Intuitively, blockwise adaptivity is less aggressive than adaptivity to individual coordinates, and can have a better balance between adaptivity and generalization. We show theoretically that the proposed blockwise adaptive gradient descent has comparable convergence rate as its counterpart with coordinate-wise adaptive stepsize, but is faster up to some constant. We also study its uniform stability and show that blockwise adaptivity can lead to lower generalization error than coordinate-wise adaptivity. Experimental results show that blockwise adaptive gradient descent converges faster and improves generalization performance over Nesterov's accelerated gradient and Adam.	Adagrad @cite_25 is the first adaptive gradient method in online convex learning with coordinate-wise stepsize. It is particularly useful for sparse learning, as parameters for the rare features can take large steps. Its stepsize schedule is competitive with the best coordinate-wise stepsize in hindsight @cite_20 . Recently, its convergence rate with a global adaptive stepsize in nonconvex optimization is established @cite_32 . It is shown that Adagrad converges to a stationary point at the optimal @math rate (up to a factor @math ), where @math is the total number of iterations.	{ "cite_N": [ "@cite_32", "@cite_25", "@cite_20" ], "mid": [ "2805767638", "2146502635", "2951395930" ], "abstract": [ "Adaptive gradient methods such as AdaGrad and its variants update the stepsize in stochastic gradient descent on the fly according to the gradients received along the way; such methods have gained widespread use in large-scale optimization for their ability to converge robustly, without the need to fine-tune the stepsize schedule. Yet, the theoretical guarantees to date for AdaGrad are for online and convex optimization. We bridge this gap by providing theoretical guarantees for the convergence of AdaGrad for smooth, nonconvex functions. We show that the norm version of AdaGrad (AdaGrad-Norm) converges to a stationary point at the @math rate in the stochastic setting, and at the optimal @math rate in the batch (non-stochastic) setting -- in this sense, our convergence guarantees are 'sharp'. In particular, the convergence of AdaGrad-Norm is robust to the choice of all hyper-parameters of the algorithm, in contrast to stochastic gradient descent whose convergence depends crucially on tuning the step-size to the (generally unknown) Lipschitz smoothness constant and level of stochastic noise on the gradient. Extensive numerical experiments are provided to corroborate our theory; moreover, the experiments suggest that the robustness of AdaGrad-Norm extends to state-of-the-art models in deep learning, without sacrificing generalization.", "We present a new family of subgradient methods that dynamically incorporate knowledge of the geometry of the data observed in earlier iterations to perform more informative gradient-based learning. Metaphorically, the adaptation allows us to find needles in haystacks in the form of very predictive but rarely seen features. Our paradigm stems from recent advances in stochastic optimization and online learning which employ proximal functions to control the gradient steps of the algorithm. We describe and analyze an apparatus for adaptively modifying the proximal function, which significantly simplifies setting a learning rate and results in regret guarantees that are provably as good as the best proximal function that can be chosen in hindsight. We give several efficient algorithms for empirical risk minimization problems with common and important regularization functions and domain constraints. We experimentally study our theoretical analysis and show that adaptive subgradient methods outperform state-of-the-art, yet non-adaptive, subgradient algorithms.", "We introduce a new online convex optimization algorithm that adaptively chooses its regularization function based on the loss functions observed so far. This is in contrast to previous algorithms that use a fixed regularization function such as L2-squared, and modify it only via a single time-dependent parameter. Our algorithm's regret bounds are worst-case optimal, and for certain realistic classes of loss functions they are much better than existing bounds. These bounds are problem-dependent, which means they can exploit the structure of the actual problem instance. Critically, however, our algorithm does not need to know this structure in advance. Rather, we prove competitive guarantees that show the algorithm provides a bound within a constant factor of the best possible bound (of a certain functional form) in hindsight." ] }
1905.09899	2944926927	Stochastic methods with coordinate-wise adaptive stepsize (such as RMSprop and Adam) have been widely used in training deep neural networks. Despite their fast convergence, they can generalize worse than stochastic gradient descent. In this paper, by revisiting the design of Adagrad, we propose to split the network parameters into blocks, and use a blockwise adaptive stepsize. Intuitively, blockwise adaptivity is less aggressive than adaptivity to individual coordinates, and can have a better balance between adaptivity and generalization. We show theoretically that the proposed blockwise adaptive gradient descent has comparable convergence rate as its counterpart with coordinate-wise adaptive stepsize, but is faster up to some constant. We also study its uniform stability and show that blockwise adaptivity can lead to lower generalization error than coordinate-wise adaptivity. Experimental results show that blockwise adaptive gradient descent converges faster and improves generalization performance over Nesterov's accelerated gradient and Adam.	Recall that the SGD iterate is the solution to the problem: @math , where @math is the gradient of the loss function @math at iteration @math , and @math is the parameter vector. To incorporate information about the curvature of sequence @math , the @math -norm in the SGD update can be replaced by the Mahalanobis norm, leading to @cite_25 : where @math . This is an instance of mirror descent @cite_30 . Its regret bound has a gradient-related term @math . Adagrad's stepsize can be obtained by examining a similar objective @cite_25 : where @math , and @math is some constant. At optimality, @math , where @math . As @math cannot depend on @math 's with @math , this suggests @math . Theoretically, this choice of @math leads to a regret bound that is competitive with the best post-hoc optimal bound @cite_20 .	{ "cite_N": [ "@cite_30", "@cite_25", "@cite_20" ], "mid": [ "1505731132", "2146502635", "2951395930" ], "abstract": [ "", "We present a new family of subgradient methods that dynamically incorporate knowledge of the geometry of the data observed in earlier iterations to perform more informative gradient-based learning. Metaphorically, the adaptation allows us to find needles in haystacks in the form of very predictive but rarely seen features. Our paradigm stems from recent advances in stochastic optimization and online learning which employ proximal functions to control the gradient steps of the algorithm. We describe and analyze an apparatus for adaptively modifying the proximal function, which significantly simplifies setting a learning rate and results in regret guarantees that are provably as good as the best proximal function that can be chosen in hindsight. We give several efficient algorithms for empirical risk minimization problems with common and important regularization functions and domain constraints. We experimentally study our theoretical analysis and show that adaptive subgradient methods outperform state-of-the-art, yet non-adaptive, subgradient algorithms.", "We introduce a new online convex optimization algorithm that adaptively chooses its regularization function based on the loss functions observed so far. This is in contrast to previous algorithms that use a fixed regularization function such as L2-squared, and modify it only via a single time-dependent parameter. Our algorithm's regret bounds are worst-case optimal, and for certain realistic classes of loss functions they are much better than existing bounds. These bounds are problem-dependent, which means they can exploit the structure of the actual problem instance. Critically, however, our algorithm does not need to know this structure in advance. Rather, we prove competitive guarantees that show the algorithm provides a bound within a constant factor of the best possible bound (of a certain functional form) in hindsight." ] }
1905.09899	2944926927	Stochastic methods with coordinate-wise adaptive stepsize (such as RMSprop and Adam) have been widely used in training deep neural networks. Despite their fast convergence, they can generalize worse than stochastic gradient descent. In this paper, by revisiting the design of Adagrad, we propose to split the network parameters into blocks, and use a blockwise adaptive stepsize. Intuitively, blockwise adaptivity is less aggressive than adaptivity to individual coordinates, and can have a better balance between adaptivity and generalization. We show theoretically that the proposed blockwise adaptive gradient descent has comparable convergence rate as its counterpart with coordinate-wise adaptive stepsize, but is faster up to some constant. We also study its uniform stability and show that blockwise adaptivity can lead to lower generalization error than coordinate-wise adaptivity. Experimental results show that blockwise adaptive gradient descent converges faster and improves generalization performance over Nesterov's accelerated gradient and Adam.	To solve the expected risk minimization problem in ), an Adagrad variant called weighted AdaEMA is recently proposed in @cite_38 . It employs weighted averaging of @math 's for stepsize and momentum acceleration. This is a general coordinate-wise adaptive method and includes many Adagrad variants as special cases, including Adam and RMSprop.	{ "cite_N": [ "@cite_38" ], "mid": [ "2900459988" ], "abstract": [ "Adam and RMSProp are two of the most influential adaptive stochastic algorithms for training deep neural networks, which have been pointed out to be divergent even in the convex setting via a few simple counterexamples. Many attempts, such as decreasing an adaptive learning rate, adopting a big batch size, incorporating a temporal decorrelation technique, seeking an analogous surrogate, etc., have been tried to promote Adam RMSProp-type algorithms to converge. In contrast with existing approaches, we introduce an alternative easy-to-check sufficient condition, which merely depends on the parameters of the base learning rate and combinations of historical second-order moments, to guarantee the global convergence of generic Adam RMSProp for solving large-scale non-convex stochastic optimization. Moreover, we show that the convergences of several variants of Adam, such as AdamNC, AdaEMA, etc., can be directly implied via the proposed sufficient condition in the non-convex setting. In addition, we illustrate that Adam is essentially a specifically weighted AdaGrad with exponential moving average momentum, which provides a novel perspective for understanding Adam and RMSProp. This observation coupled with this sufficient condition gives much deeper interpretations on their divergences. At last, we validate the sufficient condition by applying Adam and RMSProp to tackle a certain counterexample and train deep neural networks. Numerical results are exactly in accord with our theoretical analysis." ] }
1905.09979	2944942767	Ensembling is a universally useful approach to boost the performance of machine learning models. However, individual models in an ensemble are typically trained independently in separate stages, without information access about the overall ensemble. In this paper, model ensembles are treated as first-class citizens, and their performance is optimized end-to-end with parameter sharing and a novel loss structure that improves generalization. On large-scale datasets including ImageNet, Youtube-8M, and Kinetics, we demonstrate a procedure that starts from a strongly performing single deep neural network, and constructs an EnsembleNet that has both a smaller size and better performance. Moreover, an EnsembleNet can be trained in one stage just like a single model without manual intervention.	Sharing a common base structure among multiple individual models may produce better ensembles, and this technique is used in @cite_27 @cite_15 . Furthermore, hierarchical sharing may give additional performance boost @cite_9 .	{ "cite_N": [ "@cite_9", "@cite_27", "@cite_15" ], "mid": [ "2963373431", "2172734211", "2807912816" ], "abstract": [ "We introduce collaborative learning in which multiple classifier heads of the same network are simultaneously trained on the same training data to improve generalization and robustness to label noise with no extra inference cost. It acquires the strengths from auxiliary training, multi-task learning and knowledge distillation. There are two important mechanisms involved in collaborative learning. First, the consensus of multiple views from different classifier heads on the same example provides supplementary information as well as regularization to each classifier, thereby improving generalization. Second, intermediate-level representation (ILR) sharing with backpropagation rescaling aggregates the gradient flows from all heads, which not only reduces training computational complexity, but also facilitates supervision to the shared layers. The empirical results on CIFAR and ImageNet datasets demonstrate that deep neural networks learned as a group in a collaborative way significantly reduce the generalization error and increase the robustness to label noise.", "Convolutional Neural Networks have achieved state-of-the-art performance on a wide range of tasks. Most benchmarks are led by ensembles of these powerful learners, but ensembling is typically treated as a post-hoc procedure implemented by averaging independently trained models with model variation induced by bagging or random initialization. In this paper, we rigorously treat ensembling as a first-class problem to explicitly address the question: what are the best strategies to create an ensemble? We first compare a large number of ensembling strategies, and then propose and evaluate novel strategies, such as parameter sharing (through a new family of models we call TreeNets) as well as training under ensemble-aware and diversity-encouraging losses. We demonstrate that TreeNets can improve ensemble performance and that diverse ensembles can be trained end-to-end under a unified loss, achieving significantly higher \"oracle\" accuracies than classical ensembles.", "Knowledge distillation is effective to train small and generalisable network models for meeting the low-memory and fast running requirements. Existing offline distillation methods rely on a strong pre-trained teacher, which enables favourable knowledge discovery and transfer but requires a complex two-phase training procedure. Online counterparts address this limitation at the price of lacking a highcapacity teacher. In this work, we present an On-the-fly Native Ensemble (ONE) strategy for one-stage online distillation. Specifically, ONE trains only a single multi-branch network while simultaneously establishing a strong teacher on-the- fly to enhance the learning of target network. Extensive evaluations show that ONE improves the generalisation performance a variety of deep neural networks more significantly than alternative methods on four image classification dataset: CIFAR10, CIFAR100, SVHN, and ImageNet, whilst having the computational efficiency advantages." ] }
1905.09979	2944942767	Ensembling is a universally useful approach to boost the performance of machine learning models. However, individual models in an ensemble are typically trained independently in separate stages, without information access about the overall ensemble. In this paper, model ensembles are treated as first-class citizens, and their performance is optimized end-to-end with parameter sharing and a novel loss structure that improves generalization. On large-scale datasets including ImageNet, Youtube-8M, and Kinetics, we demonstrate a procedure that starts from a strongly performing single deep neural network, and constructs an EnsembleNet that has both a smaller size and better performance. Moreover, an EnsembleNet can be trained in one stage just like a single model without manual intervention.	In this work, we focus on a training strategy where individual models are aware of the ensembler during the optimization. A simple approach is to add a loss from the ensembler prediction. Another popular approach is co-distillation @cite_16 @cite_30 @cite_15 @cite_9 , where constituent models are encouraged to learn from each other by regressing their predictions to the ensembler prediction. We will show a deep connection between these two approaches.	{ "cite_N": [ "@cite_30", "@cite_9", "@cite_15", "@cite_16" ], "mid": [ "2787017828", "2963373431", "2807912816", "2620998106" ], "abstract": [ "Techniques such as ensembling and distillation promise model quality improvements when paired with almost any base model. However, due to increased test-time cost (for ensembles) and increased complexity of the training pipeline (for distillation), these techniques are challenging to use in industrial settings. In this paper we explore a variant of distillation which is relatively straightforward to use as it does not require a complicated multi-stage setup or many new hyperparameters. Our first claim is that online distillation enables us to use extra parallelism to fit very large datasets about twice as fast. Crucially, we can still speed up training even after we have already reached the point at which additional parallelism provides no benefit for synchronous or asynchronous stochastic gradient descent. Two neural networks trained on disjoint subsets of the data can share knowledge by encouraging each model to agree with the predictions the other model would have made. These predictions can come from a stale version of the other model so they can be safely computed using weights that only rarely get transmitted. Our second claim is that online distillation is a cost-effective way to make the exact predictions of a model dramatically more reproducible. We support our claims using experiments on the Criteo Display Ad Challenge dataset, ImageNet, and the largest to-date dataset used for neural language modeling, containing @math tokens and based on the Common Crawl repository of web data.", "We introduce collaborative learning in which multiple classifier heads of the same network are simultaneously trained on the same training data to improve generalization and robustness to label noise with no extra inference cost. It acquires the strengths from auxiliary training, multi-task learning and knowledge distillation. There are two important mechanisms involved in collaborative learning. First, the consensus of multiple views from different classifier heads on the same example provides supplementary information as well as regularization to each classifier, thereby improving generalization. Second, intermediate-level representation (ILR) sharing with backpropagation rescaling aggregates the gradient flows from all heads, which not only reduces training computational complexity, but also facilitates supervision to the shared layers. The empirical results on CIFAR and ImageNet datasets demonstrate that deep neural networks learned as a group in a collaborative way significantly reduce the generalization error and increase the robustness to label noise.", "Knowledge distillation is effective to train small and generalisable network models for meeting the low-memory and fast running requirements. Existing offline distillation methods rely on a strong pre-trained teacher, which enables favourable knowledge discovery and transfer but requires a complex two-phase training procedure. Online counterparts address this limitation at the price of lacking a highcapacity teacher. In this work, we present an On-the-fly Native Ensemble (ONE) strategy for one-stage online distillation. Specifically, ONE trains only a single multi-branch network while simultaneously establishing a strong teacher on-the- fly to enhance the learning of target network. Extensive evaluations show that ONE improves the generalisation performance a variety of deep neural networks more significantly than alternative methods on four image classification dataset: CIFAR10, CIFAR100, SVHN, and ImageNet, whilst having the computational efficiency advantages.", "Model distillation is an effective and widely used technique to transfer knowledge from a teacher to a student network. The typical application is to transfer from a powerful large network or ensemble to a small network, in order to meet the low-memory or fast execution requirements. In this paper, we present a deep mutual learning (DML) strategy. Different from the one-way transfer between a static pre-defined teacher and a student in model distillation, with DML, an ensemble of students learn collaboratively and teach each other throughout the training process. Our experiments show that a variety of network architectures benefit from mutual learning and achieve compelling results on both category and instance recognition tasks. Surprisingly, it is revealed that no prior powerful teacher network is necessary - mutual learning of a collection of simple student networks works, and moreover outperforms distillation from a more powerful yet static teacher." ] }
1905.10100	2946007935	Recently, fully convolutional neural networks (FCNs) have shown significant performance in image parsing, including scene parsing and object parsing. Different from generic object parsing tasks, hand parsing is more challenging due to small size, complex structure, heavy self-occlusion and ambiguous texture problems. In this paper, we propose a novel parsing framework, Multi-Scale Dual-Branch Fully Convolutional Network (MSDB-FCN), for hand parsing tasks. Our network employs a Dual-Branch architecture to extract features of hand area, paying attention on the hand itself. These features are used to generate multi-scale features with pyramid pooling strategy. In order to better encode multi-scale features, we design a Deconvolution and Bilinear Interpolation Block (DB-Block) for upsampling and merging the features of different scales. To address data imbalance, which is a common problem in many computer vision tasks as well as hand parsing tasks, we propose a generalization of Focal Loss, namely Multi-Class Balanced Focal Loss, to tackle data imbalance in multi-class classification. Extensive experiments on RHD-PARSING dataset demonstrate that our MSDB-FCN has achieved the state-of-the-art performance for hand parsing.	Object Extraction: Driven by the development of Convolutional Neural Networks (CNNs), researches on many computer vision tasks, such as object detection, semantic segmentation and pose estimation, have achieved remarkable success. For some specific tasks, object extraction may be required. Methods of @cite_24 @cite_22 combined object detection network for accurate segmentation. @cite_3 , a network for segmentation was employed to extract the region of a hand. Yet these approaches did not make full use of the information extracted by the network, causing a large number of redundant calculations. Our MSDB-FCN employs a mask branch to calculate the position of a hand area, providing prior information to the parsing branch. The mask branch is a small subnet embedded in the network, which avoids repeated computation of feature extraction. Compared with the aforementioned methods, the structure of our model is more complete.	{ "cite_N": [ "@cite_24", "@cite_22", "@cite_3" ], "mid": [ "2962820340", "2963773230", "2962811204" ], "abstract": [ "", "The problem of video object segmentation can become extremely challenging when multiple instances co-exist. While each instance may exhibit large scale and pose variations, the problem is compounded when instances occlude each other causing failures in tracking. In this study, we formulate a deep recurrent network that is capable of segmenting and tracking objects in video simultaneously by their temporal continuity, yet able to re-identify them when they re-appear after a prolonged occlusion. We combine temporal propagation and re-identification functionalities into a single framework that can be trained end-to-end. In particular, we present a re-identification module with template expansion to retrieve missing objects despite their large appearance changes. In addition, we contribute an attention-based recurrent mask propagation approach that is robust to distractors not belonging to the target segment. Our approach achieves a new state-of-the-art ( G )-mean of 68.2 on the challenging DAVIS 2017 benchmark (test-dev set), outperforming the winning solution. Project Page: http: mmlab.ie.cuhk.edu.hk projects DyeNet .", "Low-cost consumer depth cameras and deep learning have enabled reasonable 3D hand pose estimation from single depth images. In this paper, we present an approach that estimates 3D hand pose from regular RGB images. This task has far more ambiguities due to the missing depth information. To this end, we propose a deep network that learns a network-implicit 3D articulation prior. Together with detected keypoints in the images, this network yields good estimates of the 3D pose. We introduce a large scale 3D hand pose dataset based on synthetic hand models for training the involved networks. Experiments on a variety of test sets, including one on sign language recognition, demonstrate the feasibility of 3D hand pose estimation on single color images." ] }
