Search

Unsupervised domain adaptation (UDA) aims at reducing the distribution discrepancy when transferring knowledge from a labeled source domain to an unlabeled target domain. Previous UDA methods assume that the source and target domains share an identical label space, which is unrealistic in practice since the label information of the target domain is agnostic. This article focuses on a more realistic UDA scenario, i.e., partial domain adaptation (PDA), where the target label space is subsumed to the source label space. In the PDA scenario, the source outliers that are absent in the target domain may be wrongly matched to the target domain (technically named negative transfer), leading to performance degradation of UDA methods. This article proposes a novel target-domain-specific classifier learning-based domain adaptation (TSCDA) method. TSCDA presents a soft-weighed maximum mean discrepancy criterion to partially align feature distributions and alleviate negative transfer. Also, it learns a target-specific classifier for the target domain with pseudolabels and multiple auxiliary classifiers to further address the classifier shift. A module named peers-assisted learning is used to minimize the prediction difference between multiple target-specific classifiers, which makes the classifiers more discriminant for the target domain. Extensive experiments conducted on three PDA benchmark data sets show that TSCDA outperforms other state-of-the-art methods with a large margin, e.g., 4% and 5.6% averagely on Office-31 and Office-Home, respectively.

Unsupervised domain adaptation (UDA) is a representative problem in transfer learning, which aims to improve the classification performance on an unlabeled target domain by exploiting discriminant information from a labeled source domain. The optimal transport model has been used for UDA in the perspective of distribution matching. However, the transport distance cannot reflect the discriminant information from either domain knowledge or category prior. In this work, we propose an enhanced transport distance (ETD) for UDA. This method builds an attention-aware transport distance, which can be viewed as the prediction feedback of the iteratively learned classifier, to measure the domain discrepancy. Further, the Kantorovich potential variable is re-parameterized by deep neural networks to learn the distribution in the latent space. The entropy-based regularization is developed to explore the intrinsic structure of the target domain. The proposed method is optimized alternately in an end-to-end manner. Extensive experiments are conducted on four benchmark datasets to demonstrate the SOTA performance of ETD.

Unsupervised domain adaptation is effective in leveraging the rich information from the source domain to the unsupervised target domain. Though deep learning and adversarial strategy make an important breakthrough in the adaptability of features, there are two issues to be further explored. First, the hard-assigned pseudo labels on the target domain are risky to the intrinsic data structure. Second, the batch-wise training manner in deep learning limits the description of the global structure. In this paper, a Riemannian manifold learning framework is proposed to achieve transferability and discriminability consistently. As to the first problem, this method establishes a probabilistic discriminant criterion on the target domain via soft labels. Further, this criterion is extended to a global approximation scheme for the second issue; such approximation is also memory-saving. The manifold metric alignment is exploited to be compatible with the embedding space. A theoretical error bound is derived to facilitate the alignment. Extensive experiments have been conducted to investigate the proposal and results of the comparison study manifest the superiority of consistent manifold learning framework., Comment: Accepted to AAAI 2020. Code available: \

Image set recognition has been widely applied in many practical problems like real-time video retrieval and image caption tasks. Due to its superior performance, it has grown into a significant topic in recent years. However, images with complicated variations, e.g., postures and human ages, are difficult to address, as these variations are continuous and gradual with respect to image appearance. Consequently, the crucial point of image set recognition is to mine the intrinsic connection or structural information from the image batches with variations. In this work, a Discriminant Residual Analysis (DRA) method is proposed to improve the classification performance by discovering discriminant features in related and unrelated groups. Specifically, DRA attempts to obtain a powerful projection which casts the residual representations into a discriminant subspace. Such a projection subspace is expected to magnify the useful information of the input space as much as possible, then the relation between the training set and the test set described by the given metric or distance will be more precise in the discriminant subspace. We also propose a nonfeasance strategy by defining another approach to construct the unrelated groups, which help to reduce furthermore the cost of sampling errors. Two regularization approaches are used to deal with the probable small sample size problem. Extensive experiments are conducted on benchmark databases, and the results show superiority and efficiency of the new methods.

