arXiv subjects

Dacheng Tao

Publications and source records attributed to Dacheng Tao.

At least 937 records · Page 52Linked to original sources

arXiv2018

An Information-Theoretic View for Deep Learning

Deep learning has transformed computer vision, natural language processing, and speech recognition\cite{badrinarayanan2017segnet, dong2016image, ren2017faster, ji20133d}. However, two critical questions remain obscure: (1) why do deep neural networks generalize better than shallow networks; and (2) does it always hold that a deeper network leads to better performance? Specifically, letting $L$ be the number of convolutional and pooling layers in a deep neural network, and $n$ be the size of the training sample, we derive an upper bound on the expected generalization error for this network, i.e., \begin{eqnarray*} \mathbb{E}[R(W)-R_S(W)] \leq \exp{\left(-\frac{L}{2}\log{\frac{1}η}\right)}\sqrt{\frac{2σ^2}{n}I(S,W) } \end{eqnarray*} where $σ>0$ is a constant depending on the loss function, $0<η<1$ is a constant depending on the information loss for each convolutional or pooling layer, and $I(S, W)$ is the mutual information between the training sample $S$ and the output hypothesis $W$. This upper bound shows that as the number of convolutional and pooling layers $L$ increases in the network, the expected generalization error will decrease exponentially to zero. Layers with strict information loss, such as the convolutional layers, reduce the generalization error for the whole network; this answers the first question. However, algorithms with zero expected generalization error does not imply a small test error or $\mathbb{E}[R(W)]$. This is because $\mathbb{E}[R_S(W)]$ is large when the information for fitting the data is lost as the number of layers increases. This suggests that the claim `the deeper the better' is conditioned on a small training error or $\mathbb{E}[R_S(W)]$. Finally, we show that deep learning satisfies a weak notion of stability and the sample complexity of deep neural networks will decrease as $L$ increases.

stat.ML↗

arXiv2018

Stochastic Second-order Methods for Non-convex Optimization with Inexact Hessian and Gradient

Trust region and cubic regularization methods have demonstrated good performance in small scale non-convex optimization, showing the ability to escape from saddle points. Each iteration of these methods involves computation of gradient, Hessian and function value in order to obtain the search direction and adjust the radius or cubic regularization parameter. However, exactly computing those quantities are too expensive in large-scale problems such as training deep networks. In this paper, we study a family of stochastic trust region and cubic regularization methods when gradient, Hessian and function values are computed inexactly, and show the iteration complexity to achieve $ε$-approximate second-order optimality is in the same order with previous work for which gradient and function values are computed exactly. The mild conditions on inexactness can be achieved in finite-sum minimization using random sampling. We show the algorithm performs well on training convolutional neural networks compared with previous second-order methods.

math.OC↗

arXiv2018

Towards Query Efficient Black-box Attacks: An Input-free Perspective

Recent studies have highlighted that deep neural networks (DNNs) are vulnerable to adversarial attacks, even in a black-box scenario. However, most of the existing black-box attack algorithms need to make a huge amount of queries to perform attacks, which is not practical in the real world. We note one of the main reasons for the massive queries is that the adversarial example is required to be visually similar to the original image, but in many cases, how adversarial examples look like does not matter much. It inspires us to introduce a new attack called \emph{input-free} attack, under which an adversary can choose an arbitrary image to start with and is allowed to add perceptible perturbations on it. Following this approach, we propose two techniques to significantly reduce the query complexity. First, we initialize an adversarial example with a gray color image on which every pixel has roughly the same importance for the target model. Then we shrink the dimension of the attack space by perturbing a small region and tiling it to cover the input image. To make our algorithm more effective, we stabilize a projected gradient ascent algorithm with momentum, and also propose a heuristic approach for region size selection. Through extensive experiments, we show that with only 1,701 queries on average, we can perturb a gray image to any target class of ImageNet with a 100\% success rate on InceptionV3. Besides, our algorithm has successfully defeated two real-world systems, the Clarifai food detection API and the Baidu Animal Identification API.

stat.ML↗

arXiv2018

Stochastically Controlled Stochastic Gradient for the Convex and Non-convex Composition problem