1905.10100	2946007935	Recently, fully convolutional neural networks (FCNs) have shown significant performance in image parsing, including scene parsing and object parsing. Different from generic object parsing tasks, hand parsing is more challenging due to small size, complex structure, heavy self-occlusion and ambiguous texture problems. In this paper, we propose a novel parsing framework, Multi-Scale Dual-Branch Fully Convolutional Network (MSDB-FCN), for hand parsing tasks. Our network employs a Dual-Branch architecture to extract features of hand area, paying attention on the hand itself. These features are used to generate multi-scale features with pyramid pooling strategy. In order to better encode multi-scale features, we design a Deconvolution and Bilinear Interpolation Block (DB-Block) for upsampling and merging the features of different scales. To address data imbalance, which is a common problem in many computer vision tasks as well as hand parsing tasks, we propose a generalization of Focal Loss, namely Multi-Class Balanced Focal Loss, to tackle data imbalance in multi-class classification. Extensive experiments on RHD-PARSING dataset demonstrate that our MSDB-FCN has achieved the state-of-the-art performance for hand parsing.	Multi-Scale Features Aggregation: Multi-scale features have been shown to significantly improve the performance of networks to perform many computer vision tasks. A number of researches recently focus on investigating methods to obtain and encode multi-scale features. One way is to use an encoder-decoder structure. Chen @math @cite_25 used a GlobalNet to generate multi-scale features and a RefineNet to encode them. Lin @math @cite_7 proposed a method of feature fusion in pairs. Lin @math @cite_16 designed a multi-path RefineNet network to generate high-resolution semantic feature maps. Another way is to employ multi-scale pooling strategies, such as pyramid pooling. By controlling the pool size, we can gain features of different scales. He @math @cite_10 proposed spatial pyramid pooling for visual recognition. Zhao @math @cite_31 extended the approach and designed PSPNet for scene parsing task. Yang @math @cite_18 adjusted the pool size and designed a bidirectional structure to aggregate multi-scale features. Our work considers the difference of hand parsing from scene parsing. We then choose suitable pooling sizes and design a novel structure for upsampling and feature coding. Our model achieves better performance than using either deconvolution or bilinear interpolation alone.	{ "cite_N": [ "@cite_18", "@cite_7", "@cite_31", "@cite_16", "@cite_10", "@cite_25" ], "mid": [ "", "2895420332", "2560023338", "", "2179352600", "2964221239" ], "abstract": [ "", "Accurate semantic image segmentation requires the joint consideration of local appearance, semantic information, and global scene context. In today’s age of pre-trained deep networks and their powerful convolutional features, state-of-the-art semantic segmentation approaches differ mostly in how they choose to combine together these different kinds of information. In this work, we propose a novel scheme for aggregating features from different scales, which we refer to as Multi-Scale Context Intertwining (MSCI). In contrast to previous approaches, which typically propagate information between scales in a one-directional manner, we merge pairs of feature maps in a bidirectional and recurrent fashion, via connections between two LSTM chains. By training the parameters of the LSTM units on the segmentation task, the above approach learns how to extract powerful and effective features for pixel-level semantic segmentation, which are then combined hierarchically. Furthermore, rather than using fixed information propagation routes, we subdivide images into super-pixels, and use the spatial relationship between them in order to perform image-adapted context aggregation. Our extensive evaluation on public benchmarks indicates that all of the aforementioned components of our approach increase the effectiveness of information propagation throughout the network, and significantly improve its eventual segmentation accuracy.", "Scene parsing is challenging for unrestricted open vocabulary and diverse scenes. In this paper, we exploit the capability of global context information by different-region-based context aggregation through our pyramid pooling module together with the proposed pyramid scene parsing network (PSPNet). Our global prior representation is effective to produce good quality results on the scene parsing task, while PSPNet provides a superior framework for pixel-level prediction. The proposed approach achieves state-of-the-art performance on various datasets. It came first in ImageNet scene parsing challenge 2016, PASCAL VOC 2012 benchmark and Cityscapes benchmark. A single PSPNet yields the new record of mIoU accuracy 85.4 on PASCAL VOC 2012 and accuracy 80.2 on Cityscapes.", "", "Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g. 224×224) input image. This requirement is “artificial” and may hurt the recognition accuracy for the images or sub-images of an arbitrary size scale. In this work, we equip the networks with a more principled pooling strategy, “spatial pyramid pooling”, to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size scale. By removing the fixed-size limitation, we can improve all CNN-based image classification methods in general. Our SPP-net achieves state-of-the-art accuracy on the datasets of ImageNet 2012, Pascal VOC 2007, and Caltech101.", "The topic of multi-person pose estimation has been largely improved recently, especially with the development of convolutional neural network. However, there still exist a lot of challenging cases, such as occluded keypoints, invisible keypoints and complex background, which cannot be well addressed. In this paper, we present a novel network structure called Cascaded Pyramid Network (CPN) which targets to relieve the problem from these \"hard\" keypoints. More specifically, our algorithm includes two stages: GlobalNet and RefineNet. GlobalNet is a feature pyramid network which can successfully localize the \"simple\" keypoints like eyes and hands but may fail to precisely recognize the occluded or invisible keypoints. Our RefineNet tries explicitly handling the \"hard\" keypoints by integrating all levels of feature representations from the GlobalNet together with an online hard keypoint mining loss. In general, to address the multi-person pose estimation problem, a top-down pipeline is adopted to first generate a set of human bounding boxes based on a detector, followed by our CPN for keypoint localization in each human bounding box. Based on the proposed algorithm, we achieve state-of-art results on the COCO keypoint benchmark, with average precision at 73.0 on the COCO test-dev dataset and 72.1 on the COCO test-challenge dataset, which is a 19 relative improvement compared with 60.5 from the COCO 2016 keypoint challenge. Code1 and the detection results for person used will be publicly available for further research." ] }
1905.10100	2946007935	Recently, fully convolutional neural networks (FCNs) have shown significant performance in image parsing, including scene parsing and object parsing. Different from generic object parsing tasks, hand parsing is more challenging due to small size, complex structure, heavy self-occlusion and ambiguous texture problems. In this paper, we propose a novel parsing framework, Multi-Scale Dual-Branch Fully Convolutional Network (MSDB-FCN), for hand parsing tasks. Our network employs a Dual-Branch architecture to extract features of hand area, paying attention on the hand itself. These features are used to generate multi-scale features with pyramid pooling strategy. In order to better encode multi-scale features, we design a Deconvolution and Bilinear Interpolation Block (DB-Block) for upsampling and merging the features of different scales. To address data imbalance, which is a common problem in many computer vision tasks as well as hand parsing tasks, we propose a generalization of Focal Loss, namely Multi-Class Balanced Focal Loss, to tackle data imbalance in multi-class classification. Extensive experiments on RHD-PARSING dataset demonstrate that our MSDB-FCN has achieved the state-of-the-art performance for hand parsing.	Loss Function for Data Imbalance: Problems of data imbalance are often encountered in computer vision tasks. In order to balance different samples, Shrivastava @math @cite_21 proposed Online Hard Example Mining (OHEM), but it only handled on easy samples. Class Balanced Loss solved class imbalance by calculating the loss of positive and negative samples separately and using weights to balanced them. Focal Loss proposed by He @math @cite_15 has been shown great success in object detection, which balances the weights of different classes and samples. However, Focal Loss is only effective in binary classification. In multi-class classification, such as image parsing or semantic segmentation, the parameter @math used in Focal Loss to solving class imbalance becomes ineffective. In this paper, we propose a Multi-Class Balanced Focal Loss to enable multi-class classification, overcoming the limitation of Focal Loss.	{ "cite_N": [ "@cite_15", "@cite_21" ], "mid": [ "2963351448", "2963516811" ], "abstract": [ "The highest accuracy object detectors to date are based on a two-stage approach popularized by R-CNN, where a classifier is applied to a sparse set of candidate object locations. In contrast, one-stage detectors that are applied over a regular, dense sampling of possible object locations have the potential to be faster and simpler, but have trailed the accuracy of two-stage detectors thus far. In this paper, we investigate why this is the case. We discover that the extreme foreground-background class imbalance encountered during training of dense detectors is the central cause. We propose to address this class imbalance by reshaping the standard cross entropy loss such that it down-weights the loss assigned to well-classified examples. Our novel Focal Loss focuses training on a sparse set of hard examples and prevents the vast number of easy negatives from overwhelming the detector during training. To evaluate the effectiveness of our loss, we design and train a simple dense detector we call RetinaNet. Our results show that when trained with the focal loss, RetinaNet is able to match the speed of previous one-stage detectors while surpassing the accuracy of all existing state-of-the-art two-stage detectors.", "The field of object detection has made significant advances riding on the wave of region-based ConvNets, but their training procedure still includes many heuristics and hyperparameters that are costly to tune. We present a simple yet surprisingly effective online hard example mining (OHEM) algorithm for training region-based ConvNet detectors. Our motivation is the same as it has always been – detection datasets contain an overwhelming number of easy examples and a small number of hard examples. Automatic selection of these hard examples can make training more effective and efficient. OHEM is a simple and intuitive algorithm that eliminates several heuristics and hyperparameters in common use. But more importantly, it yields consistent and significant boosts in detection performance on benchmarks like PASCAL VOC 2007 and 2012. Its effectiveness increases as datasets become larger and more difficult, as demonstrated by the results on the MS COCO dataset. Moreover, combined with complementary advances in the field, OHEM leads to state-of-the-art results of 78.9 and 76.3 mAP on PASCAL VOC 2007 and 2012 respectively." ] }
1905.09989	2944928881	Learning Mahalanobis metric spaces is an important problem that has found numerous applications. Several algorithms have been designed for this problem, including Information Theoretic Metric Learning (ITML) by [ 2007] and Large Margin Nearest Neighbor (LMNN) classification by [Weinberger and Saul 2009]. We consider a formulation of Mahalanobis metric learning as an optimization problem, where the objective is to minimize the number of violated similarity dissimilarity constraints. We show that for any fixed ambient dimension, there exists a fully polynomial-time approximation scheme (FPTAS) with nearlylinear running time. This result is obtained using tools from the theory of linear programming in low dimensions. We also discuss improvements of the algorithm in practice, and present experimental results on synthetic and real-world data sets.	Several algorithms for learning Mahalanobis metric spaces have been proposed. Notable examples include the SDP based algorithm of @cite_14 , the algorithm of Globerson and Roweis for the fully supervised setting @cite_13 , Information Theoretic Metric Learning ( ) by @cite_2 , which casts the problem as a particular optimization minimizing LogDet divergence, as well as Large Margin Nearest Neighbor ( ) by @cite_10 , which attempts to learn a metric geared towards optimizing @math -NN classification. We refer the reader to the surveys @cite_11 @cite_4 for a detailed discussion of previous work. Our algorithm differs from previous approaches in that it seeks to directly minimize the number of violated pairwise distance constraints, which is a highly non-convex objective, without resorting to a convex relaxation of the corresponding optimization problem.	{ "cite_N": [ "@cite_13", "@cite_14", "@cite_4", "@cite_2", "@cite_10", "@cite_11" ], "mid": [ "2104752854", "2117154949", "2808979505", "", "2130556178", "2121949863" ], "abstract": [ "We present an algorithm for learning a quadratic Gaussian metric (Mahalanobis distance) for use in classification tasks. Our method relies on the simple geometric intuition that a good metric is one under which points in the same class are simultaneously near each other and far from points in the other classes. We construct a convex optimization problem whose solution generates such a metric by trying to collapse all examples in the same class to a single point and push examples in other classes infinitely far away. We show that when the metric we learn is used in simple classifiers, it yields substantial improvements over standard alternatives on a variety of problems. We also discuss how the learned metric may be used to obtain a compact low dimensional feature representation of the original input space, allowing more efficient classification with very little reduction in performance.", "Many algorithms rely critically on being given a good metric over their inputs. For instance, data can often be clustered in many \"plausible\" ways, and if a clustering algorithm such as K-means initially fails to find one that is meaningful to a user, the only recourse may be for the user to manually tweak the metric until sufficiently good clusters are found. For these and other applications requiring good metrics, it is desirable that we provide a more systematic way for users to indicate what they consider \"similar.\" For instance, we may ask them to provide examples. In this paper, we present an algorithm that, given examples of similar (and, if desired, dissimilar) pairs of points in ℝn, learns a distance metric over ℝn that respects these relationships. Our method is based on posing metric learning as a convex optimization problem, which allows us to give efficient, local-optima-free algorithms. We also demonstrate empirically that the learned metrics can be used to significantly improve clustering performance.", "Abstract Distance metric learning has been a hot research spot recently due to its high effectiveness and efficiency in improving the performance of distance related methods, such as k nearest neighbors ( k NN). Metric learning aims to learn a data-dependent metric to make intra-class distance smaller and inter-class larger. A large number of methods have been proposed for various applications and a survey to evaluate and compare these methods is imperative. The existing surveys just analyze the algorithms theoretically or compare them experimentally with a narrow time scope. Therefore, the paper reviews classical and influential methods that were proposed between 2003 and 2017 and presents a taxonomy based on the most distinct character of each method. All the methods are categorized into five classes, including pairwise cost, probabilistic framework, boost-like approaches, advantageous variants and specific applications. A comprehensive experimental study is made to compare all the selected methods, exploring the ability in improving accuracy, the relation between distance change and accuracy, the relation between accuracy and k NN neighbor size.", "", "We show how to learn a Mahanalobis distance metric for k-nearest neighbor (kNN) classification by semidefinite programming. The metric is trained with the goal that the k-nearest neighbors always belong to the same class while examples from different classes are separated by a large margin. On seven data sets of varying size and difficulty, we find that metrics trained in this way lead to significant improvements in kNN classification—for example, achieving a test error rate of 1.3 on the MNIST handwritten digits. As in support vector machines (SVMs), the learning problem reduces to a convex optimization based on the hinge loss. Unlike learning in SVMs, however, our framework requires no modification or extension for problems in multiway (as opposed to binary) classification.", "The metric learning problem is concerned with learning a distance function tuned to a particular task, and has been shown to be useful when used in conjunction with nearest-neighbor methods and other techniques that rely on distances or similarities. This survey presents an overview of existing research in metric learning, including recent progress on scaling to high-dimensional feature spaces and to data sets with an extremely large number of data points. A goal of the survey is to present as unified as possible a framework under which existing research on metric learning can be cast. The first part of the survey focuses on linear metric learning approaches, mainly concentrating on the class of Mahalanobis distance learning methods. We then discuss nonlinear metric learning approaches, focusing on the connections between the nonlinear and linear approaches. Finally, we discuss extensions of metric learning, as well as applications to a variety of problems in computer vision, text analysis, program analysis, and multimedia. Full text available at: http: dx.doi.org 10.1561 2200000019" ] }
1905.10108	2946154049	The minimization of loss functions is the heart and soul of Machine Learning. In this paper, we propose an off-the-shelf optimization approach that can minimize virtually any non-differentiable and non-decomposable loss function (e.g. Miss-classification Rate, AUC, F1, Jaccard Index, Mathew Correlation Coefficient, etc.) seamlessly. Our strategy learns smooth relaxation versions of the true losses by approximating them through a surrogate neural network. The proposed loss networks are set-wise models which are invariant to the order of mini-batch instances. Ultimately, the surrogate losses are learned jointly with the prediction model via bilevel optimization. Empirical results on multiple datasets with diverse real-life loss functions compared with state-of-the-art baselines demonstrate the efficiency of learning surrogate losses.	Due to the non-differential and non-continuous nature of most real-life losses, early works deployed the proxies of the miss-classification rate (e.g. cross-entropy) as universal proxy losses, despite their sub-optimal performance @cite_21 . Subsequent approaches relied on designing smooth relaxations of the losses. As an example, the pairwise ranking loss is a common surrogate of the AUC measure @cite_16 @cite_14 . On the other, hand, the F-measure is another typical loss that cannot be optimized directly due to its non-decomposable nature over instances. The initial papers tackling F1 focused instead on the empirical utility maximization paradigm @cite_15 . Later on, researchers addressed F1 by optimizing the hyper-parameters of standard binary classifiers, either the cost-sensitive weights of the classification loss @cite_9 , or the threshold of the estimated target values @cite_7 @cite_23 . Nevertheless, the non-decomposable trait of F1 remains unresolved @cite_12 and recent works have explored directions to improve hyper-parameter tuning with tighter bounds @cite_1 .	{ "cite_N": [ "@cite_14", "@cite_7", "@cite_9", "@cite_21", "@cite_1", "@cite_23", "@cite_15", "@cite_16", "@cite_12" ], "mid": [ "2142697503", "61851215", "2104144433", "2120100126", "2921041035", "2122770142", "1758941703", "2962977416", "2892127629" ], "abstract": [ "Learning to rank has become an important research topic in machine learning. While most learning-to-rank methods learn the ranking functions by minimizing loss functions, it is the ranking measures (such as NDCG and MAP) that are used to evaluate the performance of the learned ranking functions. In this work, we reveal the relationship between ranking measures and loss functions in learning-to-rank methods, such as Ranking SVM, RankBoost, RankNet, and ListMLE. We show that the loss functions of these methods are upper bounds of the measure-based ranking errors. As a result, the minimization of these loss functions will lead to the maximization of the ranking measures. The key to obtaining this result is to model ranking as a sequence of classification tasks, and define a so-called essential loss for ranking as the weighted sum of the classification errors of individual tasks in the sequence. We have proved that the essential loss is both an upper bound of the measure-based ranking errors, and a lower bound of the loss functions in the aforementioned methods. Our proof technique also suggests a way to modify existing loss functions to make them tighter bounds of the measure-based ranking errors. Experimental results on benchmark datasets show that the modifications can lead to better ranking performances, demonstrating the correctness of our theoretical analysis.", "This paper provides new insight into maximizing F1 measures in the context of binary classification and also in the context of multilabel classification. The harmonic mean of precision and recall, the F1 measure is widely used to evaluate the success of a binary classifier when one class is rare. Micro average, macro average, and per instance average F1 measures are used in multilabel classification. For any classifier that produces a real-valued output, we derive the relationship between the best achievable F1 value and the decision-making threshold that achieves this optimum. As a special case, if the classifier outputs are well-calibrated conditional probabilities, then the optimal threshold is half the optimal F1 value. As another special case, if the classifier is completely uninformative, then the optimal behavior is to classify all examples as positive. When the actual prevalence of positive examples is low, this behavior can be undesirable. As a case study, we discuss the results, which can be surprising, of maximizing F1 when predicting 26,853 labels for Medline documents.", "We present a theoretical analysis of F-measures for binary, multiclass and multilabel classification. These performance measures are non-linear, but in many scenarios they are pseudo-linear functions of the per-class false negative false positive rate. Based on this observation, we present a general reduction of F-measure maximization to cost-sensitive classification with unknown costs. We then propose an algorithm with provable guarantees to obtain an approximately optimal classifier for the F-measure by solving a series of cost-sensitive classification problems. The strength of our analysis is to be valid on any dataset and any class of classifiers, extending the existing theoretical results on F-measures, which are asymptotic in nature. We present numerical experiments to illustrate the relative importance of cost asymmetry and thresholding when learning linear classifiers on various F-measure optimization tasks.", "The area under an ROC curve (AUC) is a criterion used in many applications to measure the quality of a classification algorithm. However, the objective function optimized in most of these algorithms is the error rate and not the AUC value. We give a detailed statistical analysis of the relationship between the AUC and the error rate, including the first exact expression of the expected value and the variance of the AUC for a fixed error rate. Our results show that the average AUC is monotonically increasing as a function of the classification accuracy, but that the standard deviation for uneven distributions and higher error rates is noticeable. Thus, algorithms designed to minimize the error rate may not lead to the best possible AUC values. We show that, under certain conditions, the global function optimized by the RankBoost algorithm is exactly the AUC. We report the results of our experiments with RankBoost in several datasets demonstrating the benefits of an algorithm specifically designed to globally optimize the AUC over other existing algorithms optimizing an approximation of the AUC or only locally optimizing the AUC.", "The F-measure is a classification performance measure, especially suited when dealing with imbalanced datasets, which provides a compromise between the precision and the recall of a classifier. As this measure is non convex and non linear, it is often indirectly optimized using cost-sensitive learning (that affects different costs to false positives and false negatives). In this article, we derive theoretical guarantees that give tight bounds on the best F-measure that can be obtained from cost-sensitive learning. We also give an original geometric interpretation of the bounds that serves as an inspiration for CONE, a new algorithm to optimize for the F-measure. Using 10 datasets exhibiting varied class imbalance, we illustrate that our bounds are much tighter than previous work and show that CONE learns models with either superior F-measures than existing methods or comparable but in fewer iterations.", "Performance metrics for binary classification are designed to capture