Local feature descriptor plays a key role in different image classification applications. Some of these methods such as local binary pattern and image gradient orientations have been proven effective to some extent. However, such traditional descriptors which only utilize single-type features, are deficient to capture the edges and orientations information and intrinsic structure information of images. In this paper, we propose a kernel embedding multiorientation local pattern (MOLP) to address this problem. For a given image, it is first transformed by gradient operators in local regions, which generate multiorientation gradient images containing edges and orientations information of different directions. Then the histogram feature which takes into account the sign component and magnitude component, is extracted to form the refined feature from each orientation gradient image. The refined feature captures more information of the intrinsic structure, and is effective for image representation and classification. Finally, the multiorientation refined features are automatically fused in the kernel embedding discriminant subspace learning model. The extensive experiments on various image classification tasks, such as face recognition, texture classification, object categorization, and palmprint recognition show that MOLP could achieve competitive performance with those state-of-the art methods.

Tackling illumination variation is a major problem and it is also an important challenge for practical face recognition systems. Some related methods consider that lighting intensity components mainly lie in large-scale features, and they use a lot of image decomposition techniques to extract the small-scale illumination-invariant features and remove the large-scale features from original face images. However, it argues that the large-scale features contain a lot useful information which can be further extracted, and the small-scale illumination-invariant features are not robust enough due to they contain some detrimental features (noise, etc.). In this paper, we propose a discriminative multi-layer illumination-robust feature extraction (DMI) model to address this problem. First, we decompose the large-scale features into multi-layer small-scale illumination-robust features as a linear combination, and then a weight is assigned to each layer to adjust its importance and influence. The idea is to take full advantage of these useful information in large-scale features for face recognition. Second, we learn a discriminant filter to improve the robustness and statistical discriminative ability of the reconstructed illumination-robust face for face recognition under poor lighting conditions. Extensive experiments on three benchmark face databases and a video image database show that DMI performs better than the related methods, especially in difficult lighting conditions.

Space-borne equipments produce very big images while their capacities of storage, calculation and transmission are limited, so low-complexity image compression algorithms are necessary. In this paper, we develop an efficient image compression algorithm based on quadtree in wavelet domain for this mission. First, we propose an adaptive scanning order for quadtree, which traverses prior the neighbors of previous significant nodes from bottom to the top of quadtree, so that more significant coefficients are encoded at a specified bit rate. Second, we divide the entire wavelet image to several blocks and encode them individually. Because the distortion-rate usually decreases as the level of the quadtree increases with the adaptive scanning order, to control bit rate for each block, we set the points exactly after coding each level of the quadtree as the candidate truncation points. The proposed method can provide quality, position and resolution scalability, which is simple and fast without any entropy coding, so it is very suitable for space-borne equipments. Experimental results show that it attains better performance compared with some state-of-the-art algorithms.

A sparse representation classifier (SRC) and a kernel discriminant analysis (KDA) are two successful methods for face recognition. An SRC is good at dealing with occlusion, while a KDA does well in suppressing intraclass variations. In this paper, we propose kernel extended dictionary (KED) for face recognition, which provides an efficient way for combining KDA and SRC. We first learn several kernel principal components of occlusion variations as an occlusion model, which can represent the possible occlusion variations efficiently. Then, the occlusion model is projected by KDA to get the KED, which can be computed via the same kernel trick as new testing samples. Finally, we use structured SRC for classification, which is fast as only a small number of atoms are appended to the basic dictionary, and the feature dimension is low. We also extend KED to multikernel space to fuse different types of features at kernel level. Experiments are done on several large-scale data sets, demonstrating that not only does KED get impressive results for nonoccluded samples, but it also handles the occlusion well without overfitting, even with a single gallery sample per subject.

The remote sensing image data is so vast that it requires compression by low-complexity algorithm on space-borne equipment. Binary tree coding with adaptive scanning order (BTCA) is an effective algorithm for the mission. However, for large-scale remote sensing images, BTCA requires a lot of memory, and does not provide random access property. In this paper, we propose a new coding method based on BTCA and optimize truncation. The wavelet image is first divided into several blocks which are encoded individually by BTCA. According the property of BTCA, we select the valid truncation points for each block carefully to optimize the ratio of rate-distortion, so that a higher compression ratio, lower memory requirement and random access property are attained. Without any entropy coding, the proposed method is simple and fast, which is very suitable for space-borne equipment. Experiments are conducted on three remote sensing image sets, and the results show that it can significantly improve PSNR, SSIM and VIF, as well as subjective visual experience.