In this paper, we consider the convex and non-convex composition problem with the structure $\frac{1}{n}\sum\nolimits_{i = 1}^n {{F_i}( {G( x )} )}$, where $G( x )=\frac{1}{n}\sum\nolimits_{j = 1}^n {{G_j}( x )} $ is the inner function, and $F_i(\cdot)$ is the outer function. We explore the variance reduction based method to solve the composition optimization. Due to the fact that when the number of inner function and outer function are large, it is not reasonable to estimate them directly, thus we apply the stochastically controlled stochastic gradient (SCSG) method to estimate the gradient of the composition function and the value of the inner function. The query complexity of our proposed method for the convex and non-convex problem is equal to or better than the current method for the composition problem. Furthermore, we also present the mini-batch version of the proposed method, which has the improved the query complexity with related to the size of the mini-batch.

math.OC↗

arXiv2018

Transfer Learning with Label Noise

Transfer learning aims to improve learning in target domain by borrowing knowledge from a related but different source domain. To reduce the distribution shift between source and target domains, recent methods have focused on exploring invariant representations that have similar distributions across domains. However, when learning this invariant knowledge, existing methods assume that the labels in source domain are uncontaminated, while in reality, we often have access to source data with noisy labels. In this paper, we first show how label noise adversely affect the learning of invariant representations and the correcting of label shift in various transfer learning scenarios. To reduce the adverse effects, we propose a novel Denoising Conditional Invariant Component (DCIC) framework, which provably ensures (1) extracting invariant representations given examples with noisy labels in source domain and unlabeled examples in target domain; (2) estimating the label distribution in target domain with no bias. Experimental results on both synthetic and real-world data verify the effectiveness of the proposed method.

stat.ML↗

arXiv2018

Learning with Biased Complementary Labels

In this paper, we study the classification problem in which we have access to easily obtainable surrogate for true labels, namely complementary labels, which specify classes that observations do \textbf{not} belong to. Let $Y$ and $\bar{Y}$ be the true and complementary labels, respectively. We first model the annotation of complementary labels via transition probabilities $P(\bar{Y}=i|Y=j), i\neq j\in\{1,\cdots,c\}$, where $c$ is the number of classes. Previous methods implicitly assume that $P(\bar{Y}=i|Y=j), \forall i\neq j$, are identical, which is not true in practice because humans are biased toward their own experience. For example, as shown in Figure 1, if an annotator is more familiar with monkeys than prairie dogs when providing complementary labels for meerkats, she is more likely to employ "monkey" as a complementary label. We therefore reason that the transition probabilities will be different. In this paper, we propose a framework that contributes three main innovations to learning with \textbf{biased} complementary labels: (1) It estimates transition probabilities with no bias. (2) It provides a general method to modify traditional loss functions and extends standard deep neural network classifiers to learn with biased complementary labels. (3) It theoretically ensures that the classifier learned with complementary labels converges to the optimal one learned with true labels. Comprehensive experiments on several benchmark datasets validate the superiority of our method to current state-of-the-art methods.

stat.ML↗

arXiv2018

Stochastic Zeroth-order Optimization via Variance Reduction method

Derivative-free optimization has become an important technique used in machine learning for optimizing black-box models. To conduct updates without explicitly computing gradient, most current approaches iteratively sample a random search direction from Gaussian distribution and compute the estimated gradient along that direction. However, due to the variance in the search direction, the convergence rates and query complexities of existing methods suffer from a factor of $d$, where $d$ is the problem dimension. In this paper, we introduce a novel Stochastic Zeroth-order method with Variance Reduction under Gaussian smoothing (SZVR-G) and establish the complexity for optimizing non-convex problems. With variance reduction on both sample space and search space, the complexity of our algorithm is sublinear to $d$ and is strictly better than current approaches, in both smooth and non-smooth cases. Moreover, we extend the proposed method to the mini-batch version. Our experimental results demonstrate the superior performance of the proposed method over existing derivative-free optimization techniques. Furthermore, we successfully apply our method to conduct a universal black-box attack to deep neural networks and present some interesting results.