tradeoffs between four fundamental population quantities: true positives, false positives, true negatives and false negatives. Despite significant interest from theoretical and applied communities, little is known about either optimal classifiers or consistent algorithms for optimizing binary classification performance metrics beyond a few special cases. We consider a fairly large family of performance metrics given by ratios of linear combinations of the four fundamental population quantities. This family includes many well known binary classification metrics such as classification accuracy, AM measure, F-measure and the Jaccard similarity coefficient as special cases. Our analysis identifies the optimal classifiers as the sign of the thresholded conditional probability of the positive class, with a performance metric-dependent threshold. The optimal threshold can be constructed using simple plug-in estimators when the performance metric is a linear combination of the population quantities, but alternative techniques are required for the general case. We propose two algorithms for estimating the optimal classifiers, and prove their statistical consistency. Both algorithms are straightforward modifications of standard approaches to address the key challenge of optimal threshold selection, thus are simple to implement in practice. The first algorithm combines a plug-in estimate of the conditional probability of the positive class with optimal threshold selection. The second algorithm leverages recent work on calibrated asymmetric surrogate losses to construct candidate classifiers. We present empirical comparisons between these algorithms on benchmark datasets.", "F-measures are popular performance metrics, particularly for tasks with imbalanced data sets. Algorithms for learning to maximize F-measures follow two approaches: the empirical utility maximization (EUM) approach learns a classifier having optimal performance on training data, while the decision-theoretic approach learns a probabilistic model and then predicts labels with maximum expected F-measure. In this paper, we investigate the theoretical justifications and connections for these two approaches, and we study the conditions under which one approach is preferable to the other using synthetic and real datasets. Given accurate models, our results suggest that the two approaches are asymptotically equivalent given large training and test sets. Nevertheless, empirically, the EUM approach appears to be more robust against model misspecification, and given a good model, the decision-theoretic approach appears to be better for handling rare classes and a common domain adaptation scenario.", "AUC (Area Under ROC Curve) has been an important criterion widely used in diverse learning tasks. To optimize AUC, many learning approaches have been developed, most working with pairwise surrogate losses. Thus, it is important to study the AUC consistency based on minimizing pairwise surrogate losses. In this paper, we introduce the generalized calibration for AUC optimization, and prove that it is a necessary condition for AUC consistency. We then provide a sufficient condition for AUC consistency, and show its usefulness in studying the consistency of various surrogate losses, as well as the invention of new consistent losses. We further derive regret bounds for exponential and logistic losses, and present regret bounds for more general surrogate losses in the realizable setting. Finally, we prove regret bounds that disclose the equivalence between the pairwise exponential loss of AUC and univariate exponential loss of accuracy.", "In this paper, we consider online F-measure optimization (OFO). Unlike traditional performance metrics (e.g., classification error rate), F-measure is non-decomposable over training examples and is a non-convex function of model parameters, making it much more difficult to be optimized in an online fashion. Most existing results of OFO usually suffer from high memory computational costs and or lack statistical consistency guarantee for optimizing F-measure at the population level. To advance OFO, we propose an efficient online algorithm based on simultaneously learning a posterior probability of class and learning an optimal threshold by minimizing a stochastic strongly convex function with unknown strong convexity parameter. A key component of the proposed method is a novel stochastic algorithm with low memory and computational costs, which can enjoy a convergence rate of ˜ O (1 √ n ) for learning the optimal threshold under a mild condition on the convergence of the posterior probability, where n is the number of processed examples. It is provably faster than its predecessor based on a heuristic for updating the threshold. The experiments verify the efficiency of the proposed algorithm in comparison with state-of-the-art OFO algorithms." ] }
1905.10108	2946154049	The minimization of loss functions is the heart and soul of Machine Learning. In this paper, we propose an off-the-shelf optimization approach that can minimize virtually any non-differentiable and non-decomposable loss function (e.g. Miss-classification Rate, AUC, F1, Jaccard Index, Mathew Correlation Coefficient, etc.) seamlessly. Our strategy learns smooth relaxation versions of the true losses by approximating them through a surrogate neural network. The proposed loss networks are set-wise models which are invariant to the order of mini-batch instances. Ultimately, the surrogate losses are learned jointly with the prediction model via bilevel optimization. Empirical results on multiple datasets with diverse real-life loss functions compared with state-of-the-art baselines demonstrate the efficiency of learning surrogate losses.	Instead of relying on explicit surrogates, another research direction handles non-convex losses by means of the method @cite_5 , which minimizes a surrogate loss by embedding the true loss as a correction term. This method was recently extended to optimize neural networks @cite_0 . It assumes the loss can be decomposed into per-instance sub-losses and the authors derived an explicit decomposition of the average precision @cite_0 . Unfortunately, per-instance dis-aggregations are not trivially feasible in other cases (e.g. F1), making the direct loss optimization technique an impractical off-the-shelf option.	{ "cite_N": [ "@cite_0", "@cite_5" ], "mid": [ "2963789586", "2099119623" ], "abstract": [ "Supervised training of deep neural nets typically relies on minimizing cross-entropy. However, in many domains, we are interested in performing well on metrics specific to the application. In this paper we propose a direct loss minimization approach to train deep neural networks, which provably minimizes the application-specific loss function. This is often non-trivial, since these functions are neither smooth nor decomposable and thus are not amenable to optimization with standard gradient-based methods. We demonstrate the effectiveness of our approach in the context of maximizing average precision for ranking problems. Towards this goal, we develop a novel dynamic programming algorithm that can efficiently compute the weight updates. Our approach proves superior to a variety of baselines in the context of action classification and object detection, especially in the presence of label noise.", "In discriminative machine learning one is interested in training a system to optimize a certain desired measure of performance, or loss. In binary classification one typically tries to minimizes the error rate. But in structured prediction each task often has its own measure of performance such as the BLEU score in machine translation or the intersection-over-union score in PASCAL segmentation. The most common approaches to structured prediction, structural SVMs and CRFs, do not minimize the task loss: the former minimizes a surrogate loss with no guarantees for task loss and the latter minimizes log loss independent of task loss. The main contribution of this paper is a theorem stating that a certain perceptron-like learning rule, involving features vectors derived from loss-adjusted inference, directly corresponds to the gradient of task loss. We give empirical results on phonetic alignment of a standard test set from the TIMIT corpus, which surpasses all previously reported results on this problem." ] }
1905.10108	2946154049	The minimization of loss functions is the heart and soul of Machine Learning. In this paper, we propose an off-the-shelf optimization approach that can minimize virtually any non-differentiable and non-decomposable loss function (e.g. Miss-classification Rate, AUC, F1, Jaccard Index, Mathew Correlation Coefficient, etc.) seamlessly. Our strategy learns smooth relaxation versions of the true losses by approximating them through a surrogate neural network. The proposed loss networks are set-wise models which are invariant to the order of mini-batch instances. Ultimately, the surrogate losses are learned jointly with the prediction model via bilevel optimization. Empirical results on multiple datasets with diverse real-life loss functions compared with state-of-the-art baselines demonstrate the efficiency of learning surrogate losses.	Two more recent papers have offered relaxation surrogates for non-decomposable losses, concretely the AUC and the Jaccard Index (known as Intersection over Union in the computer vision community). The first method defines relaxation forms for the building blocks of a confusion matrix (e.g. true positives, true negatives etc.) and combines the building block relaxations to create a final surrogate for losses like the AUC @cite_10 . However, this model does not handle cases where the loss is not expressible into confusion matrix blocks, for instance the Jaccard index. On the other hand, the second paper proposes the Lovasz soft-max as a smooth approximation to the Jaccard index @cite_8 , which is based on a generic decomposition of sub-modular (decomposable) losses for sets @cite_6 .	{ "cite_N": [ "@cite_10", "@cite_6", "@cite_8" ], "mid": [ "2964027161", "1863660263", "2795587607" ], "abstract": [ "", "Learning with non-modular losses is an important problem when sets of predictions are made simultaneously. The main tools for constructing convex surrogate loss functions for set prediction are margin rescaling and slack rescaling. In this work, we show that these strategies lead to tight convex surrogates iff the underlying loss function is increasing in the number of incorrect predictions. However, gradient or cutting-plane computation for these functions is NP-hard for nonsupermodular loss functions. We propose instead a novel convex surrogate loss function for submodular losses, the Lovasz hinge, which leads to O(p log p) complexity with O(p) oracle accesses to the loss function to compute a gradient or cutting-plane. As a result, we have developed the first tractable convex surrogates in the literature for submodular losses. We demonstrate the utility of this novel convex surrogate through a real world image labeling task.", "The Jaccard index, also referred to as the intersection-over-union score, is commonly employed in the evaluation of image segmentation results given its perceptual qualities, scale invariance - which lends appropriate relevance to small objects, and appropriate counting of false negatives, in comparison to per-pixel losses. We present a method for direct optimization of the mean intersection-over-union loss in neural networks, in the context of semantic image segmentation, based on the convex LovAisz extension of submodular losses. The loss is shown to perform better with respect to the Jaccard index measure than the traditionally used cross-entropy loss. We show quantitative and qualitative differences between optimizing the Jaccard index per image versus optimizing the Jaccard index taken over an entire dataset. We evaluate the impact of our method in a semantic segmentation pipeline and show substantially improved intersection-over-union segmentation scores on the Pascal VOC and Cityscapes datasets using state-of-the-art deep learning segmentation architectures." ] }
1905.10108	2946154049	The minimization of loss functions is the heart and soul of Machine Learning. In this paper, we propose an off-the-shelf optimization approach that can minimize virtually any non-differentiable and non-decomposable loss function (e.g. Miss-classification Rate, AUC, F1, Jaccard Index, Mathew Correlation Coefficient, etc.) seamlessly. Our strategy learns smooth relaxation versions of the true losses by approximating them through a surrogate neural network. The proposed loss networks are set-wise models which are invariant to the order of mini-batch instances. Ultimately, the surrogate losses are learned jointly with the prediction model via bilevel optimization. Empirical results on multiple datasets with diverse real-life loss functions compared with state-of-the-art baselines demonstrate the efficiency of learning surrogate losses.	Besides that, a stream of recent papers has focused on meta-learning for loss functions. The "Learning to Teach" paradigm @cite_4 proposes a meta-level teacher controller that continuously updates the loss function for a prediction model based on its progress. The work has been recently extended to enable a gradient-based learning of the teacher controller @cite_17 . However, this approach does not extend to non-decomposable loss functions which are defined over the full set of instances. A parallel stream of research has elaborated the concept of "Learning to Learn" @cite_13 , or "Learning to Optimize" @cite_19 , which proposes to directly learn the amount of update values that are applied to the parameters of the prediction model. In the proposed framework a controller uses per-parameter learning curves comprised of the loss values and derivatives of the loss with respect to each parameters @cite_19 . This method suffers from two drawbacks that prohibit its direct applicability to arbitrary losses: i) for large prediction models it is computationally infeasible to store the learning curve of every parameter, and ii) there is no gradient information for non-differentiable losses.	{ "cite_N": [ "@cite_13", "@cite_19", "@cite_4", "@cite_17" ], "mid": [ "2962880633", "2624086852", "", "2963780286" ], "abstract": [ "Algorithm design is a laborious process and often requires many iterations of ideation and validation. In this paper, we explore automating algorithm design and present a method to learn an optimization algorithm. We approach this problem from a reinforcement learning perspective and represent any particular optimization algorithm as a policy. We learn an optimization algorithm using guided policy search and demonstrate that the resulting algorithm outperforms existing hand-engineered algorithms in terms of convergence speed and or the final objective value.", "We learn recurrent neural network optimizers trained on simple synthetic functions by gradient descent. We show that these learned optimizers exhibit a remarkable degree of transfer in that they can be used to efficiently optimize a broad range of derivative-free black-box functions, including Gaussian process bandits, simple control objectives, global optimization benchmarks and hyper-parameter tuning tasks. Up to the training horizon, the learned optimizers learn to tradeoff exploration and exploitation, and compare favourably with heavily engineered Bayesian optimization packages for hyper-parameter tuning.", "", "Teaching is critical to human society: it is with teaching that prospective students are educated and human civilization can be inherited and advanced. A good teacher not only provides his her students with qualified teaching materials (e.g., textbooks), but also sets up appropriate learning objectives (e.g., course projects and exams) considering different situations of a student. When it comes to artificial intelligence, treating machine learning models as students, the loss functions that are optimized act as perfect counterparts of the learning objective set by the teacher. In this work, we explore the possibility of imitating human teaching behaviors by dynamically and automatically outputting appropriate loss functions to train machine learning models. Different from typical learning settings in which the loss function of a machine learning model is predefined and fixed, in our framework, the loss function of a machine learning model (we call it student) is defined by another machine learning model (we call it teacher). The ultimate goal of teacher model is cultivating the student to have better performance measured on development dataset. Towards that end, similar to human teaching, the teacher, a parametric model, dynamically outputs different loss functions that will be used and optimized by its student model at different training stages. We develop an efficient learning method for the teacher model that makes gradient based optimization possible, exempt of the ineffective solutions such as policy optimization. We name our method as learning to teach with dynamic loss functions'' (L2T-DLF for short). Extensive experiments on real world tasks including image classification and neural machine translation demonstrate that our method significantly improves the quality of various student models." ] }
1905.09634	2946257523	Outlier feature matches and loop-closures that survived front-end data association can lead to catastrophic failures in the back-end optimization of large-scale point cloud based 3D reconstruction. To alleviate this problem, we propose a probabilistic approach for robust back-end optimization in the presence of outliers. More specifically, we model the problem as a Bayesian network and solve it using the Expectation-Maximization algorithm. Our approach leverages on a long-tail Cauchy distribution to suppress outlier feature matches in the odometry constraints, and a Cauchy-Uniform mixture model with a set of binary latent variables to simultaneously suppress outlier loop-closure constraints and outlier feature matches in the inlier loop-closure constraints. Furthermore, we show that by using a Gaussian-Uniform mixture model, our approach degenerates to the formulation of a state-of-the-art approach for robust indoor reconstruction. Experimental results demonstrate that our approach has comparable performance with the state-of-the-art on a benchmark indoor dataset, and outperforms it on a large-scale outdoor dataset. Our source code can be found on the project website.	Reconstruction of outdoor scenes has been studied in @cite_27 @cite_16 . Sch "o @cite_30 propose a set of filtering steps to detect and discard unreliable depth measurements acquired from a RGB-D camera. However, loop-closures is not detected and this can lead to reconstruction failures. Relying on very accurate GPS INS, @cite_27 propose a 3D reconstruction system from RGB images. However, GPS INS signal may be unavailable or unreliable, especially on cloudy days or in urban canyons. Our work relies on neither GPS INS nor RGB images. In contrast, we focus on reconstruction from point cloud data acquired from 3D scanners that is less sensitive to weather or lighting changes. There are also many works on indoor scene reconstruction. Since the seminal KinectFusion @cite_0 , there are several follow-up algorithms @cite_40 @cite_19 @cite_23 . Unfortunately, these methods do not detect loop-closures. Nonetheless, there are many RGB-D reconstruction methods with loop-closure detection @cite_10 @cite_2 @cite_35 @cite_4 @cite_36 @cite_8 @cite_32 @cite_24 @cite_6 .	{ "cite_N": [ "@cite_30", "@cite_35", "@cite_4", "@cite_8", "@cite_36", "@cite_32", "@cite_6", "@cite_0", "@cite_19", "@cite_40", "@cite_27", "@cite_24", "@cite_23", "@cite_2", "@cite_16", "@cite_10" ], "mid": [ "2519897181", "", "", "", "", "", "", "1987648924", "", "2071398263", "2059496137", "", "", "", "", "1957167950" ], "abstract": [ "Approach for reconstructing large-scale outdoor scenes using monocular motion stereo.Runs at interactive frame rates on a Google Project Tango Development Kit Tablet.Can use strict outlier filtering, as the user can interactively record more data.Missing parts in the live preview guide the user to places which require more data. This paper presents an approach for reconstructing large-scale outdoor scenes through monocular motion stereo at interactive frame rates on a modern mobile device (Google Project Tango Development Kit Tablet). The devices fisheye camera enables a user to reconstruct large scenes in only a few minutes by simply walking through the scene. We utilize the devices GPU to compute depth maps via plane sweep stereo. In contrast to reconstructing small objects, we observe that in large-scale scenarios using motion stereo, free-space measurements are less effective for suppressing outliers due to limited possibilities for camera placement and an unbounded reconstruction volume. Furthermore, the outlier ratio in depth maps from stereo matching is much higher compared to images from depth sensors. Consequently, we propose a set of filtering steps to detect and discard unreliable depth measurements. The remaining parts of the depth maps are then integrated into a volumetric representation of the scene using a truncated signed distance function. Ours is the first method to enable live reconstruction of large outdoor scenes on a mobile device. We extensively evaluate our approach, demonstrating the benefit of rigorously filtering depth maps.", "", "", "", "", "", "", "We present a system for accurate real-time mapping of complex and arbitrary indoor scenes in variable lighting conditions, using only a moving low-cost depth camera and commodity graphics hardware. We fuse all of the depth data streamed from a Kinect sensor into a single global implicit surface model of the observed scene in real-time. The current sensor pose is simultaneously obtained by tracking the live depth frame relative to the global model using a coarse-to-fine iterative closest point (ICP) algorithm, which uses all of the observed depth data available. We demonstrate the advantages of tracking against the growing full surface model compared with frame-to-frame tracking, obtaining tracking and mapping results in constant time within room sized scenes with limited drift and high accuracy. We also show both qualitative and quantitative results relating to various aspects of our tracking and mapping system. Modelling of natural scenes, in real-time with only commodity sensor and GPU hardware, promises an exciting step forward in augmented reality (AR), in particular, it allows dense surfaces to be reconstructed in real-time, with a level of detail and robustness beyond any solution yet presented using passive computer vision.", "", "We address the fundamental challenge of scalability for real-time volumetric surface reconstruction methods. We design a memory efficient, hierarchical data structure for commodity graphics hardware, which supports live reconstruction of large-scale scenes with fine geometric details. Our sparse data structure fuses overlapping depth maps from a moving depth camera into a single volumetric representation, from which detailed surface models are extracted. Our hierarchy losslessly streams data bidirectionally between GPU and host, allowing for unbounded reconstructions. Our pipeline, comprised of depth map post-processing, camera pose estimation, volumetric fusion, surface extraction, and streaming, runs entirely in real-time. We experimentally demonstrate that a shallow hierarchy with relatively large branching factors yields the best memory speed tradeoff, consuming an order of magnitude less memory than a regular grid. We compare an implementation of our data structure to existing methods and demonstrate higher-quality reconstructions on a variety of large-scale scenes, all captured in real-time.", "The paper presents a system for automatic, geo-registered, real-time 3D reconstruction from video of urban scenes. The system collects video streams, as well as GPS and inertia measurements in order to place the reconstructed models in geo-registered coordinates. It is designed using current state of the art real-time modules for all processing steps. It employs commodity graphics hardware and standard CPU's to achieve real-time performance. We present the main considerations in designing the system and the steps of the processing pipeline. Our system extends existing algorithms to meet the robustness and variability necessary to operate out of the lab. To account for the large dynamic range of outdoor videos the processing pipeline estimates global camera gain changes in the feature tracking stage and efficiently compensates for these in stereo estimation without impacting the real-time performance. The required accuracy for many applications is achieved with a two-step stereo reconstruction process exploiting the redundancy across frames. We show results on real video sequences comprising hundreds of thousands of frames.", "", "", "", "", "We present an approach to indoor scene reconstruction from RGB-D video. The key idea is to combine geometric registration of scene fragments with robust global optimization based on line processes. Geometric registration is error-prone due to sensor noise, which leads to aliasing of geometric detail and inability to disambiguate different surfaces in the scene. The presented optimization approach disables erroneous geometric alignments even when they significantly outnumber correct ones. Experimental results demonstrate that the presented approach substantially increases the accuracy of reconstructed scene models." ] }