Because of the dramatic intra-class variations in lighting, expression and pose of face images, no single feature is rich enough to capture all the discriminant information, fusing multiple features is an efficient way to improve performance for face recognition. But most of the existing fusing methods use features sampling at fixed gird and manually set too many parameters, thus their performances are limited. In this paper, we first propose an improved landmark-based multi-scale LBP feature to address the dramatic pose and expression variations, which samples features around landmarks instead of fixed grid. Then we propose a novel model which fuses LBP feature and Gabor feature at kernel-level to capture the information of facial texture and facial shape, where the weighted coefficients between kernels, the discriminant projection matrix and the standard deviations of RBF kernel are simultaneously learnt by the proposed optimization algorithm. Experiments are done on LFW, AR and Extended Yale B datasets, and results show that not only does the proposed method get much better identification performance than some state-of-the-art methods, but it also achieves competitive result for verification task.

The dimensionality reduction methods based on linear embedding, such as neighborhood preserving embedding (NPE), sparsity preserving projections (SPP) and collaborative representation based projections (CRP), try to preserve a certain kind of linear representation for each sample after projection. However, in the transformed low-dimensional space, the linear relationship between the samples may be changed, which cannot make the linear representation-based classifiers, such as sparse representation-based classifier (SRC), to achieve higher recognition accuracy. In this paper, we propose a new linear dimensionality reduction algorithm, called Regularized Coplanar Discriminant Analysis (RCDA) to address this problem. It simultaneously seeks a linear projection matrix and some linear representation coefficients that make the samples from the same class coplanar and the samples from different classes not coplanar. The proposed regularization term balances the bias from the optimal linear representation and that from the class mean to avoid overfitting the training data, and overcomes the matrix singularity in solving the linear representation coefficients. An alternative optimization approach is proposed to solve the RCDA model. Experiments are done on several benchmark face databases and hyperspectral image databases, and results show that RCDA can obtain better performance than other dimensionality reduction methods.

Domain adaptation (DA) and transfer learning with statistical property description is very important in image analysis and data classification. This article studies the domain adaptive feature representation problem for the heterogeneous data, of which both the feature dimensions and the sample distributions across domains are so different that their features cannot be matched directly. To transfer the discriminant information efficiently from the source domain to the target domain, and then enhance the classification performance for the target data, we first introduce two projection matrices specified for different domains to transform the heterogeneous features into a shared space. We then propose a joint kernel regression model to learn the regression variable, which is called feature translator in this article. The novelty focuses on the exploration of optimal experimental design (OED) to deal with the heterogeneous and nonlinear DA by seeking the covariance structured feature translators (CSFTs). An approximate and efficient method is proposed to compute the optimal data projections. Comprehensive experiments are conducted to validate the effectiveness and efficacy of the proposed model. The results show the state-of-the-art performance of our method in heterogeneous DA.

As a powerful approach for exploratory data analysis, unsupervised clustering is a fundamental task in computer vision and pattern recognition. Many clustering algorithms have been developed, but most of them perform unsatisfactorily on the data with complex structures. Recently, Adversarial Auto-Encoder (AAE) shows effectiveness on tackling such data by combining Auto-Encoder (AE) and adversarial training, but it cannot effectively extract classification information from the unlabeled data. In this work, we propose Dual Adversarial Auto-encoder (Dual-AAE) which simultaneously maximizes the likelihood function and mutual information between observed examples and a subset of latent variables. By performing variational inference on the objective function of Dual-AAE, we derive a new reconstruction loss which can be optimized by training a pair of Auto-encoders. Moreover, to avoid mode collapse, we introduce the clustering regularization term for the category variable. Experiments on four benchmarks show that Dual-AAE achieves superior performance over state-of-the-art clustering methods. Besides, by adding a reject option, the clustering accuracy of Dual-AAE can reach that of supervised CNN algorithms. Dual-AAE can also be used for disentangling style and content of images without using supervised information.

Conditional maximum mean discrepancy (CMMD) can capture the discrepancy between conditional distributions by drawing support from nonlinear kernel functions; thus, it has been successfully used for pattern classification. However, CMMD does not work well on complex distributions, especially when the kernel function fails to correctly characterize the difference between intraclass similarity and interclass similarity. In this paper, a new kernel learning method is proposed to improve the discrimination performance of CMMD. It can be operated with deep network features iteratively and thus denoted as KLN for abbreviation. The CMMD loss and an autoencoder (AE) are used to learn an injective function. By considering the compound kernel, that is, the injective function with a characteristic kernel, the effectiveness of CMMD for data category description is enhanced. KLN can simultaneously learn a more expressive kernel and label prediction distribution; thus, it can be used to improve the classification performance in both supervised and semisupervised learning scenarios. In particular, the kernel-based similarities are iteratively learned on the deep network features, and the algorithm can be implemented in an end-to-end manner. Extensive experiments are conducted on four benchmark datasets, including MNIST, SVHN, CIFAR-10, and CIFAR-100. The results indicate that KLN achieves the state-of-the-art classification performance.