stat.ML↗

arXiv2018

Robust Student Network Learning

Deep neural networks bring in impressive accuracy in various applications, but the success often relies on the heavy network architecture. Taking well-trained heavy networks as teachers, classical teacher-student learning paradigm aims to learn a student network that is lightweight yet accurate. In this way, a portable student network with significantly fewer parameters can achieve a considerable accuracy which is comparable to that of teacher network. However, beyond accuracy, robustness of the learned student network against perturbation is also essential for practical uses. Existing teacher-student learning frameworks mainly focus on accuracy and compression ratios, but ignore the robustness. In this paper, we make the student network produce more confident predictions with the help of the teacher network, and analyze the lower bound of the perturbation that will destroy the confidence of the student network. Two important objectives regarding prediction scores and gradients of examples are developed to maximize this lower bound, so as to enhance the robustness of the student network without sacrificing the performance. Experiments on benchmark datasets demonstrate the efficiency of the proposed approach to learn robust student networks which have satisfying accuracy and compact sizes.

cs.LG↗

arXiv2018

Improved Techniques for Learning to Dehaze and Beyond: A Collective Study

Here we explore two related but important tasks based on the recently released REalistic Single Image DEhazing (RESIDE) benchmark dataset: (i) single image dehazing as a low-level image restoration problem; and (ii) high-level visual understanding (e.g., object detection) of hazy images. For the first task, we investigated a variety of loss functions and show that perception-driven loss significantly improves dehazing performance. In the second task, we provide multiple solutions including using advanced modules in the dehazing-detection cascade and domain-adaptive object detectors. In both tasks, our proposed solutions significantly improve performance. GitHub repository URL is: https://github.com/guanlongzhao/dehaze

cs.CV↗

arXiv2018

Discriminative multi-view Privileged Information learning for image re-ranking

Conventional multi-view re-ranking methods usually perform asymmetrical matching between the region of interest (ROI) in the query image and the whole target image for similarity computation. Due to the inconsistency in the visual appearance, this practice tends to degrade the retrieval accuracy particularly when the image ROI, which is usually interpreted as the image objectness, accounts for a smaller region in the image. Since Privileged Information (PI), which can be viewed as the image prior, enables well characterizing the image objectness, we are aiming at leveraging PI for further improving the performance of the multi-view re-ranking accuracy in this paper. Towards this end, we propose a discriminative multi-view re-ranking approach in which both the original global image visual contents and the local auxiliary PI features are simultaneously integrated into a unified training framework for generating the latent subspaces with sufficient discriminating power. For the on-the-fly re-ranking, since the multi-view PI features are unavailable, we only project the original multi-view image representations onto the latent subspace, and thus the re-ranking can be achieved by computing and sorting the distances from the multi-view embeddings to the separating hyperplane. Extensive experimental evaluations on the two public benchmarks Oxford5k and Paris6k reveal our approach provides further performance boost for accurate image re-ranking, whilst the comparative study demonstrates the advantage of our method against other multi-view re-ranking methods.

cs.CV↗

arXiv2018

Domain Generalization via Conditional Invariant Representation

Domain generalization aims to apply knowledge gained from multiple labeled source domains to unseen target domains. The main difficulty comes from the dataset bias: training data and test data have different distributions, and the training set contains heterogeneous samples from different distributions. Let $X$ denote the features, and $Y$ be the class labels. Existing domain generalization methods address the dataset bias problem by learning a domain-invariant representation $h(X)$ that has the same marginal distribution $\mathbb{P}(h(X))$ across multiple source domains. The functional relationship encoded in $\mathbb{P}(Y|X)$ is usually assumed to be stable across domains such that $\mathbb{P}(Y|h(X))$ is also invariant. However, it is unclear whether this assumption holds in practical problems. In this paper, we consider the general situation where both $\mathbb{P}(X)$ and $\mathbb{P}(Y|X)$ can change across all domains. We propose to learn a feature representation which has domain-invariant class conditional distributions $\mathbb{P}(h(X)|Y)$. With the conditional invariant representation, the invariance of the joint distribution $\mathbb{P}(h(X),Y)$ can be guaranteed if the class prior $\mathbb{P}(Y)$ does not change across training and test domains. Extensive experiments on both synthetic and real data demonstrate the effectiveness of the proposed method.