1905.09634	2946257523	Outlier feature matches and loop-closures that survived front-end data association can lead to catastrophic failures in the back-end optimization of large-scale point cloud based 3D reconstruction. To alleviate this problem, we propose a probabilistic approach for robust back-end optimization in the presence of outliers. More specifically, we model the problem as a Bayesian network and solve it using the Expectation-Maximization algorithm. Our approach leverages on a long-tail Cauchy distribution to suppress outlier feature matches in the odometry constraints, and a Cauchy-Uniform mixture model with a set of binary latent variables to simultaneously suppress outlier loop-closure constraints and outlier feature matches in the inlier loop-closure constraints. Furthermore, we show that by using a Gaussian-Uniform mixture model, our approach degenerates to the formulation of a state-of-the-art approach for robust indoor reconstruction. Experimental results demonstrate that our approach has comparable performance with the state-of-the-art on a benchmark indoor dataset, and outperforms it on a large-scale outdoor dataset. Our source code can be found on the project website.	@cite_10 achieve the state-of-the-art performance for indoor reconstruction with robust loop-closure. However, they assume no outlier feature matches in the odometry and inlier loop-closure constraints. We relax this assumption to achieve robust feature matching. More specifically, @cite_10 estimates a switch variable @cite_7 for each loop-closure constraint using line processes @cite_26 . Outlier loop-closures are disabled by setting the respective switch variables to zero. Additional switch prior terms are imposed and chosen empirically @cite_7 to prevent a trivial solution of removing all loop-closure constraints. In comparison, our approach does not require the additional prior terms. We estimate the posterior of a loop-closure being an inlier constraint in the Expectation step shown in Sec. . The EM approach is also used by @cite_29 . However, they solve a robust pose-graph optimization problem without coping with the feature matches for reconstruction.	{ "cite_N": [ "@cite_29", "@cite_26", "@cite_10", "@cite_7" ], "mid": [ "1990976732", "2088616581", "1957167950", "1999399477" ], "abstract": [ "In this paper, we model the robust loop-closure pose-graph SLAM problem as a Bayesian network and show that it can be solved with the Classification Expectation-Maximization (EM) algorithm. In particular, we express our robust pose-graph SLAM as a Bayesian network where the robot poses and constraints are latent and observed variables. An additional set of latent variables is introduced as weights for the loop-constraints. We show that the weights can be chosen as the Cauchy function, which are iteratively computed from the errors between the predicted robot poses and observed loop-closure constraints in the Expectation step, and used to weigh the cost functions from the pose-graph loop-closure constraints in the Maximization step. As a result, outlier loop-closure constraints are assigned low weights and exert less influences in the pose-graph optimization within the EM iterations. To prevent the EM algorithm from getting stuck at local minima, we perform the EM algorithm multiple times where the loop constraints with very low weights are removed after each EM process. This is repeated until there are no more changes to the weights. We show proofs of the conceptual similarity between our EM algorithm and the M-Estimator. Specifically, we show that the weight function in our EM algorithm is equivalent to the robust residual function in the M-Estimator. We verify our proposed algorithm with experimental results from multiple simulated and real-world datasets, and comparisons with other existing works.", "The modeling of spatial discontinuities for problems such as surface recovery, segmentation, image reconstruction, and optical flow has been intensely studied in computer vision. While “line-process” models of discontinuities have received a great deal of attention, there has been recent interest in the use of robust statistical techniques to account for discontinuities. This paper unifies the two approaches. To achieve this we generalize the notion of a “line process” to that of an analog “outlier process” and show how a problem formulated in terms of outlier processes can be viewed in terms of robust statistics. We also characterize a class of robust statistical problems for which an equivalent outlier-process formulation exists and give a straightforward method for converting a robust estimation problem into an outlier-process formulation. We show how prior assumptions about the spatial structure of outliers can be expressed as constraints on the recovered analog outlier processes and how traditional continuation methods can be extended to the explicit outlier-process formulation. These results indicate that the outlier-process approach provides a general framework which subsumes the traditional line-process approaches as well as a wide class of robust estimation problems. Examples in surface reconstruction, image segmentation, and optical flow are presented to illustrate the use of outlier processes and to show how the relationship between outlier processes and robust statistics can be exploited. An appendix provides a catalog of common robust error norms and their equivalent outlier-process formulations.", "We present an approach to indoor scene reconstruction from RGB-D video. The key idea is to combine geometric registration of scene fragments with robust global optimization based on line processes. Geometric registration is error-prone due to sensor noise, which leads to aliasing of geometric detail and inability to disambiguate different surfaces in the scene. The presented optimization approach disables erroneous geometric alignments even when they significantly outnumber correct ones. Experimental results demonstrate that the presented approach substantially increases the accuracy of reconstructed scene models.", "Current SLAM back-ends are based on least squares optimization and thus are not robust against outliers like data association errors and false positive loop closure detections. Our paper presents and evaluates a robust back-end formulation for SLAM using switchable constraints. Instead of proposing yet another appearance-based data association technique, our system is able to recognize and reject outliers during the optimization. This is achieved by making the topology of the underlying factor graph representation subject to the optimization instead of keeping it fixed. The evaluation shows that the approach can deal with up to 1000 false positive loop closure constraints on various datasets. This largely increases the robustness of the overall SLAM system and closes a gap between the sensor-driven front-end and the back-end optimizers." ] }
1905.09797	2951652972	We attempt to interpret how adversarially trained convolutional neural networks (AT-CNNs) recognize objects. We design systematic approaches to interpret AT-CNNs in both qualitative and quantitative ways and compare them with normally trained models. Surprisingly, we find that adversarial training alleviates the texture bias of standard CNNs when trained on object recognition tasks, and helps CNNs learn a more shape-biased representation. We validate our hypothesis from two aspects. First, we compare the salience maps of AT-CNNs and standard CNNs on clean images and images under different transformations. The comparison could visually show that the prediction of the two types of CNNs is sensitive to dramatically different types of features. Second, to achieve quantitative verification, we construct additional test datasets that destroy either textures or shapes, such as style-transferred version of clean data, saturated images and patch-shuffled ones, and then evaluate the classification accuracy of AT-CNNs and normal CNNs on these datasets. Our findings shed some light on why AT-CNNs are more robust than those normally trained ones and contribute to a better understanding of adversarial training over CNNs from an interpretation perspective.	In this work, we mainly interpret the AT-CNNs based on norm-constrained perturbation over the original images. It is worthy of noting that the difference between normally trained and adversarially trained CNNs may highly depends on the type of adversaries. Models trained against spatially-transformed adversary @cite_39 , denoted as ST-ST-CNNs, have similar robustness towards PGD attack with standard models, and their salience maps are still quite different as shown in Figure . Also the average distance between salience maps is close to that of standard CNN, which is much higher than that of PGD-AT-CNN. There exists a variety of generalized types of attacks, @math parameterized by @math , such as spatially transformed @cite_39 and GAN-based adversarial examples @cite_3 . We leave interpreting the AT-CNNs based on these generalized types of attacks as future work.	{ "cite_N": [ "@cite_3", "@cite_39" ], "mid": [ "2964238361", "2963297642" ], "abstract": [ "Adversarial examples are typically constructed by perturbing an existing data point within a small matrix norm, and current defense methods are focused on guarding against this type of attack. In this paper, we propose a new class of adversarial examples that are synthesized entirely from scratch using a conditional generative model, without being restricted to norm-bounded perturbations. We first train an Auxiliary Classifier Generative Adversarial Network (AC-GAN) to model the class-conditional distribution over data samples. Then, conditioned on a desired class, we search over the AC-GAN latent space to find images that are likely under the generative model and are misclassified by a target classifier. We demonstrate through human evaluation that these new kind of adversarial images, which we call Generative Adversarial Examples, are legitimate and belong to the desired class. Our empirical results on the MNIST, SVHN, and CelebA datasets show that generative adversarial examples can bypass strong adversarial training and certified defense methods designed for traditional adversarial attacks.", "Recent studies show that widely used Deep neural networks (DNNs) are vulnerable to the carefully crafted adversarial examples. Many advanced algorithms have been proposed to generate adversarial examples by leveraging the L_p distance for penalizing perturbations. Different defense methods have also been explored to defend against such adversarial attacks. While the effectiveness of L_p distance as a metric of perceptual quality remains an active research area, in this paper we will instead focus on a different type of perturbation, namely spatial transformation, as opposed to manipulating the pixel values directly as in prior works. Perturbations generated through spatial transformation could result in large L_p distance measures, but our extensive experiments show that such spatially transformed adversarial examples are perceptually realistic and more difficult to defend against with existing defense systems. This potentially provides a new direction in adversarial example generation and the design of corresponding defenses. We visualize the spatial transformation based perturbation for different examples and show that our technique can produce realistic adversarial examples with smooth image deformation. Finally, we visualize the attention of deep networks with different types of adversarial examples to better understand how these examples are interpreted." ] }
1905.09763	2945245453	Graph embedding seeks to build a low-dimensional representation of a graph G. This low-dimensional representation is then used for various downstream tasks. One popular approach is Laplacian Eigenmaps, which constructs a graph embedding based on the spectral properties of the Laplacian matrix of G. The intuition behind it, and many other embedding techniques, is that the embedding of a graph must respect node similarity: similar nodes must have embeddings that are close to one another. Here, we dispose of this distance-minimization assumption. Instead, we use the Laplacian matrix to find an embedding with geometric properties instead of spectral ones, by leveraging the so-called simplex geometry of G. We introduce a new approach, Geometric Laplacian Eigenmap Embedding (or GLEE for short), and demonstrate that it outperforms various other techniques (including Laplacian Eigenmaps) in the tasks of graph reconstruction and link prediction.	Within the methods depending on random walks we find DeepWalk @cite_17 and node2vec @cite_24 , both of which adapt the framework of word embeddings to graphs by using (biased) random walks and optimize a shallow architecture. It's also worth mentioning NetMF @cite_23 which unifies several methods such as DeepWalk and node2vec in a single algorithm that depends on matrix factorization and thus unifies the two previous categories.	{ "cite_N": [ "@cite_24", "@cite_23", "@cite_17" ], "mid": [ "2962756421", "2761896323", "2154851992" ], "abstract": [ "Prediction tasks over nodes and edges in networks require careful effort in engineering features used by learning algorithms. Recent research in the broader field of representation learning has led to significant progress in automating prediction by learning the features themselves. However, present feature learning approaches are not expressive enough to capture the diversity of connectivity patterns observed in networks. Here we propose node2vec, an algorithmic framework for learning continuous feature representations for nodes in networks. In node2vec, we learn a mapping of nodes to a low-dimensional space of features that maximizes the likelihood of preserving network neighborhoods of nodes. We define a flexible notion of a node's network neighborhood and design a biased random walk procedure, which efficiently explores diverse neighborhoods. Our algorithm generalizes prior work which is based on rigid notions of network neighborhoods, and we argue that the added flexibility in exploring neighborhoods is the key to learning richer representations. We demonstrate the efficacy of node2vec over existing state-of-the-art techniques on multi-label classification and link prediction in several real-world networks from diverse domains. Taken together, our work represents a new way for efficiently learning state-of-the-art task-independent representations in complex networks.", "Since the invention of word2vec, the skip-gram model has significantly advanced the research of network embedding, such as the recent emergence of the DeepWalk, LINE, PTE, and node2vec approaches. In this work, we show that all of the aforementioned models with negative sampling can be unified into the matrix factorization framework with closed forms. Our analysis and proofs reveal that: (1) DeepWalk empirically produces a low-rank transformation of a network's normalized Laplacian matrix; (2) LINE, in theory, is a special case of DeepWalk when the size of vertices' context is set to one; (3) As an extension of LINE, PTE can be viewed as the joint factorization of multiple networks» Laplacians; (4) node2vec is factorizing a matrix related to the stationary distribution and transition probability tensor of a 2nd-order random walk. We further provide the theoretical connections between skip-gram based network embedding algorithms and the theory of graph Laplacian. Finally, we present the NetMF method as well as its approximation algorithm for computing network embedding. Our method offers significant improvements over DeepWalk and LINE for conventional network mining tasks. This work lays the theoretical foundation for skip-gram based network embedding methods, leading to a better understanding of latent network representation learning.", "We present DeepWalk, a novel approach for learning latent representations of vertices in a network. These latent representations encode social relations in a continuous vector space, which is easily exploited by statistical models. DeepWalk generalizes recent advancements in language modeling and unsupervised feature learning (or deep learning) from sequences of words to graphs. DeepWalk uses local information obtained from truncated random walks to learn latent representations by treating walks as the equivalent of sentences. We demonstrate DeepWalk's latent representations on several multi-label network classification tasks for social networks such as BlogCatalog, Flickr, and YouTube. Our results show that DeepWalk outperforms challenging baselines which are allowed a global view of the network, especially in the presence of missing information. DeepWalk's representations can provide F1 scores up to 10 higher than competing methods when labeled data is sparse. In some experiments, DeepWalk's representations are able to outperform all baseline methods while using 60 less training data. DeepWalk is also scalable. It is an online learning algorithm which builds useful incremental results, and is trivially parallelizable. These qualities make it suitable for a broad class of real world applications such as network classification, and anomaly detection." ] }
1905.09763	2945245453	Graph embedding seeks to build a low-dimensional representation of a graph G. This low-dimensional representation is then used for various downstream tasks. One popular approach is Laplacian Eigenmaps, which constructs a graph embedding based on the spectral properties of the Laplacian matrix of G. The intuition behind it, and many other embedding techniques, is that the embedding of a graph must respect node similarity: similar nodes must have embeddings that are close to one another. Here, we dispose of this distance-minimization assumption. Instead, we use the Laplacian matrix to find an embedding with geometric properties instead of spectral ones, by leveraging the so-called simplex geometry of G. We introduce a new approach, Geometric Laplacian Eigenmap Embedding (or GLEE for short), and demonstrate that it outperforms various other techniques (including Laplacian Eigenmaps) in the tasks of graph reconstruction and link prediction.	Among the methods using that use architectures we have the deep autoencoder Structural Deep Network Embedding (SDNE) @cite_2 . It penalizes representations of similar nodes that are far from each other using the same objective as LE. Thus, SDNE is also based on the distance-minimization approach. There is also @cite_0 which obtains a non-linear mapping between the probabilistic mutual information matrix (PMI) of a sampled network and the embedding space. This is akin to applying the distance-minimization assumption not to the graph directly but to the PMI matrix.	{ "cite_N": [ "@cite_0", "@cite_2" ], "mid": [ "2415243320", "2393319904" ], "abstract": [ "In this paper, we propose a novel model for learning graph representations, which generates a low-dimensional vector representation for each vertex by capturing the graph structural information. Different from other previous research efforts, we adopt a random surfing model to capture graph structural information directly, instead of using the sampling-based method for generating linear sequences proposed by (2014). The advantages of our approach will be illustrated from both theorical and empirical perspectives. We also give a new perspective for the matrix factorization method proposed by Levy and Goldberg (2014), in which the pointwise mutual information (PMI) matrix is considered as an analytical solution to the objective function of the skip-gram model with negative sampling proposed by (2013). Unlike their approach which involves the use of the SVD for finding the low-dimensitonal projections from the PMI matrix, however, the stacked denoising autoencoder is introduced in our model to extract complex features and model non-linearities. To demonstrate the effectiveness of our model, we conduct experiments on clustering and visualization tasks, employing the learned vertex representations as features. Empirical results on datasets of varying sizes show that our model outperforms other stat-of-the-art models in such tasks.", "Network embedding is an important method to learn low-dimensional representations of vertexes in networks, aiming to capture and preserve the network structure. Almost all the existing network embedding methods adopt shallow models. However, since the underlying network structure is complex, shallow models cannot capture the highly non-linear network structure, resulting in sub-optimal network representations. Therefore, how to find a method that is able to effectively capture the highly non-linear network structure and preserve the global and local structure is an open yet important problem. To solve this problem, in this paper we propose a Structural Deep Network Embedding method, namely SDNE. More specifically, we first propose a semi-supervised deep model, which has multiple layers of non-linear functions, thereby being able to capture the highly non-linear network structure. Then we propose to exploit the first-order and second-order proximity jointly to preserve the network structure. The second-order proximity is used by the unsupervised component to capture the global network structure. While the first-order proximity is used as the supervised information in the supervised component to preserve the local network structure. By jointly optimizing them in the semi-supervised deep model, our method can preserve both the local and global network structure and is robust to sparse networks. Empirically, we conduct the experiments on five real-world networks, including a language network, a citation network and three social networks. The results show that compared to the baselines, our method can reconstruct the original network significantly better and achieves substantial gains in three applications, i.e. multi-label classification, link prediction and visualization." ] }
1905.09747	2946182708	Knowledge distillation is effective for producing small high-performance neural networks for classification, but these small networks are vulnerable to adversarial attacks. We first study how robustness transfers from robust teacher to student network during knowledge distillation. We find that a large amount of robustness may be inherited by the student even when distilled on only clean images. Second, we introduce Adversarially Robust Distillation (ARD) for distilling robustness onto small student networks. ARD is an analogue of adversarial training but for distillation. In addition to producing small models with high test accuracy like conventional distillation, ARD also passes the superior robustness of large networks onto the student. In our experiments, we find that ARD student models decisively outperform adversarially trained networks of identical architecture on robust accuracy. Finally, we adapt recent fast adversarial training methods to ARD for accelerated robust distillation.	Early schemes for compressing neural networks involved binarized weights to reduce storage and computation costs @cite_32 . Other early efforts focused on speeding up calculations via low-rank regularization and pruning weights to reduce computation costs @cite_0 @cite_9 . Knowledge distillation teaches a student network to mimic a more powerful teacher @cite_11 . The student network is usually a small, lightweight architecture like MobileNetV2 @cite_36 .	{ "cite_N": [ "@cite_36", "@cite_9", "@cite_32", "@cite_0", "@cite_11" ], "mid": [ "2963163009", "", "2319920447", "2181101938", "1821462560" ], "abstract": [ "In this paper we describe a new mobile architecture, MobileNetV2, that improves the state of the art performance of mobile models on multiple tasks and benchmarks as well as across a spectrum of different model sizes. We also describe efficient ways of applying these mobile models to object detection in a novel framework we call SSDLite. Additionally, we demonstrate how to build mobile semantic segmentation models through a reduced form of DeepLabv3 which we call Mobile DeepLabv3. is based on an inverted residual structure where the shortcut connections are between the thin bottleneck layers. The intermediate expansion layer uses lightweight depthwise convolutions to filter features as a source of non-linearity. Additionally, we find that it is important to remove non-linearities in the narrow layers in order to maintain representational power. We demonstrate that this improves performance and provide an intuition that led to this design. Finally, our approach allows decoupling of the input output domains from the expressiveness of the transformation, which provides a convenient framework for further analysis. We measure our performance on ImageNet [1] classification, COCO object detection [2], VOC image segmentation [3]. We evaluate the trade-offs between accuracy, and number of operations measured by multiply-adds (MAdd), as well as actual latency, and the number of parameters.", "", "We introduce a method to train Binarized Neural Networks (BNNs) - neural networks with binary weights and activations at run-time. At training-time the binary weights and activations are used for computing the parameters gradients. During the forward pass, BNNs drastically reduce memory size and accesses, and replace most arithmetic operations with bit-wise operations, which is expected to substantially improve power-efficiency. To validate the effectiveness of BNNs we conduct two sets of experiments on the Torch7 and Theano frameworks. On both, BNNs achieved nearly state-of-the-art results over the MNIST, CIFAR-10 and SVHN datasets. Last but not least, we wrote a binary matrix multiplication GPU kernel with which it is possible to run our MNIST BNN 7 times faster than with an unoptimized GPU kernel, without suffering any loss in classification accuracy. The code for training and running our BNNs is available on-line.", "Large CNNs have delivered impressive performance in various computer vision applications. But the storage and computation requirements make it problematic for deploying these models on mobile devices. Recently, tensor decompositions have been used for speeding up CNNs. In this paper, we further develop the tensor decomposition technique. We propose a new algorithm for computing the low-rank tensor decomposition for removing the redundancy in the convolution kernels. The algorithm finds the exact global optimizer of the decomposition and is more effective than iterative methods. Based on the decomposition, we further propose a new method for training low-rank constrained CNNs from scratch. Interestingly, while achieving a significant speedup, sometimes the low-rank constrained CNNs delivers significantly better performance than their non-constrained counterparts. On the CIFAR-10 dataset, the proposed low-rank NIN model achieves @math accuracy (without data augmentation), which also improves upon state-of-the-art result. We evaluated the proposed method on CIFAR-10 and ILSVRC12 datasets for a variety of modern CNNs, including AlexNet, NIN, VGG and GoogleNet with success. For example, the forward time of VGG-16 is reduced by half while the performance is still comparable. Empirical success suggests that low-rank tensor decompositions can be a very useful tool for speeding up large CNNs.", "A very simple way to improve the performance of almost any machine learning algorithm is to train many different models on the same data and then to average their predictions. Unfortunately, making predictions using a whole ensemble of models is cumbersome and may be too computationally expensive to allow deployment to a large number of users, especially if the individual models are large neural nets. Caruana and his collaborators have shown that it is possible to compress the knowledge in an ensemble into a single model which is much easier to deploy and we develop this approach further using a different compression technique. We achieve some surprising results on MNIST and we show that we can significantly improve the acoustic model of a heavily used commercial system by distilling the knowledge in an ensemble of models into a single model. We also introduce a new type of ensemble composed of one or more full models and many specialist models which learn to distinguish fine-grained classes that the full models confuse. Unlike a mixture of experts, these specialist models can be trained rapidly and in parallel." ] }