cs.LG↗

arXiv2018

Selective Zero-Shot Classification with Augmented Attributes

In this paper, we introduce a selective zero-shot classification problem: how can the classifier avoid making dubious predictions? Existing attribute-based zero-shot classification methods are shown to work poorly in the selective classification scenario. We argue the under-complete human defined attribute vocabulary accounts for the poor performance. We propose a selective zero-shot classifier based on both the human defined and the automatically discovered residual attributes. The proposed classifier is constructed by firstly learning the defined and the residual attributes jointly. Then the predictions are conducted within the subspace of the defined attributes. Finally, the prediction confidence is measured by both the defined and the residual attributes. Experiments conducted on several benchmarks demonstrate that our classifier produces a superior performance to other methods under the risk-coverage trade-off metric.

cs.CV↗

arXiv2018

Transforming Cooling Optimization for Green Data Center via Deep Reinforcement Learning

Cooling system plays a critical role in a modern data center (DC). Developing an optimal control policy for DC cooling system is a challenging task. The prevailing approaches often rely on approximating system models that are built upon the knowledge of mechanical cooling, electrical and thermal management, which is difficult to design and may lead to sub-optimal or unstable performances. In this paper, we propose utilizing the large amount of monitoring data in DC to optimize the control policy. To do so, we cast the cooling control policy design into an energy cost minimization problem with temperature constraints, and tap it into the emerging deep reinforcement learning (DRL) framework. Specifically, we propose an end-to-end cooling control algorithm (CCA) that is based on the actor-critic framework and an off-policy offline version of the deep deterministic policy gradient (DDPG) algorithm. In the proposed CCA, an evaluation network is trained to predict an energy cost counter penalized by the cooling status of the DC room, and a policy network is trained to predict optimized control settings when gave the current load and weather information. The proposed algorithm is evaluated on the EnergyPlus simulation platform and on a real data trace collected from the National Super Computing Centre (NSCC) of Singapore. Our results show that the proposed CCA can achieve about 11% cooling cost saving on the simulation platform compared with a manually configured baseline control algorithm. In the trace-based study, we propose a de-underestimation validation mechanism as we cannot directly test the algorithm on a real DC. Even though with DUE the results are conservative, we can still achieve about 15% cooling energy saving on the NSCC data trace if we set the inlet temperature threshold at 26.6 degree Celsius.

cs.AI↗

arXiv2018

Causal Generative Domain Adaptation Networks

An essential problem in domain adaptation is to understand and make use of distribution changes across domains. For this purpose, we first propose a flexible Generative Domain Adaptation Network (G-DAN) with specific latent variables to capture changes in the generating process of features across domains. By explicitly modeling the changes, one can even generate data in new domains using the generating process with new values for the latent variables in G-DAN. In practice, the process to generate all features together may involve high-dimensional latent variables, requiring dealing with distributions in high dimensions and making it difficult to learn domain changes from few source domains. Interestingly, by further making use of the causal representation of joint distributions, we then decompose the joint distribution into separate modules, each of which involves different low-dimensional latent variables and can be learned separately, leading to a Causal G-DAN (CG-DAN). This improves both statistical and computational efficiency of the learning procedure. Finally, by matching the feature distribution in the target domain, we can recover the target-domain joint distribution and derive the learning machine for the target domain. We demonstrate the efficacy of both G-DAN and CG-DAN in domain generation and cross-domain prediction on both synthetic and real data experiments.

stat.ML↗

arXiv2018

MoE-SPNet: A Mixture-of-Experts Scene Parsing Network

Scene parsing is an indispensable component in understanding the semantics within a scene. Traditional methods rely on handcrafted local features and probabilistic graphical models to incorporate local and global cues. Recently, methods based on fully convolutional neural networks have achieved new records on scene parsing. An important strategy common to these methods is the aggregation of hierarchical features yielded by a deep convolutional neural network. However, typical algorithms usually aggregate hierarchical convolutional features via concatenation or linear combination, which cannot sufficiently exploit the diversities of contextual information in multi-scale features and the spatial inhomogeneity of a scene. In this paper, we propose a mixture-of-experts scene parsing network (MoE-SPNet) that incorporates a convolutional mixture-of-experts layer to assess the importance of features from different levels and at different spatial locations. In addition, we propose a variant of mixture-of-experts called the adaptive hierarchical feature aggregation (AHFA) mechanism which can be incorporated into existing scene parsing networks that use skip-connections to fuse features layer-wisely. In the proposed networks, different levels of features at each spatial location are adaptively re-weighted according to the local structure and surrounding contextual information before aggregation. We demonstrate the effectiveness of the proposed methods on two scene parsing datasets including PASCAL VOC 2012 and SceneParse150 based on two kinds of baseline models FCN-8s and DeepLab-ASPP.

cs.CV↗

arXiv2018

Deep Blur Mapping: Exploiting High-Level Semantics by Deep Neural Networks

The human visual system excels at detecting local blur of visual images, but the underlying mechanism is not well understood. Traditional views of blur such as reduction in energy at high frequencies and loss of phase coherence at localized features have fundamental limitations. For example, they cannot well discriminate flat regions from blurred ones. Here we propose that high-level semantic information is critical in successfully identifying local blur. Therefore, we resort to deep neural networks that are proficient at learning high-level features and propose the first end-to-end local blur mapping algorithm based on a fully convolutional network. By analyzing various architectures with different depths and design philosophies, we empirically show that high-level features of deeper layers play a more important role than low-level features of shallower layers in resolving challenging ambiguities for this task. We test the proposed method on a standard blur detection benchmark and demonstrate that it significantly advances the state-of-the-art (ODS F-score of 0.853). Furthermore, we explore the use of the generated blur maps in three applications, including blur region segmentation, blur degree estimation, and blur magnification.

cs.CV↗

arXiv2018

A Compromise Principle in Deep Monocular Depth Estimation

Monocular depth estimation, which plays a key role in understanding 3D scene geometry, is fundamentally an ill-posed problem. Existing methods based on deep convolutional neural networks (DCNNs) have examined this problem by learning convolutional networks to estimate continuous depth maps from monocular images. However, we find that training a network to predict a high spatial resolution continuous depth map often suffers from poor local solutions. In this paper, we hypothesize that achieving a compromise between spatial and depth resolutions can improve network training. Based on this "compromise principle", we propose a regression-classification cascaded network (RCCN), which consists of a regression branch predicting a low spatial resolution continuous depth map and a classification branch predicting a high spatial resolution discrete depth map. The two branches form a cascaded structure allowing the classification and regression branches to benefit from each other. By leveraging large-scale raw training datasets and some data augmentation strategies, our network achieves top or state-of-the-art results on the NYU Depth V2, KITTI, and Make3D benchmarks.

cs.CV↗

arXiv2018

Deep Ordinal Regression Network for Monocular Depth Estimation

Monocular depth estimation, which plays a crucial role in understanding 3D scene geometry, is an ill-posed problem. Recent methods have gained significant improvement by exploring image-level information and hierarchical features from deep convolutional neural networks (DCNNs). These methods model depth estimation as a regression problem and train the regression networks by minimizing mean squared error, which suffers from slow convergence and unsatisfactory local solutions. Besides, existing depth estimation networks employ repeated spatial pooling operations, resulting in undesirable low-resolution feature maps. To obtain high-resolution depth maps, skip-connections or multi-layer deconvolution networks are required, which complicates network training and consumes much more computations. To eliminate or at least largely reduce these problems, we introduce a spacing-increasing discretization (SID) strategy to discretize depth and recast depth network learning as an ordinal regression problem. By training the network using an ordinary regression loss, our method achieves much higher accuracy and \dd{faster convergence in synch}. Furthermore, we adopt a multi-scale network structure which avoids unnecessary spatial pooling and captures multi-scale information in parallel. The method described in this paper achieves state-of-the-art results on four challenging benchmarks, i.e., KITTI [17], ScanNet [9], Make3D [50], and NYU Depth v2 [42], and win the 1st prize in Robust Vision Challenge 2018. Code has been made available at: https://github.com/hufu6371/DORN.

cs.CV↗