1905.09684	2945142306	This work addresses a new problem of learning generative adversarial networks (GANs) from multiple data collections that are each i) owned separately and privately by different clients and ii) drawn from a non-identical distribution that comprises different classes. Given such multi-client and non-iid data as input, we aim to achieve a distribution involving all the classes input data can belong to, while keeping the data decentralized and private in each client storage. Our key contribution to this end is a new decentralized approach for learning GANs from non-iid data called Forgiver-First Update (F2U), which a) asks clients to train an individual discriminator with their own data and b) updates a generator to fool the most forgiving' discriminators who deem generated samples as the most real. Our theoretical analysis proves that this updating strategy indeed allows the decentralized GAN to learn a generator's distribution with all the input classes as its global optimum based on f-divergence minimization. Moreover, we propose a relaxed version of F2U called Forgiver-First Aggregation (F2A), which adaptively aggregates the discriminators while emphasizing forgiving ones to perform well in practice. Our empirical evaluations with image generation tasks demonstrated the effectiveness of our approach over state-of-the-art decentralized learning methods.	Our work on decentralized learning of GANs has a connection to existing literature in several aspects. In terms of the formulation of GANs, recent work has also tried to involve multiple discriminators and or multiple generators. The motivations behind such works are, however, not to enable decentralized learning for multi-client data but to stabilize the training process @cite_5 @cite_14 , to avoid mode collapses @cite_10 @cite_1 , or to model multi-domain data @cite_20 @cite_16 @cite_38 , with the centralized setting.	{ "cite_N": [ "@cite_38", "@cite_14", "@cite_1", "@cite_5", "@cite_16", "@cite_10", "@cite_20" ], "mid": [ "2963784072", "2963706720", "2962928367", "2554506842", "2768626898", "2962900302", "2962793481" ], "abstract": [ "We propose coupled generative adversarial network (CoGAN) for learning a joint distribution of multi-domain images. In contrast to the existing approaches, which require tuples of corresponding images in different domains in the training set, CoGAN can learn a joint distribution without any tuple of corresponding images. It can learn a joint distribution with just samples drawn from the marginal distributions. This is achieved by enforcing a weight-sharing constraint that limits the network capacity and favors a joint distribution solution over a product of marginal distributions one. We apply CoGAN to several joint distribution learning tasks, including learning a joint distribution of color and depth images, and learning a joint distribution of face images with different attributes. For each task it successfully learns the joint distribution without any tuple of corresponding images. We also demonstrate its applications to domain adaptation and image transformation.", "The Generative Adversarial Networks (GANs) have demonstrated impressive performance for data synthesis, and are now used in a wide range of computer vision tasks. In spite of this success, they gained a reputation for being difficult to train, what results in a time-consuming and human-involved development process to use them. We consider an alternative training process, named SGAN, in which several adversarial \"local\" pairs of networks are trained independently so that a \"global\" supervising pair of networks can be trained against them. The goal is to train the global pair with the corresponding ensemble opponent for improved performances in terms of mode coverage. This approach aims at increasing the chances that learning will not stop for the global pair, preventing both to be trapped in an unsatisfactory local minimum, or to face oscillations often observed in practice. To guarantee the latter, the global pair never affects the local ones. The rules of SGAN training are thus as follows: the global generator and discriminator are trained using the local discriminators and generators, respectively, whereas the local networks are trained with their fixed local opponent. Experimental results on both toy and real-world problems demonstrate that this approach outperforms standard training in terms of better mitigating mode collapse, stability while converging and that it surprisingly, increases the convergence speed as well.", "We propose MAD-GAN, an intuitive generalization to the Generative Adversarial Networks (GANs) and its conditional variants to address the well known problem of mode collapse. First, MAD-GAN is a multi-agent GAN architecture incorporating multiple generators and one discriminator. Second, to enforce that different generators capture diverse high probability modes, the discriminator of MAD-GAN is designed such that along with finding the real and fake samples, it is also required to identify the generator that generated the given fake sample. Intuitively, to succeed in this task, the discriminator must learn to push different generators towards different identifiable modes. We perform extensive experiments on synthetic and real datasets and compare MAD-GAN with different variants of GAN. We show high quality diverse sample generations for challenging tasks such as image-to-image translation and face generation. In addition, we also show that MAD-GAN is able to disentangle different modalities when trained using highly challenging diverse-class dataset (e.g. dataset with images of forests, icebergs, and bedrooms). In the end, we show its efficacy on the unsupervised feature representation task.", "Generative adversarial networks (GANs) are a framework for producing a generative model by way of a two-player minimax game. In this paper, we propose the (GMAN), a framework that extends GANs to multiple discriminators. In previous work, the successful training of GANs requires modifying the minimax objective to accelerate training early on. In contrast, GMAN can be reliably trained with the original, untampered objective. We explore a number of design perspectives with the discriminator role ranging from formidable adversary to forgiving teacher. Image generation tasks comparing the proposed framework to standard GANs demonstrate GMAN produces higher quality samples in a fraction of the iterations when measured by a pairwise GAM-type metric.", "Recent studies have shown remarkable success in image-to-image translation for two domains. However, existing approaches have limited scalability and robustness in handling more than two domains, since different models should be built independently for every pair of image domains. To address this limitation, we propose StarGAN, a novel and scalable approach that can perform image-to-image translations for multiple domains using only a single model. Such a unified model architecture of StarGAN allows simultaneous training of multiple datasets with different domains within a single network. This leads to StarGAN's superior quality of translated images compared to existing models as well as the novel capability of flexibly translating an input image to any desired target domain. We empirically demonstrate the effectiveness of our approach on a facial attribute transfer and a facial expression synthesis tasks.", "We propose in this paper a novel approach to tackle the problem of mode collapse encountered in generative adversarial network (GAN). Our idea is intuitive but proven to be very effective, especially in addressing some key limitations of GAN. In essence, it combines the Kullback-Leibler (KL) and reverse KL divergences into a unified objective function, thus it exploits the complementary statistical properties from these divergences to effectively diversify the estimated density in capturing multi-modes. We term our method dual discriminator generative adversarial nets (D2GAN) which, unlike GAN, has two discriminators; and together with a generator, it also has the analogy of a minimax game, wherein a discriminator rewards high scores for samples from data distribution whilst another discriminator, conversely, favoring data from the generator, and the generator produces data to fool both two discriminators. We develop theoretical analysis to show that, given the maximal discriminators, optimizing the generator of D2GAN reduces to minimizing both KL and reverse KL divergences between data distribution and the distribution induced from the data generated by the generator, hence effectively avoiding the mode collapsing problem. We conduct extensive experiments on synthetic and real-world large-scale datasets (MNIST, CIFAR-10, STL-10, ImageNet), where we have made our best effort to compare our D2GAN with the latest state-of-the-art GAN's variants in comprehensive qualitative and quantitative evaluations. The experimental results demonstrate the competitive and superior performance of our approach in generating good quality and diverse samples over baselines, and the capability of our method to scale up to ImageNet database.", "Image-to-image translation is a class of vision and graphics problems where the goal is to learn the mapping between an input image and an output image using a training set of aligned image pairs. However, for many tasks, paired training data will not be available. We present an approach for learning to translate an image from a source domain X to a target domain Y in the absence of paired examples. Our goal is to learn a mapping G : X → Y such that the distribution of images from G(X) is indistinguishable from the distribution Y using an adversarial loss. Because this mapping is highly under-constrained, we couple it with an inverse mapping F : Y → X and introduce a cycle consistency loss to push F(G(X)) ≈ X (and vice versa). Qualitative results are presented on several tasks where paired training data does not exist, including collection style transfer, object transfiguration, season transfer, photo enhancement, etc. Quantitative comparisons against several prior methods demonstrate the superiority of our approach." ] }
1905.09684	2945142306	This work addresses a new problem of learning generative adversarial networks (GANs) from multiple data collections that are each i) owned separately and privately by different clients and ii) drawn from a non-identical distribution that comprises different classes. Given such multi-client and non-iid data as input, we aim to achieve a distribution involving all the classes input data can belong to, while keeping the data decentralized and private in each client storage. Our key contribution to this end is a new decentralized approach for learning GANs from non-iid data called Forgiver-First Update (F2U), which a) asks clients to train an individual discriminator with their own data and b) updates a generator to fool the most forgiving' discriminators who deem generated samples as the most real. Our theoretical analysis proves that this updating strategy indeed allows the decentralized GAN to learn a generator's distribution with all the input classes as its global optimum based on f-divergence minimization. Moreover, we propose a relaxed version of F2U called Forgiver-First Aggregation (F2A), which adaptively aggregates the discriminators while emphasizing forgiving ones to perform well in practice. Our empirical evaluations with image generation tasks demonstrated the effectiveness of our approach over state-of-the-art decentralized learning methods.	To the best of our knowledge, our work is the first to address the problem of unsupervised decentralized learning from non-iid data. Much work has focused on how computations for large-scale training can be decentralized to multiple clients (see @cite_28 , @cite_35 , and @cite_26 for a summary of recent work). Among these studies, the most relevant approach is federated learning @cite_2 that addressed the problem of learning from non-iid data. More recent work has tried to make federated learning more communication efficient @cite_36 @cite_29 @cite_0 , secure @cite_9 @cite_27 , and applicable to a practical wireless setting @cite_8 @cite_41 , but under the setting of standard supervised learning. One exception presented recently is MD-GAN @cite_17 , which however worked only when client data were iid. Another interesting attempt on the decentralized learning of GANs but with a different objective was @cite_23 , which tried to allow clients to jointly' learn their own GANs by exchanging discriminators.	{ "cite_N": [ "@cite_35", "@cite_26", "@cite_8", "@cite_28", "@cite_36", "@cite_29", "@cite_9", "@cite_41", "@cite_0", "@cite_27", "@cite_23", "@cite_2", "@cite_17" ], "mid": [ "2913570153", "2788193959", "", "2963228337", "2535838896", "", "2767079719", "2793925626", "2903471046", "2810065831", "2889968450", "", "2899730059" ], "abstract": [ "Federated Learning is a distributed machine learning approach which enables model training on a large corpus of decentralized data. We have built a scalable production system for Federated Learning in the domain of mobile devices, based on TensorFlow. In this paper, we describe the resulting high-level design, sketch some of the challenges and their solutions, and touch upon the open problems and future directions.", "Deep Neural Networks (DNNs) are becoming an important tool in modern computing applications. Accelerating their training is a major challenge and techniques range from distributed algorithms to low-level circuit design. In this survey, we describe the problem from a theoretical perspective, followed by approaches for its parallelization. Specifically, we present trends in DNN architectures and the resulting implications on parallelization strategies. We discuss the different types of concurrency in DNNs; synchronous and asynchronous stochastic gradient descent; distributed system architectures; communication schemes; and performance modeling. Based on these approaches, we extrapolate potential directions for parallelism in deep learning.", "", "Most distributed machine learning systems nowadays, including TensorFlow and CNTK, are built in a centralized fashion. One bottleneck of centralized algorithms lies on high communication cost on the central node. Motivated by this, we ask, can decentralized algorithms be faster than its centralized counterpart? Although decentralized PSGD (D-PSGD) algorithms have been studied by the control community, existing analysis and theory do not show any advantage over centralized PSGD (C-PSGD) algorithms, simply assuming the application scenario where only the decentralized network is available. In this paper, we study a D-PSGD algorithm and provide the first theoretical analysis that indicates a regime in which decentralized algorithms might outperform centralized algorithms for distributed stochastic gradient descent. This is because D-PSGD has comparable total computational complexities to C-PSGD but requires much less communication cost on the busiest node. We further conduct an empirical study to validate our theoretical analysis across multiple frameworks (CNTK and Torch), different network configurations, and computation platforms up to 112 GPUs. On network configurations with low bandwidth or high latency, D-PSGD can be up to one order of magnitude faster than its well-optimized centralized counterparts.", "Federated Learning is a machine learning setting where the goal is to train a high-quality centralized model while training data remains distributed over a large number of clients each with unreliable and relatively slow network connections. We consider learning algorithms for this setting where on each round, each client independently computes an update to the current model based on its local data, and communicates this update to a central server, where the client-side updates are aggregated to compute a new global model. The typical clients in this setting are mobile phones, and communication efficiency is of the utmost importance. In this paper, we propose two ways to reduce the uplink communication costs: structured updates, where we directly learn an update from a restricted space parametrized using a smaller number of variables, e.g. either low-rank or a random mask; and sketched updates, where we learn a full model update and then compress it using a combination of quantization, random rotations, and subsampling before sending it to the server. Experiments on both convolutional and recurrent networks show that the proposed methods can reduce the communication cost by two orders of magnitude.", "", "We design a novel, communication-efficient, failure-robust protocol for secure aggregation of high-dimensional data. Our protocol allows a server to compute the sum of large, user-held data vectors from mobile devices in a secure manner (i.e. without learning each user's individual contribution), and can be used, for example, in a federated learning setting, to aggregate user-provided model updates for a deep neural network. We prove the security of our protocol in the honest-but-curious and active adversary settings, and show that security is maintained even if an arbitrarily chosen subset of users drop out at any time. We evaluate the efficiency of our protocol and show, by complexity analysis and a concrete implementation, that its runtime and communication overhead remain low even on large data sets and client pools. For 16-bit input values, our protocol offers $1.73 x communication expansion for 210 users and 220-dimensional vectors, and 1.98 x expansion for 214 users and 224-dimensional vectors over sending data in the clear.", "Emerging technologies and applications including Internet of Things (IoT), social networking, and crowd-sourcing generate large amounts of data at the network edge. Machine learning models are often built from the collected data, to enable the detection, classification, and prediction of future events. Due to bandwidth, storage, and privacy concerns, it is often impractical to send all the data to a centralized location. In this paper, we consider the problem of learning model parameters from data distributed across multiple edge nodes, without sending raw data to a centralized place. Our focus is on a generic class of machine learning models that are trained using gradient-descent based approaches. We analyze the convergence rate of distributed gradient descent from a theoretical point of view, based on which we propose a control algorithm that determines the best trade-off between local update and global parameter aggregation to minimize the loss function under a given resource budget. The performance of the proposed algorithm is evaluated via extensive experiments with real datasets, both on a networked prototype system and in a larger-scale simulated environment. The experimentation results show that our proposed approach performs near to the optimum with various machine learning models and different data distributions.", "On-device machine learning (ML) enables the training process to exploit a massive amount of user-generated private data samples. To enjoy this benefit, inter-device communication overhead should be minimized. With this end, we propose federated distillation (FD), a distributed model training algorithm whose communication payload size is much smaller than a benchmark scheme, federated learning (FL), particularly when the model size is large. Moreover, user-generated data samples are likely to become non-IID across devices, which commonly degrades the performance compared to the case with an IID dataset. To cope with this, we propose federated augmentation (FAug), where each device collectively trains a generative model, and thereby augments its local data towards yielding an IID dataset. Empirical studies demonstrate that FD with FAug yields around 26x less communication overhead while achieving 95-98 test accuracy compared to FL.", "Federated learning enables multiple participants to jointly construct a deep learning model without sharing their private training data with each other. For example, multiple smartphones can jointly train a predictive keyboard model without revealing what individual users type into their phones. We demonstrate that any participant in federated learning can introduce hidden backdoor functionality into the joint global model, e.g., to ensure that an image classifier assigns an attacker-chosen label to images with certain features, or that a next-word predictor completes certain sentences with an attacker-chosen word. We design and evaluate a new \"constrain-and-scale\" model-poisoning methodology and show that it greatly outperforms data poisoning. An attacker selected just once, in a single round of federated learning, can cause the global model to reach 100 accuracy on the backdoor task. We evaluate the attack under different assumptions and attack scenarios for standard federated learning tasks. We also show how to evade anomaly detection-based defenses by incorporating the evasion into the loss function when training the attack model.", "Decentralized computation is crucial for training on large data sets, which are stored in various locations. This paper proposes a method for collaboratively training generative adversarial networks (GANs) on several nodes to approach this problem. Each node has access to its local private dataset and model, but can influence and can be influenced by neighboring nodes. Such an approach avoids the need of a fusion center, offers better network load balance and improves data privacy. This work proposes an initial approach for decentralized GAN network optimization which is based on discriminator exchange among neighbors using an information gain criterion. The initial experiments are presented to verify whether such a decentralized architecture can provide useful results.", "", "A recent technical breakthrough in the domain of machine learning is the discovery and the multiple applications of Generative Adversarial Networks (GANs). Those generative models are computationally demanding, as a GAN is composed of two deep neural networks, and because it trains on large datasets. A GAN is generally trained on a single server. In this paper, we address the problem of distributing GANs so that they are able to train over datasets that are spread on multiple workers. MD-GAN is exposed as the first solution for this problem: we propose a novel learning procedure for GANs so that they fit this distributed setup. We then compare the performance of MD-GAN to an adapted version of Federated Learning to GANs, using the MNIST and CIFAR10 datasets. MD-GAN exhibits a reduction by a factor of two of the learning complexity on each worker node, while providing better performances than federated learning on both datasets. We finally discuss the practical implications of distributing GANs." ] }
1905.09684	2945142306	This work addresses a new problem of learning generative adversarial networks (GANs) from multiple data collections that are each i) owned separately and privately by different clients and ii) drawn from a non-identical distribution that comprises different classes. Given such multi-client and non-iid data as input, we aim to achieve a distribution involving all the classes input data can belong to, while keeping the data decentralized and private in each client storage. Our key contribution to this end is a new decentralized approach for learning GANs from non-iid data called Forgiver-First Update (F2U), which a) asks clients to train an individual discriminator with their own data and b) updates a generator to fool the most forgiving' discriminators who deem generated samples as the most real. Our theoretical analysis proves that this updating strategy indeed allows the decentralized GAN to learn a generator's distribution with all the input classes as its global optimum based on f-divergence minimization. Moreover, we propose a relaxed version of F2U called Forgiver-First Aggregation (F2A), which adaptively aggregates the discriminators while emphasizing forgiving ones to perform well in practice. Our empirical evaluations with image generation tasks demonstrated the effectiveness of our approach over state-of-the-art decentralized learning methods.	Finally, our work has several limitations. i) Our empirical evaluation is based on a simulation in a single workstation. Practical implementations of the proposed approach as well as other baselines will come with problems of communication, security, and scalability for a large number of clients, as discussed in @cite_2 . ii) Our approach by itself is not designed to resolve common problems observed in GAN training, such as mode collapse and theoretical guarantee for convergence. We still require additional contributions to make GANs perform well in practice. iii) The current formulation can be applied only to standard GANs with a generator and a discriminator. One interesting extension for future work is to deal with various architectures such as conditional GANs @cite_40 @cite_44 @cite_3 @cite_32 @cite_6 and GANs with multiple generators @cite_20 @cite_16 @cite_14 @cite_1 @cite_38 .	{ "cite_N": [ "@cite_38", "@cite_14", "@cite_1", "@cite_32", "@cite_3", "@cite_6", "@cite_44", "@cite_40", "@cite_2", "@cite_16", "@cite_20" ], "mid": [ "2963784072", "2963706720", "2962928367", "2548275288", "2405756170", "2962754210", "2125389028", "2963073614", "", "2768626898", "2962793481" ], "abstract": [ "We propose coupled generative adversarial network (CoGAN) for learning a joint distribution of multi-domain images. In contrast to the existing approaches, which require tuples of corresponding images in different domains in the training set, CoGAN can learn a joint distribution without any tuple of corresponding images. It can learn a joint distribution with just samples drawn from the marginal distributions. This is achieved by enforcing a weight-sharing constraint that limits the network capacity and favors a joint distribution solution over a product of marginal distributions one. We apply CoGAN to several joint distribution learning tasks, including learning a joint distribution of color and depth images, and learning a joint distribution of face images with different attributes. For each task it successfully learns the joint distribution without any tuple of corresponding images. We also demonstrate its applications to domain adaptation and image transformation.", "The Generative Adversarial Networks (GANs) have demonstrated impressive performance for data synthesis, and are now used in a wide range of computer vision tasks. In spite of this success, they gained a reputation for being difficult to train, what results in a time-consuming and human-involved development process to use them. We consider an alternative training process, named SGAN, in which several adversarial \"local\" pairs of networks are trained independently so that a \"global\" supervising pair of networks can be trained against them. The goal is to train the global pair with the corresponding ensemble opponent for improved performances in terms of mode coverage. This approach aims at increasing the chances that learning will not stop for the global pair, preventing both to be trapped in an unsatisfactory local minimum, or to face oscillations often observed in practice. To guarantee the latter, the global pair never affects the local ones. The rules of SGAN training are thus as follows: the global generator and discriminator are trained using the local discriminators and generators, respectively, whereas the local networks are trained with their fixed local opponent. Experimental results on both toy and real-world problems demonstrate that this approach outperforms standard training in terms of better mitigating mode collapse, stability while converging and that it surprisingly, increases the convergence speed as well.", "We propose MAD-GAN, an intuitive generalization to the Generative Adversarial Networks (GANs) and its conditional variants to address the well known problem of mode collapse. First, MAD-GAN is a multi-agent GAN architecture incorporating multiple generators and one discriminator. Second, to enforce that different generators capture diverse high probability modes, the discriminator of MAD-GAN is designed such that along with finding the real and fake samples, it is also required to identify the generator that generated the given fake sample. Intuitively, to succeed in this task, the discriminator must learn to push different generators towards different identifiable modes. We perform extensive experiments on synthetic and real datasets and compare MAD-GAN with different variants of GAN. We show high quality diverse sample generations for challenging tasks such as image-to-image translation and face generation. In addition, we also show that MAD-GAN is able to disentangle different modalities when trained using highly challenging diverse-class dataset (e.g. dataset with images of forests, icebergs, and bedrooms). In the end, we show its efficacy on the unsupervised feature representation task.", "Synthesizing high resolution photorealistic images has been a long-standing challenge in machine learning. In this paper we introduce new methods for the improved training of generative adversarial networks (GANs) for image synthesis. We construct a variant of GANs employing label conditioning that results in 128x128 resolution image samples exhibiting global coherence. We expand on previous work for image quality assessment to provide two new analyses for assessing the discriminability and diversity of samples from class-conditional image synthesis models. These analyses demonstrate that high resolution samples provide class information not present in low resolution samples. Across 1000 ImageNet classes, 128x128 samples are more than twice as discriminable as artificially resized 32x32 samples. In addition, 84.7 of the classes have samples exhibiting diversity comparable to real ImageNet data.", "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.", "We propose a novel, projection based way to incorporate the conditional information into the discriminator of GANs that respects the role of the conditional information in the underlining probabilistic model. This approach is in contrast with most frameworks of conditional GANs used in application today, which use the conditional information by concatenating the (embedded) conditional vector to the feature vectors. With this modification, we were able to significantly improve the quality of the class conditional image generation on ILSVRC2012 (ImageNet) dataset from the current state-of-the-art result, and we achieved this with a single pair of a discriminator and a generator. We were also able to extend the application to super-resolution and succeeded in producing highly discriminative super-resolution images. This new structure also enabled high quality category transformation based on parametric functional transformation of conditional batch normalization layers in the generator.", "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. Moreover, since the release of the pix2pix software associated with this paper, hundreds of twitter users have posted their own artistic experiments using our system. As a community, we no longer hand-engineer our mapping functions, and this work suggests we can achieve reasonable results without handengineering our loss functions either.", "", "Recent studies have shown remarkable success in image-to-image translation for two domains. However, existing approaches have limited scalability and robustness in handling more than two domains, since different models should be built independently for every pair of image domains. To address this limitation, we propose StarGAN, a novel and scalable approach that can perform image-to-image translations for multiple domains using only a single model. Such a unified model architecture of StarGAN allows simultaneous training of multiple datasets with different domains within a single network. This leads to StarGAN's superior quality of translated images compared to existing models as well as the novel capability of flexibly translating an input image to any desired target domain. We empirically demonstrate the effectiveness of our approach on a facial attribute transfer and a facial expression synthesis tasks.", "Image-to-image translation is a class of vision and graphics problems where the goal is to learn the mapping between an input image and an output image using a training set of aligned image pairs. However, for many tasks, paired training data will not be available. We present an approach for learning to translate an image from a source domain X to a target domain Y in the absence of paired examples. Our goal is to learn a mapping G : X → Y such that the distribution of images from G(X) is indistinguishable from the distribution Y using an adversarial loss. Because this mapping is highly under-constrained, we couple it with an inverse mapping F : Y → X and introduce a cycle consistency loss to push F(G(X)) ≈ X (and vice versa). Qualitative results are presented on several tasks where paired training data does not exist, including collection style transfer, object transfiguration, season transfer, photo enhancement, etc. Quantitative comparisons against several prior methods demonstrate the superiority of our approach." ] }
1905.09760	2946338065	In many domains, software systems cannot be deployed until authorities judge them fit for use in an intended operating environment. Certification standards and processes have been devised and deployed to regulate operations of software systems and prevent their failures. However, practitioners are often unsatisfied with the efficiency and value proposition of certification efforts. In this study, we compare two certification standards, Common Criteria and DO-178C, and collect insights from literature and from interviews with subject-matter experts to identify design options relevant to the design of standards. The results of the comparison of certification efforts---leading to the identification of design dimensions that affect their quality---serve as a framework to guide the comparison, creation, and revision of certification standards and processes. This paper puts software engineering research in context and discusses key issues around process and quality assurance and includes observations from industry about relevant topics such as recertification, timely evaluations, but also technical discussions around model-driven approaches and formal methods. Our initial characterization of the design space of certification efforts can be used to inform technical discussions and to influence the directions of new or existing certification efforts. Practitioners, technical commissions, and government can directly benefit from our analytical framework.	There are also papers that perform standards comparison, but much more focused and fine-grained. compared both IEC 60880 and IEC 61508 standards, aiming at identifying differences in strictness and scope between the shall'' requirements as defined by the two standards. The results indicate that such fine-grained comparison is difficult, but they could still identify that even though the two standards are fairly similar, they complement each other and should be used together to achieve higher safety assurance. @cite_18 investigated differences in certifying a wheel braking system when using a prescriptive approach (DO-178C) and when using assurance cases. The authors suggest linking assurance cases and DO-178C processes to increase safety assurance, since using assurance cases alone might be challenging, especially for vendors without experience in developing cases. Our study complements these papers by contrasting two distinct standards, which contributes to a broader discussion about certification efforts across communities. We are not aware of any other papers that compare standards across communities, focusing on contrasting cases to confront community-specific culture.	{ "cite_N": [ "@cite_18" ], "mid": [ "2157607087" ], "abstract": [ "In safety–critical applications, it is necessary to justify, prior to deployment, why software behaviour is to be trusted. This is normally referred to as software safety assurance. Within certification standards, developers demonstrate this by appealing to the satisfaction of objectives that the safety assurance standards require for compliance. In some standards the objectives can be very detailed in nature, prescribing specific processes and techniques that must be followed. This approach to certification is often described as prescriptive or process-based certification. Other standards set out much more high-level objectives and are less prescriptive about the particular processes and techniques to be used. These standards instead explicitly require the submission of an assurance argument which communicates how evidence, generated during development (for example from testing, analysis and review) satisfies claims concerning the safety of the software. There has been much debate surrounding the relative merits of prescriptive and safety assurance argument approaches to certification. In many ways this debate can lead to confusion. There can in fact be seen to be a role for both approaches in a successful software assurance regime. In this paper, we provide a comparative examination of these two approaches, and seek to identify the relative merits of each. We first introduce the concepts of assurance cases and prescriptive software assurance. We describe how an assurance case could be generated for the software of an aircraft wheel braking system. We then describe how prescriptive certification guidelines could be used in order to gain assurance in the same system. Finally, we compare the results of the two approaches and explain how these approaches may complement each other. This comparison highlights the crucial role that an assurance argument can play in explaining and justifying how the software evidence supports the assurance argument, even when a prescriptive safety standard is being followed." ] }
1905.09760	2946338065	In many domains, software systems cannot be deployed until authorities judge them fit for use in an intended operating environment. Certification standards and processes have been devised and deployed to regulate operations of software systems and prevent their failures. However, practitioners are often unsatisfied with the efficiency and value proposition of certification efforts. In this study, we compare two certification standards, Common Criteria and DO-178C, and collect insights from literature and from interviews with subject-matter experts to identify design options relevant to the design of standards. The results of the comparison of certification efforts---leading to the identification of design dimensions that affect their quality---serve as a framework to guide the comparison, creation, and revision of certification standards and processes. This paper puts software engineering research in context and discusses key issues around process and quality assurance and includes observations from industry about relevant topics such as recertification, timely evaluations, but also technical discussions around model-driven approaches and formal methods. Our initial characterization of the design space of certification efforts can be used to inform technical discussions and to influence the directions of new or existing certification efforts. Practitioners, technical commissions, and government can directly benefit from our analytical framework.	Design spaces have been widely studied by the design community, where exploring alternatives more systematically is considered good practice @cite_47 @cite_1 @cite_44 . Designers are often described as explorers that use their abilities to assess design options and make informed decisions based on the trade-offs that these options entail. With our study, we attempt to provide an initial contribution to building a design space for software certification standards, so that stakeholders can also make informed decisions when creating or updating standards, empowering them to reason more systematically about the options in the space, similar to how many other areas have done in the past with architectural styles @cite_77 , computer input devices @cite_72 @cite_4 , wireless communication @cite_70 , 3D architectures @cite_30 , and self-adaptive systems @cite_0 .	{ "cite_N": [ "@cite_30", "@cite_4", "@cite_70", "@cite_1", "@cite_44", "@cite_77", "@cite_0", "@cite_72", "@cite_47" ], "mid": [ "1987131925", "", "2137486085", "", "", "2150889907", "2042259499", "2000823235", "2081017800" ], "abstract": [ "As technology scales, interconnects have become a major performance bottleneck and a major source of power consumption for microprocessors. Increasing interconnect costs make it necessary to consider alternate ways of building modern microprocessors. One promising option is 3D architectures where a stack of multiple device layers with direct vertical tunneling through them are put together on the same chip. As fabrication of 3D integrated circuits has become viable, developing CAD tools and architectural techniques is imperative to explore the design space to 3D microarchitectures. In this article, we give a brief introduction to 3D integration technology, discuss the EDA design tools that can enable the adoption of 3D ICs, and present the implementation of various microprocessor components using 3D technology. An industrial case study is presented as an initial attempt to design 3D microarchitectures.", "", "In the recent past, wireless sensor networks have found their way into a wide variety of applications and systems with vastly varying requirements and characteristics. As a consequence, it is becoming increasingly difficult to discuss typical requirements regarding hardware issues and software support. This is particularly problematic in a multidisciplinary research area such as wireless sensor networks, where close collaboration between users, application domain experts, hardware designers, and software developers is needed to implement efficient systems. In this article we discuss the consequences of this fact with regard to the design space of wireless sensor networks by considering its various dimensions. We justify our view by demonstrating that specific existing applications occupy different points in the design space.", "", "", "Software architects use a number of commonly-recognized \"styles\" to guide their design of system structures. Each of these is appropriate for some classes of problems, but none is suitable for all problems. How, then, does a software designer choose an architecture suitable for the problem at hand? Two kinds of information are required: (1) careful discrimination among the candidate architectures and (2) design guidance on how to make appropriate choices. In this paper, we support careful discrimination with a preliminary classification of styles. We use a 2D classification strategy with control and data issues as the dominant organizing axes. We position the major styles within this space and use finer-grained discriminations to elaborate variations on the styles. This provides a framework for organizing design guidance, which we partially flesh out with rules of thumb.", "Over the last 100 years it has become much easier to operate a car. However in recent years the number of functions a user can control while driving has greatly increased. Infotainment, entertainment and comfort systems as well as driver assistance contribute to this trend. Interaction with these systems plays an important role, as on one hand this can improve the user experience while driving but on the other hand it may distract from the primary task of driving. User interfaces in cars differ regarding the number of input and output devices and their placement in the car to a great extent. In this paper, we introduce a first design space for driver-based automotive user interfaces that allows a comprehensive description of input and output devices in a car with regard to placement and modality. This design space is intended to provide a basis for analyzing and discussing different user interface arrangements in cars, to compare alternative user interface setups, and to identify new opportunities for interaction and placement of controls. We present a graphical representation of the design space and discuss its usage in detail based on several examples. To assess the completeness of the proposed design space we used it to classify and compare user interfaces from more than 100 cars shown at IAA2007, cars from the BMW museum, and from the A2Mac1 image database.", "A bewildering variety of devices for communication from humans to computers now exists on the market. In order to make sense of this variety, and to aid in the design of new input devices, we propose a framework for describing and analyzing input devices. Following Mackinlay's semantic analysis of the design space for graphical presentations, our goal is to provide tools for the generation and test of input device designs. The descriptive tools we have created allow us to describe the semantics of a device and measure its expressiveness . Using these tools, we have built a taxonomy of input devices that goes beyond earlier taxonomies of Buxton & Baecker and Foley, Wallace, & Chan. In this paper, we build on these descriptive tools, and proceed to the use of human performance theories and data for evaluation of the effectiveness of points in this design space. We focus on two figures of merit, footprint and bandwidth, to illustrate this evaluation. The result is the systematic integration of methods for both generating and testing the design space of input devices.", "A central task in design is deciding what artifact will best satisfy the client's needs, whether that requires creating an artifact or choosing from existing alternatives. A design space identifies and organizes the decisions that must be made, together with the alternatives for those decisions, thereby providing guidance for creating artifacts or a framework for comparing them. The Studying Professional Software Design workshop studied three pairs of professional software designers sketching designs for a traffic signal simulator. A discussion of the design space for the simulation task shows how this design space enables comparison of the designs. It also illustrates the benefits of explicitly considering the design space during design and the risks of failing to do so." ] }
1905.09642	2949135528	Text classification approaches have usually required task-specific model architectures and huge labeled datasets. Recently, thanks to the rise of text-based transfer learning techniques, it is possible to pre-train a language model in an unsupervised manner and leverage them to perform effective on downstream tasks. In this work we focus on Japanese and show the potential use of transfer learning techniques in text classification. Specifically, we perform binary and multi-class sentiment classification on the Rakuten product review and Yahoo movie review datasets. We show that transfer learning-based approaches perform better than task-specific models trained on 3 times as much data. Furthermore, these approaches perform just as well for language modeling pre-trained on only 1 30 of the data. We release our pre-trained models and code as open source.	Word embedding is defined as the representation of a word as a dense vector. There have been many neural network implementations, including word2vec @cite_16 , fasttext @cite_12 and Glove @cite_20 that embed using a single layer and achieve state-of-the-art performance in various NLP tasks. However, these embeddings are not context-specific: in the phrases "I washed my dish" and "I ate my dish", the word "dish" refers to different things but are still represented by the same embedding.	{ "cite_N": [ "@cite_16", "@cite_20", "@cite_12" ], "mid": [ "1614298861", "2250539671", "2468328197" ], "abstract": [ "", "Recent methods for learning vector space representations of words have succeeded in capturing fine-grained semantic and syntactic regularities using vector arithmetic, but the origin of these regularities has remained opaque. We analyze and make explicit the model properties needed for such regularities to emerge in word vectors. The result is a new global logbilinear regression model that combines the advantages of the two major model families in the literature: global matrix factorization and local context window methods. Our model efficiently leverages statistical information by training only on the nonzero elements in a word-word cooccurrence matrix, rather than on the entire sparse matrix or on individual context windows in a large corpus. The model produces a vector space with meaningful substructure, as evidenced by its performance of 75 on a recent word analogy task. It also outperforms related models on similarity tasks and named entity recognition.", "This paper explores a simple and efficient baseline for text classification. Our experiments show that our fast text classifier fastText is often on par with deep learning classifiers in terms of accuracy, and many orders of magnitude faster for training and evaluation. We can train fastText on more than one billion words in less than ten minutes using a standard multicore CPU, and classify half a million sentences among 312K classes in less than a minute." ] }
1905.09642	2949135528	Text classification approaches have usually required task-specific model architectures and huge labeled datasets. Recently, thanks to the rise of text-based transfer learning techniques, it is possible to pre-train a language model in an unsupervised manner and leverage them to perform effective on downstream tasks. In this work we focus on Japanese and show the potential use of transfer learning techniques in text classification. Specifically, we perform binary and multi-class sentiment classification on the Rakuten product review and Yahoo movie review datasets. We show that transfer learning-based approaches perform better than task-specific models trained on 3 times as much data. Furthermore, these approaches perform just as well for language modeling pre-trained on only 1 30 of the data. We release our pre-trained models and code as open source.	Instead of fixed vector embeddings, Cove @cite_0 uses a machine translation model to embed each word within the context of its sentence. The model includes a bidirectional LSTM encoder and a unidirectional LSTM decoder with attention, and only the encoder is used for downstream task-specific models. However, pretraining is limited by the availability of parallel corpora. (e.g. English-French)	{ "cite_N": [ "@cite_0" ], "mid": [ "2963756346" ], "abstract": [ "Computer vision has benefited from initializing multiple deep layers with weights pretrained on large supervised training sets like ImageNet. Natural language processing (NLP) typically sees initialization of only the lowest layer of deep models with pretrained word vectors. In this paper, we use a deep LSTM encoder from an attentional sequence-to-sequence model trained for machine translation (MT) to contextualize word vectors. We show that adding these context vectors (CoVe) improves performance over using only unsupervised word and character vectors on a wide variety of common NLP tasks: sentiment analysis (SST, IMDb), question classification (TREC), entailment (SNLI), and question answering (SQuAD). For fine-grained sentiment analysis and entailment, CoVe improves performance of our baseline models to the state of the art." ] }
1905.09642	2949135528	Text classification approaches have usually required task-specific model architectures and huge labeled datasets. Recently, thanks to the rise of text-based transfer learning techniques, it is possible to pre-train a language model in an unsupervised manner and leverage them to perform effective on downstream tasks. In this work we focus on Japanese and show the potential use of transfer learning techniques in text classification. Specifically, we perform binary and multi-class sentiment classification on the Rakuten product review and Yahoo movie review datasets. We show that transfer learning-based approaches perform better than task-specific models trained on 3 times as much data. Furthermore, these approaches perform just as well for language modeling pre-trained on only 1 30 of the data. We release our pre-trained models and code as open source.	ELMo, short for Embeddings from Language Model @cite_7 overcomes this issue by taking advantage of large monolingual data in an unsupervised way. The core foundation of ELMo is the bidirectional language model which learns to predict the probability of a target word given a sentence by combining forward and backward language models. ELMo also requires task-specific models for downstream tasks.	{ "cite_N": [ "@cite_7" ], "mid": [ "2962739339" ], "abstract": [ "We introduce a new type of deep contextualized word representation that models both (1) complex characteristics of word use (e.g., syntax and semantics), and (2) how these uses vary across linguistic contexts (i.e., to model polysemy). Our word vectors are learned functions of the internal states of a deep bidirectional language model (biLM), which is pretrained on a large text corpus. We show that these representations can be easily added to existing models and significantly improve the state of the art across six challenging NLP problems, including question answering, textual entailment and sentiment analysis. We also present an analysis showing that exposing the deep internals of the pretrained network is crucial, allowing downstream models to mix different types of semi-supervision signals." ] }
1905.09432	2949110894	We address the problem of unsupervised disentanglement of discrete and continuous explanatory factors of data. We first show a simple procedure for minimizing the total correlation of the continuous latent variables without having to use a discriminator network or perform importance sampling, via cascading the information flow in the @math -vae framework. Furthermore, we propose a method which avoids offloading the entire burden of jointly modeling the continuous and discrete factors to the variational encoder by employing a separate discrete inference procedure. This leads to an interesting alternating minimization problem which switches between finding the most likely discrete configuration given the continuous factors and updating the variational encoder based on the computed discrete factors. Experiments show that the proposed method clearly disentangles discrete factors and significantly outperforms current disentanglement methods based on the disentanglement score and inference network classification score. The source code is available at this https URL.	The recent follow up works from then analyzed the KL divergence term under the expectation over the data distribution could be broken down into the mutual information term between the data and the latent code, and the KL divergence term between the latent distribution and the factorial prior often denoted as total correlation (TC) @cite_16 . The idea was that regularizing the KL divergence term with high @math in the VAE framework not only penalizes for TC but also inevitably penalizes the mutual information between the data and the latent variables.	{ "cite_N": [ "@cite_16" ], "mid": [ "2095439994" ], "abstract": [ "A set λ of stochastic variables, y1, y2,..., yn, is grouped into subsets, µ1, µ2,..., µk. The correlation existing in λ with respect to the µ's is adequately expressed by C= Σi=1k S(µi)-S(λ)≥0, where S(v) is the entropy function defined with reference to the variables y in subset v. For a given λ, C becomes maximum when each µi consists of only one variable, (n=k). The value Cis then called fhe total correlation in λ, Ctot(λ). The present paper gives various theorems, according to which Ctot(λ) can be decomposed in terms of the partial correlations existing in subsets of λ, and of quantities derivable therefrom. The information-theoretical meaning of each decomposition is carefully explained. As illustrations, two problems are discussed at the end of the paper: (1) redundancy in geometrical figures in pattern recognition, and (2) randomization effect of shuffling cards marked \"zero' or \"one.\"" ] }
1905.09432	2949110894	We address the problem of unsupervised disentanglement of discrete and continuous explanatory factors of data. We first show a simple procedure for minimizing the total correlation of the continuous latent variables without having to use a discriminator network or perform importance sampling, via cascading the information flow in the @math -vae framework. Furthermore, we propose a method which avoids offloading the entire burden of jointly modeling the continuous and discrete factors to the variational encoder by employing a separate discrete inference procedure. This leads to an interesting alternating minimization problem which switches between finding the most likely discrete configuration given the continuous factors and updating the variational encoder based on the computed discrete factors. Experiments show that the proposed method clearly disentangles discrete factors and significantly outperforms current disentanglement methods based on the disentanglement score and inference network classification score. The source code is available at this https URL.	To this end, the recent works proposed decreasing @math but more explicitly penalizing for TC. FactorVAE estimates the total correlation via density ratio trick which utilizes a discriminator network @cite_11 . @math -TCVAE employs mini-batch weighted sampling to estimate TC @cite_6 . On the contrary, we observe a factorization @math and show we can penalize TC without any explicit computation by incrementally penalizing each summand in the factorization by cascading the information flow.	{ "cite_N": [ "@cite_6", "@cite_11" ], "mid": [ "2787273002", "2785519580" ], "abstract": [ "We decompose the evidence lower bound to show the existence of a term measuring the total correlation between latent variables. We use this to motivate our @math -TCVAE (Total Correlation Variational Autoencoder), a refinement of the state-of-the-art @math -VAE objective for learning disentangled representations, requiring no additional hyperparameters during training. We further propose a principled classifier-free measure of disentanglement called the mutual information gap (MIG). We perform extensive quantitative and qualitative experiments, in both restricted and non-restricted settings, and show a strong relation between total correlation and disentanglement, when the latent variables model is trained using our framework.", "We define and address the problem of unsupervised learning of disentangled representations on data generated from independent factors of variation. We propose FactorVAE, a method that disentangles by encouraging the distribution of representations to be factorial and hence independent across the dimensions. We show that it improves upon @math -VAE by providing a better trade-off between disentanglement and reconstruction quality. Moreover, we highlight the problems of a commonly used disentanglement metric and introduce a new metric that does not suffer from them." ] }
1905.09432	2949110894	We address the problem of unsupervised disentanglement of discrete and continuous explanatory factors of data. We first show a simple procedure for minimizing the total correlation of the continuous latent variables without having to use a discriminator network or perform importance sampling, via cascading the information flow in the @math -vae framework. Furthermore, we propose a method which avoids offloading the entire burden of jointly modeling the continuous and discrete factors to the variational encoder by employing a separate discrete inference procedure. This leads to an interesting alternating minimization problem which switches between finding the most likely discrete configuration given the continuous factors and updating the variational encoder based on the computed discrete factors. Experiments show that the proposed method clearly disentangles discrete factors and significantly outperforms current disentanglement methods based on the disentanglement score and inference network classification score. The source code is available at this https URL.	On the other hand, NVIL @cite_0 and VIMCO @cite_4 have explored training VAEs with only discrete latent variables via REINFORCE @cite_20 with variance reduction techniques. VQVAE @cite_21 learns discrete latent variables via vector quantization. However, modeling via binary latent variables alone can be inappropriate since the underlying modalities would be a mix of both continuous and discrete factors for high dimensional complex data.	{ "cite_N": [ "@cite_0", "@cite_21", "@cite_4", "@cite_20" ], "mid": [ "2952264928", "2963799213", "2278112683", "2119717200" ], "abstract": [ "Highly expressive directed latent variable models, such as sigmoid belief networks, are difficult to train on large datasets because exact inference in them is intractable and none of the approximate inference methods that have been applied to them scale well. We propose a fast non-iterative approximate inference method that uses a feedforward network to implement efficient exact sampling from the variational posterior. The model and this inference network are trained jointly by maximizing a variational lower bound on the log-likelihood. Although the naive estimator of the inference model gradient is too high-variance to be useful, we make it practical by applying several straightforward model-independent variance reduction techniques. Applying our approach to training sigmoid belief networks and deep autoregressive networks, we show that it outperforms the wake-sleep algorithm on MNIST and achieves state-of-the-art results on the Reuters RCV1 document dataset.", "Learning useful representations without supervision remains a key challenge in machine learning. In this paper, we propose a simple yet powerful generative model that learns such discrete representations. Our model, the Vector Quantised-Variational AutoEncoder (VQ-VAE), differs from VAEs in two key ways: the encoder network outputs discrete, rather than continuous, codes; and the prior is learnt rather than static. In order to learn a discrete latent representation, we incorporate ideas from vector quantisation (VQ). Using the VQ method allows the model to circumvent issues of posterior collapse'' -— where the latents are ignored when they are paired with a powerful autoregressive decoder -— typically observed in the VAE framework. Pairing these representations with an autoregressive prior, the model can generate high quality images, videos, and speech as well as doing high quality speaker conversion and unsupervised learning of phonemes, providing further evidence of the utility of the learnt representations.", "Recent progress in deep latent variable models has largely been driven by the development of flexible and scalable variational inference methods. Variational training of this type involves maximizing a lower bound on the log-likelihood, using samples from the variational posterior to compute the required gradients. Recently, (2016) have derived a tighter lower bound using a multi-sample importance sampling estimate of the likelihood and showed that optimizing it yields models that use more of their capacity and achieve higher likelihoods. This development showed the importance of such multi-sample objectives and explained the success of several related approaches. We extend the multi-sample approach to discrete latent variables and analyze the difficulty encountered when estimating the gradients involved. We then develop the first unbiased gradient estimator designed for importance-sampled objectives and evaluate it at training generative and structured output prediction models. The resulting estimator, which is based on low-variance per-sample learning signals, is both simpler and more effective than the NVIL estimator proposed for the single-sample variational objective, and is competitive with the currently used biased estimators.", "This article presents a general class of associative reinforcement learning algorithms for connectionist networks containing stochastic units. These algorithms, called REINFORCE algorithms, are shown to make weight adjustments in a direction that lies along the gradient of expected reinforcement in both immediate-reinforcement tasks and certain limited forms of delayed-reinforcement tasks, and they do this without explicitly computing gradient estimates or even storing information from which such estimates could be computed. Specific examples of such algorithms are presented, some of which bear a close relationship to certain existing algorithms while others are novel but potentially interesting in their own right. Also given are results that show how such algorithms can be naturally integrated with backpropagation. We close with a brief discussion of a number of additional issues surrounding the use of such algorithms, including what is known about their limiting behaviors as well as further considerations that might be used to help develop similar but potentially more powerful reinforcement learning algorithms." ] }
1905.09432	2949110894	We address the problem of unsupervised disentanglement of discrete and continuous explanatory factors of data. We first show a simple procedure for minimizing the total correlation of the continuous latent variables without having to use a discriminator network or perform importance sampling, via cascading the information flow in the @math -vae framework. Furthermore, we propose a method which avoids offloading the entire burden of jointly modeling the continuous and discrete factors to the variational encoder by employing a separate discrete inference procedure. This leads to an interesting alternating minimization problem which switches between finding the most likely discrete configuration given the continuous factors and updating the variational encoder based on the computed discrete factors. Experiments show that the proposed method clearly disentangles discrete factors and significantly outperforms current disentanglement methods based on the disentanglement score and inference network classification score. The source code is available at this https URL.	Jointly modeling the continuous and discrete generative latent factors has been much less explored. InfoGAN @cite_19 models both the factors and is based on Generative Adversarial Network (GAN) framework @cite_14 . InfoGAN aims at disentangling the factors by maximizing the mutual information between a subset of latent dimensions and the generated samples. However, @cite_11 showed the learning process can be very unstable and significantly reduces the sample diversity in contrast to the @math -VAE framework @cite_8 . Empirically, InfoGAN tends to mix discrete and continuous factors which results in lower disentanglement score @cite_11 than @math -VAE based methods. JointVAE proposed jointly modeling both the continuous and discrete factors by augmenting the continuous and discrete latent variables. This introduces an additional KL divergence term for discrete latent variables which is optimized by a continuous reparameterization via Gumbel softmax trick @cite_17 @cite_5 @cite_1 . Our method, on the other hand, decouples the task of jointly modeling the continuous and discrete factors of data via alternating maximization and shows significant gains in the disentanglement performance.	{ "cite_N": [ "@cite_14", "@cite_11", "@cite_8", "@cite_1", "@cite_19", "@cite_5", "@cite_17" ], "mid": [ "2099471712", "2785519580", "2753738274", "2547875792", "2434741482", "2952165242", "2797569913" ], "abstract": [ "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.", "We define and address the problem of unsupervised learning of disentangled representations on data generated from independent factors of variation. We propose FactorVAE, a method that disentangles by encouraging the distribution of representations to be factorial and hence independent across the dimensions. We show that it improves upon @math -VAE by providing a better trade-off between disentanglement and reconstruction quality. Moreover, we highlight the problems of a commonly used disentanglement metric and introduce a new metric that does not suffer from them.", "Learning an interpretable factorised representation of the independent data generative factors of the world without supervision is an important precursor for the development of artificial intelligence that is able to learn and reason in the same way that humans do. We introduce beta-VAE, a new state-of-the-art framework for automated discovery of interpretable factorised latent representations from raw image data in a completely unsupervised manner. Our approach is a modification of the variational autoencoder (VAE) framework. We introduce an adjustable hyperparameter beta that balances latent channel capacity and independence constraints with reconstruction accuracy. We demonstrate that beta-VAE with appropriately tuned beta > 1 qualitatively outperforms VAE (beta = 1), as well as state of the art unsupervised (InfoGAN) and semi-supervised (DC-IGN) approaches to disentangled factor learning on a variety of datasets (celebA, faces and chairs). Furthermore, we devise a protocol to quantitatively compare the degree of disentanglement learnt by different models, and show that our approach also significantly outperforms all baselines quantitatively. Unlike InfoGAN, beta-VAE is stable to train, makes few assumptions about the data and relies on tuning a single hyperparameter, which can be directly optimised through a hyper parameter search using weakly labelled data or through heuristic visual inspection for purely unsupervised data.", "Categorical variables are a natural choice for representing discrete structure in the world. However, stochastic neural networks rarely use categorical latent variables due to the inability to backpropagate through samples. In this work, we present an efficient gradient estimator that replaces the non-differentiable sample from a categorical distribution with a differentiable sample from a novel Gumbel-Softmax distribution. This distribution has the essential property that it can be smoothly annealed into a categorical distribution. We show that our Gumbel-Softmax estimator outperforms state-of-the-art gradient estimators on structured output prediction and unsupervised generative modeling tasks with categorical latent variables, and enables large speedups on semi-supervised classification.", "This paper describes InfoGAN, an information-theoretic extension to the Generative Adversarial Network that is able to learn disentangled representations in a completely unsupervised manner. InfoGAN is a generative adversarial network that also maximizes the mutual information between a small subset of the latent variables and the observation. We derive a lower bound to the mutual information objective that can be optimized efficiently, and show that our training procedure can be interpreted as a variation of the Wake-Sleep algorithm. Specifically, InfoGAN successfully disentangles writing styles from digit shapes on the MNIST dataset, pose from lighting of 3D rendered images, and background digits from the central digit on the SVHN dataset. It also discovers visual concepts that include hair styles, presence absence of eyeglasses, and emotions on the CelebA face dataset. Experiments show that InfoGAN learns interpretable representations that are competitive with representations learned by existing fully supervised methods.", "The reparameterization trick enables optimizing large scale stochastic computation graphs via gradient descent. The essence of the trick is to refactor each stochastic node into a differentiable function of its parameters and a random variable with fixed distribution. After refactoring, the gradients of the loss propagated by the chain rule through the graph are low variance unbiased estimators of the gradients of the expected loss. While many continuous random variables have such reparameterizations, discrete random variables lack useful reparameterizations due to the discontinuous nature of discrete states. In this work we introduce Concrete random variables---continuous relaxations of discrete random variables. The Concrete distribution is a new family of distributions with closed form densities and a simple reparameterization. Whenever a discrete stochastic node of a computation graph can be refactored into a one-hot bit representation that is treated continuously, Concrete stochastic nodes can be used with automatic differentiation to produce low-variance biased gradients of objectives (including objectives that depend on the log-probability of latent stochastic nodes) on the corresponding discrete graph. We demonstrate the effectiveness of Concrete relaxations on density estimation and structured prediction tasks using neural networks.", "" ] }
1905.09624	2945318807	We present COBS, a compact bit-sliced signature index, which is a cross-over between an inverted index and Bloom filters. Our target application is to index @math -mers of DNA samples or @math -grams from text documents and process approximate pattern matching queries on the corpus with a user-chosen coverage threshold. Query results may contain a number of false positives which decreases exponentially with the query length and the false positive rate of the index determined at construction time. We compare COBS to seven other index software packages on 100 000 microbial DNA samples. COBS' compact but simple data structure outperforms the other indexes in construction time and query performance with Mantis by on second place. However, different from Mantis and other previous work, COBS does not need the complete index in RAM and is thus designed to scale to larger document sets.	Considering @math -grams or @math -mers of a sequence are a staple in bioinformatics @cite_28 . For example, BLAST @cite_10 uses @math -mers for seeding its search for maximal local alignments.	{ "cite_N": [ "@cite_28", "@cite_10" ], "mid": [ "2933804994", "2055043387" ], "abstract": [ "The analysis of biological sequencing data has been one of the biggest applications of string algorithms. The approaches used in many such applications are based on the analysis of k-mers, which are short fixed-length strings present in a dataset. While these approaches are rather diverse, storing and querying k-mer sets has emerged as a shared underlying component. Sets of k-mers have unique features and applications that, over the last ten years, have resulted in many specialized approaches for their representation. In this survey, we give a unified presentation and comparison of the data structures that have been proposed to store and query k-mer sets. We hope this survey will not only serve as a resource for researchers in the field but also make the area more accessible to outsiders", "A new approach to rapid sequence comparison, basic local alignment search tool (BLAST), directly approximates alignments that optimize a measure of local similarity, the maximal segment pair (MSP) score. Recent mathematical results on the stochastic properties of MSP scores allow an analysis of the performance of this method as well as the statistical significance of alignments it generates. The basic algorithm is simple and robust; it can be implemented in a number of ways and applied in a variety of contexts including straight-forward DNA and protein sequence database searches, motif searches, gene identification searches, and in the analysis of multiple regions of similarity in long DNA sequences. In addition to its flexibility and tractability to mathematical analysis, BLAST is an order of magnitude faster than existing sequence comparison tools of comparable sensitivity." ] }
1905.09624	2945318807	We present COBS, a compact bit-sliced signature index, which is a cross-over between an inverted index and Bloom filters. Our target application is to index @math -mers of DNA samples or @math -grams from text documents and process approximate pattern matching queries on the corpus with a user-chosen coverage threshold. Query results may contain a number of false positives which decreases exponentially with the query length and the false positive rate of the index determined at construction time. We compare COBS to seven other index software packages on 100 000 microbial DNA samples. COBS' compact but simple data structure outperforms the other indexes in construction time and query performance with Mantis by on second place. However, different from Mantis and other previous work, COBS does not need the complete index in RAM and is thus designed to scale to larger document sets.	The earliest use of Bloom filters as an index for a collection of independent documents we could find is called by Crainiceanu and Lemire @cite_34 . They propose to use a Bloom filter for each document and to arrange them either in a B-tree or as a . The latter is similar to BIGSI and COBS without compaction.	{ "cite_N": [ "@cite_34" ], "mid": [ "2005228765" ], "abstract": [ "Abstract Bloom filters are probabilistic data structures commonly used for approximate membership problems in many areas of Computer Science (networking, distributed systems, databases, etc.). With the increase in data size and distribution of data, problems arise where a large number of Bloom filters are available, and all of them need to be searched for potential matches. As an example, in a federated cloud environment, each cloud provider could encode the information using Bloom filters and share the Bloom filters with a central coordinator. The problem of interest is not only whether a given element is in any of the sets represented by the Bloom filters, but also which of the existing sets contain the given element. This problem cannot be solved by just constructing a Bloom filter on the union of all the sets. Instead, we effectively have a multidimensional Bloom filter problem: given an element, we wish to receive a list of candidate sets where the element might be. To solve this problem, we consider three alternatives. Firstly, we can naively check many Bloom filters. Secondly, we propose to organize the Bloom filters in a hierarchical index structure akin to a B+ tree that we call Bloofi. Finally, we propose another data structure that packs the Bloom filters in such a way as to exploit bit-level parallelism, which we call Flat-Bloofi. Our theoretical and experimental results show that Bloofi and Flat-Bloofi provide scalable and efficient solutions alternatives to search through a large number of Bloom filters." ] }
1905.09624	2945318807	We present COBS, a compact bit-sliced signature index, which is a cross-over between an inverted index and Bloom filters. Our target application is to index @math -mers of DNA samples or @math -grams from text documents and process approximate pattern matching queries on the corpus with a user-chosen coverage threshold. Query results may contain a number of false positives which decreases exponentially with the query length and the false positive rate of the index determined at construction time. We compare COBS to seven other index software packages on 100 000 microbial DNA samples. COBS' compact but simple data structure outperforms the other indexes in construction time and query performance with Mantis by on second place. However, different from Mantis and other previous work, COBS does not need the complete index in RAM and is thus designed to scale to larger document sets.	The currently most cited line of work on DNA @math -mer indices for approximate search are the Sequence Bloom Trees (SBTs) first proposed by Solomon and Kingsford @cite_16 . In an SBT the @math -mers of each document are individually indexed into a Bloom filter, which are then arranged as the leaves of a binary tree. The inner nodes of the binary tree are union Bloom filters of their descendants. A query can then breadth-first traverse the tree pruning search paths which no longer sufficiently cover a given threshold @math of the query @math -mers.	{ "cite_N": [ "@cite_16" ], "mid": [ "2266239166" ], "abstract": [ "A data indexing method enables fast and computationally efficient searching of short-read sequencing archives." ] }
1905.09624	2945318807	We present COBS, a compact bit-sliced signature index, which is a cross-over between an inverted index and Bloom filters. Our target application is to index @math -mers of DNA samples or @math -grams from text documents and process approximate pattern matching queries on the corpus with a user-chosen coverage threshold. Query results may contain a number of false positives which decreases exponentially with the query length and the false positive rate of the index determined at construction time. We compare COBS to seven other index software packages on 100 000 microbial DNA samples. COBS' compact but simple data structure outperforms the other indexes in construction time and query performance with Mantis by on second place. However, different from Mantis and other previous work, COBS does not need the complete index in RAM and is thus designed to scale to larger document sets.	In the original SBT @cite_16 a simple greedy clustering method is used, the bit union is stored in each inner node, and all nodes are RRR compressed @cite_3 using SDSL @cite_24 . The first improvement, the Split Sequence Bloom Tree (SSBT) @cite_12 , splits the inner nodes into two Bloom filters: a filter and a filter, where the first contains all bits in both child filters and the second those set in either child minus the filter. Simultaneously, Sun, Harris, Chikhi, and Medvedev proposed the AllSome Sequence Bloom Tree (AllSome-SBT) @cite_7 , which splits each inner node into an and a subfilter. The filter contains bits in all leaves below the node, excluding those already set in the parent node, and the filter all bits in some leaves but not all. The currently smallest SBT variant is called HowDe-SBT by Harris and Medvedev @cite_20 . It decomposes the Bloom filters in each inner node into two bit vectors: the vector signals if a particular bit is at this inner node, meaning that it is equal in all descendant leaves, and a vector signaling if it is determined as zero or one. All determined bits can be omitted from any children.	{ "cite_N": [ "@cite_7", "@cite_3", "@cite_24", "@cite_16", "@cite_20", "@cite_12" ], "mid": [ "2795650864", "1985174631", "1962019683", "2266239166", "", "2884435343" ], "abstract": [ "Abstract The ubiquity of next-generation sequencing has transformed the size and nature of many databases, pushing the boundaries of current indexing and searching methods. One particular example is a database of 2652 human RNA-seq experiments uploaded to the Sequence Read Archive (SRA). Recently, Solomon and Kingsford proposed the Sequence Bloom Tree data structure and demonstrated how it can be used to accurately identify SRA samples that have a transcript of interest potentially expressed. In this article, we propose an improvement called the AllSome Sequence Bloom Tree. Results show that our new data structure significantly improves performance, reducing the tree construction time by 52.7 and query time by 39 –85 , with a price of upto 3 × memory consumption during queries. Notably, it can query a batch of 198,074 queries in <8 hours (compared with around 2 days previously) and a whole set of k-mers from a sequencing experiment (about 27 million k-mers) in <11 minutes.", "We consider the indexable dictionary problem which consists in storing a set S ⊆ 0,…, m - 1 for some integer m, while supporting the operations of rank(x), which returns the number of elements in S that are less than x if x e S, and -1 otherwise; and select(i) which returns the i-th smallest element in S.We give a structure that supports both operations in O(1) time on the RAM model and requires B(n,m) + o(n) + O(lg lg m) bits to store a set of size n, where B(n,m) = ⌈lg ( n m)⌉ is the minimum number of bits required to store any n-element subset from a universe of size m. Previous dictionaries taking this space only supported (yes no) membership queries in O(1) time. In the cell probe model we can remove the O(lg lg m) additive term in the space bound, answering a question raised by Fich and Miltersen, and Pagh.We also present two applications of our dictionary structure:• An information-theoretically optimal representation for k-ary cardinal trees (aka k-ary tries). Our structure uses C(n,k) + o(n + lg k) bits to store a k-ary tree with n nodes and can support parent, i-th child, child labeled i, and the degree of a node in constant time, where C(n,k) is the minimum number of bits to store any n-node k-ary tree. Previous space efficient representations for cardinal k-ary trees required C(n,k) + Ω(n) bits.• An optimal representation for multisets where (appropriate generalisations of) the select and rank operations can be supported in O(1) time. Our structure uses B(n, m + n) + o(n) + O(lg lg m) bits to represent a multiset of size n from an m element set; the first term is the minimum number of bits required to represent such a multiset.", "Engineering efficient implementations of compact and succinct structures is time-consuming and challenging, since there is no standard library of easy-to-use, highly optimized, and composable components. One consequence is that measuring the practical impact of new theoretical proposals is difficult, since older baseline implementations may not rely on the same basic components, and reimplementing from scratch can be time-consuming. In this paper we present a framework for experimentation with succinct data structures, providing a large set of configurable components, together with tests, benchmarks, and tools to analyze resource requirements. We demonstrate the functionality of the framework by recomposing two succinct solutions for top-k document retrieval which can operate on both character and integer alphabets.", "A data indexing method enables fast and computationally efficient searching of short-read sequencing archives.", "", "Abstract Enormous databases of short-read RNA-seq experiments such as the NIH Sequencing Read Archive are now available. These databases could answer many questions about condition-specific express..." ] }
1905.09624	2945318807	We present COBS, a compact bit-sliced signature index, which is a cross-over between an inverted index and Bloom filters. Our target application is to index @math -mers of DNA samples or @math -grams from text documents and process approximate pattern matching queries on the corpus with a user-chosen coverage threshold. Query results may contain a number of false positives which decreases exponentially with the query length and the false positive rate of the index determined at construction time. We compare COBS to seven other index software packages on 100 000 microbial DNA samples. COBS' compact but simple data structure outperforms the other indexes in construction time and query performance with Mantis by on second place. However, different from Mantis and other previous work, COBS does not need the complete index in RAM and is thus designed to scale to larger document sets.	@cite_30 is another @math -mer index software package. It contains an ensemble of encoding techniques for compressing the occurrence maps of @math -mers in the document set. To locate the correct occurrence map a hierarchy of s is built, which are minimum perfect hash function mappings and can introduce false positive results due to mapping of alien @math -mers to random results.	{ "cite_N": [ "@cite_30" ], "mid": [ "2788348538" ], "abstract": [ "We present SeqOthello, an ultra-fast and memory-efficient indexing structure to support arbitrary sequence query against large collections of RNA-seq experiments. It takes SeqOthello only 5 min and 19.1 GB memory to conduct a global survey of 11,658 fusion events against 10,113 TCGA Pan-Cancer RNA-seq datasets. The query recovers 92.7 of tier-1 fusions curated by TCGA Fusion Gene Database and reveals 270 novel occurrences, all of which are present as tumor-specific. By providing a reference-free, alignment-free, and parameter-free sequence search system, SeqOthello will enable large-scale integrative studies using sequence-level data, an undertaking not previously practicable for many individual labs." ] }
1905.09624	2945318807	We present COBS, a compact bit-sliced signature index, which is a cross-over between an inverted index and Bloom filters. Our target application is to index @math -mers of DNA samples or @math -grams from text documents and process approximate pattern matching queries on the corpus with a user-chosen coverage threshold. Query results may contain a number of false positives which decreases exponentially with the query length and the false positive rate of the index determined at construction time. We compare COBS to seven other index software packages on 100 000 microbial DNA samples. COBS' compact but simple data structure outperforms the other indexes in construction time and query performance with Mantis by on second place. However, different from Mantis and other previous work, COBS does not need the complete index in RAM and is thus designed to scale to larger document sets.	(BItsliced Genomic Signature Index) by @cite_13 is the direct ancestor of COBS and also a combination of Bloom filters and inverted indexes. BIGSI however is a prototype programmed in Python and uses a key-value database such as BerkeleyDB or RocksDB as storage back-end. It also does not contain the compaction feature introduced in COBS.	{ "cite_N": [ "@cite_13" ], "mid": [ "2913847081" ], "abstract": [ "Exponentially increasing amounts of unprocessed bacterial and viral genomic sequence data are stored in the global archives. The ability to query these data for sequence search terms would facilitate both basic research and applications such as real-time genomic epidemiology and surveillance. However, this is not possible with current methods. To solve this problem, we combine knowledge of microbial population genomics with computational methods devised for web search to produce a searchable data structure named BItsliced Genomic Signature Index (BIGSI). We indexed the entire global corpus of 447,833 bacterial and viral whole-genome sequence datasets using four orders of magnitude less storage than previous methods. We applied our BIGSI search function to rapidly find resistance genes MCR-1, MCR-2, and MCR-3, determine the host-range of 2,827 plasmids, and quantify antibiotic resistance in archived datasets. Our index can grow incrementally as new (unprocessed or assembled) sequence datasets are deposited and can scale to millions of datasets." ] }
1905.09751	2946146170	Many applied decision-making problems have a dynamic component: The policymaker needs not only to choose whom to treat, but also when to start which treatment. For example, a medical doctor may see a patient many times and, at each visit, need to choose between prescribing either an invasive or a non-invasive procedure and postponing the decision to the next visit. In this paper, we develop an advantage doubly robust estimator for learning such dynamic treatment rules using observational data under sequential ignorability. We prove welfare regret bounds that generalize results for doubly robust learning in the single-step setting, and show promising empirical performance in several different contexts. Our approach is practical for policy optimization, and does not need any structural (e.g., Markovian) assumptions.	In the special case of offline policy learning in a single timestep where a policy only needs to decide whether to treat a subject but not , substantial progress has been made in how to derive an optimal regret and performing optimization in finding the optimal policy * athey2017efficient, kallus2018confounding, kitagawa2015should, swaminathan2015batch,zhang2012estimating,zhao2012estimating, zhou2018offline, zhou2015residual . In particular, @cite_2 establishes a lower bound @math on learning the regret in the case of binary treatment. @cite_1 and @cite_6 show a matching upperbound assuming the nuisance components can be learned at a much slower rate in the setting where the treatment consists of binary actions and multiple actions respectively. Extending this line of result to sequential multi-step settings is nontrivial for several reasons: First, it is unclear how to optimize efficiently across all policies in a policy class, especially given the increasing complexity with long horizons. Second, the regret results used in @cite_1 and @cite_6 rely on a chaining argument in which the special form of the estimator ensures values of policies close to each other is close. It is not obvious whether existing doubly-robust estimators in the sequential settings *[e.g.,][] thomas2016data,zhang2013robust have such form.	{ "cite_N": [ "@cite_1", "@cite_6", "@cite_2" ], "mid": [ "2591915067", "2951503583", "2103959887" ], "abstract": [ "In many areas, practitioners seek to use observational data to learn a treatment assignment policy that satisfies application-specific constraints, such as budget, fairness, simplicity, or other functional form constraints. For example, policies may be restricted to take the form of decision trees based on a limited set of easily observable individual characteristics. We propose a new approach to this problem motivated by the theory of semiparametrically efficient estimation. Our method can be used to optimize either binary treatments or infinitesimal nudges to continuous treatments, and can leverage observational data where causal effects are identified using a variety of strategies, including selection on observables and instrumental variables. Given a doubly robust estimator of the causal effect of assigning everyone to treatment, we develop an algorithm for choosing whom to treat, and establish strong guarantees for the asymptotic utilitarian regret of the resulting policy.", "In many settings, a decision-maker wishes to learn a rule, or policy, that maps from observable characteristics of an individual to an action. Examples include selecting offers, prices, advertisements, or emails to send to consumers, as well as the problem of determining which medication to prescribe to a patient. While there is a growing body of literature devoted to this problem, most existing results are focused on the case where data comes from a randomized experiment, and further, there are only two possible actions, such as giving a drug to a patient or not. In this paper, we study the offline multi-action policy learning problem with observational data and where the policy may need to respect budget constraints or belong to a restricted policy class such as decision trees. We build on the theory of efficient semi-parametric inference in order to propose and implement a policy learning algorithm that achieves asymptotically minimax-optimal regret. To the best of our knowledge, this is the first result of this type in the multi-action setup, and it provides a substantial performance improvement over the existing learning algorithms. We then consider additional computational challenges that arise in implementing our method for the case where the policy is restricted to take the form of a decision tree. We propose two different approaches, one using a mixed integer program formulation and the other using a tree-search based algorithm.", "One of the main objectives of empirical analysis of experiments and quasi-experiments is to inform policy decisions that determine the allocation of treatments to individuals with dierent observable covariates. We propose the Empirical Welfare Maximization (EWM) method, which estimates a treatment assignment policy by maximizing the sample analog of average social welfare over a class of candidate treatment policies. The EWM approach is attractive in terms of both statistical performance and practical implementation in realistic settings of policy design. Common features of these settings include: (i) feasible treatment assignment rules are constrained exogenously for ethical, legislative, or political reasons, (ii) a policy maker wants a simple treatment assignment rule based on one or more eligibility scores in order to reduce the dimensionality of individual observable characteristics, and or (iii) the proportion of individuals who can receive the treatment is a priori limited due to a budget or a capacity constraint. We show that when the propensity score is known, the average social welfare attained by EWM rules converges at least at n 1=2 rate to the maximum obtainable welfare uniformly over a minimally constrained class of data distributions, and this uniform convergence rate is minimax optimal. In comparison with this benchmark rate, we examine how the uniform convergence rate of the average welfare improves or deteriorates depending on the richness of the class of candidate decision rules, the distribution of conditional treatment eects, and the lack of knowledge of the propensity score. We provide an asymptotically valid inference procedure for the population welfare gain obtained by exercising the EWM rule. We oer easily implementable algorithms" ] }
1905.09751	2946146170	Many applied decision-making problems have a dynamic component: The policymaker needs not only to choose whom to treat, but also when to start which treatment. For example, a medical doctor may see a patient many times and, at each visit, need to choose between prescribing either an invasive or a non-invasive procedure and postponing the decision to the next visit. In this paper, we develop an advantage doubly robust estimator for learning such dynamic treatment rules using observational data under sequential ignorability. We prove welfare regret bounds that generalize results for doubly robust learning in the single-step setting, and show promising empirical performance in several different contexts. Our approach is practical for policy optimization, and does not need any structural (e.g., Markovian) assumptions.	In this paper, we focus on the problem of making treatment decisions once and for all and with multiple actions at the decision point, which we note is a strict generalization of the setting in @cite_6 . We propose an advantage doubly robust estimator that draws upon the semiparametrics and orthogonal moments literature . There is also a growing number of existing works that have applied orthogonal moments construction to policy evaluation .	{ "cite_N": [ "@cite_6" ], "mid": [ "2951503583" ], "abstract": [ "In many settings, a decision-maker wishes to learn a rule, or policy, that maps from observable characteristics of an individual to an action. Examples include selecting offers, prices, advertisements, or emails to send to consumers, as well as the problem of determining which medication to prescribe to a patient. While there is a growing body of literature devoted to this problem, most existing results are focused on the case where data comes from a randomized experiment, and further, there are only two possible actions, such as giving a drug to a patient or not. In this paper, we study the offline multi-action policy learning problem with observational data and where the policy may need to respect budget constraints or belong to a restricted policy class such as decision trees. We build on the theory of efficient semi-parametric inference in order to propose and implement a policy learning algorithm that achieves asymptotically minimax-optimal regret. To the best of our knowledge, this is the first result of this type in the multi-action setup, and it provides a substantial performance improvement over the existing learning algorithms. We then consider additional computational challenges that arise in implementing our method for the case where the policy is restricted to take the form of a decision tree. We propose two different approaches, one using a mixed integer program formulation and the other using a tree-search based algorithm." ] }
1905.09751	2946146170	Many applied decision-making problems have a dynamic component: The policymaker needs not only to choose whom to treat, but also when to start which treatment. For example, a medical doctor may see a patient many times and, at each visit, need to choose between prescribing either an invasive or a non-invasive procedure and postponing the decision to the next visit. In this paper, we develop an advantage doubly robust estimator for learning such dynamic treatment rules using observational data under sequential ignorability. We prove welfare regret bounds that generalize results for doubly robust learning in the single-step setting, and show promising empirical performance in several different contexts. Our approach is practical for policy optimization, and does not need any structural (e.g., Markovian) assumptions.	Finally, our proposed estimator heavily relies on an advantage decomposition in @cite_7 . @cite_7 focuses on the generalization error on a variant of Q-learning, and we turn such a decomposition into a practical and efficient estimator for policy value search.	{ "cite_N": [ "@cite_7" ], "mid": [ "2168551972" ], "abstract": [ "Planning problems that involve learning a policy from a single training set of finite horizon trajectories arise in both social science and medical fields. We consider Q-learning with function approximation for this setting and derive an upper bound on the generalization error. This upper bound is in terms of quantities minimized by a Q-learning algorithm, the complexity of the approximation space and an approximation term due to the mismatch between Q-learning and the goal of learning a policy that maximizes the value function." ] }

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.