Learning domain adaptive features aims to enhance the classification performance of the target domain by exploring the discriminant information from an auxiliary source set. Let ${X}$ denote the feature and ${Y}$ as the label. The most typical problem to be addressed is that ${P} _{\textit {XY}}$ has a so large variation between different domains that classification in the target domain is difficult. In this paper, we study the generalized conditional domain adaptation (DA) problem, in which both ${P} _{Y}$ and ${P} _{X|Y}$ change across domains, in a causal perspective. We propose transforming the class conditional probability matching to the marginal probability matching problem, under a proper assumption. We build an intermediate domain by employing a regression model. In order to enforce the most relevant data to reconstruct the intermediate representations, a low-rank constraint is placed on the regression model for regularization. The low-rank constraint underlines a global algebraic structure between different domains, and stresses the group compactness in representing the samples. The new model is considered under the discriminant subspace framework, which is favorable in simultaneously extracting the classification information from the source domain and adaptation information across domains. The model can be solved by an alternative optimization manner of quadratic programming and the alternative Lagrange multiplier method. To the best of our knowledge, this paper is the first to exploit low-rank representation, from the source domain to the intermediate domain, to learn the domain adaptive features. Comprehensive experimental results validate that the proposed method provides better classification accuracies with DA, compared with well-established baselines.

We propose a set of novel radial basis functions with adaptive input and composite trend representation (AICTR) for portfolio selection (PS). Trend representation of asset price is one of the main information to be exploited in PS. However, most state-of-the-art trend representation-based systems exploit only one kind of trend information and lack effective mechanisms to construct a composite trend representation. The proposed system exploits a set of RBFs with multiple trend representations, which improves the effectiveness and robustness in price prediction. Moreover, the input of the RBFs automatically switches to the best trend representation according to the recent investing performance of different price predictions. We also propose a novel objective to combine these RBFs and select the portfolio. Extensive experiments on six benchmark data sets (including a new challenging data set that we propose) from different real-world stock markets indicate that the proposed RBFs effectively combine different trend representations and AICTR achieves state-of-the-art investing performance and risk control. Besides, AICTR withstands the reasonable transaction costs and runs fast; hence, it is applicable to real-world financial environments.

Generative adversarial networks are useful tools in image generation task, but training and running them are relatively slow due to the large amount parameters introduced by their generators. In this paper, S-Deconv, a sparse drop-in substitute for deconvolution layers, is proposed to alleviate this issue. S-Deconv decouples reshaping input tensor from reweighing it by first processing it with a sparse fixed filter into desired form then reweighing them using learnable one. By doing so, S-Deconv reduces the numbers of learnable and total parameters with sparsity. Our experiments on Fashion-MNIST, CelebA and Anime-Faces verify the feasibility of our method. We also give another interpretation of our method from the perspective of regularization.

Image sets-based face recognition receives growing research interest in pattern recognition and machine learning. The most challenging problem focuses on how to formulate a computable and discriminative model by using given data sets. In this paper, we propose a new method, which is called Bilinear Regression Classifier (BLRC) for short, to address the image sets-based face recognition problem. BLRC classifies a given test set by choosing the category that simultaneously maximizes the unrelated subspace and minimize the related subspace. In particular, the unrelated subspace is used to characterize the distances between the query set and the unrelated image sets, while the related subspace is used to characterize the distances between the query set and the related sets. In our work, the Mahalanobis metric, rather than the Euclidean metric, is exploited to compute the subspace distance. The subspace coefficient vectors are obtained by solving an Elastic-Net regularized regression model. Extensive experiments are conducted on several benchmark datasets to evaluate the real recognition performance of the new method. The results show that our BLRC method obtains competitive accuracies with some state-of-the-art methods.

We propose a novel linear learning system based on the peak price tracking (PPT) strategy for portfolio selection (PS). Recently, the topic of tracking control attracts intensive attention and some novel models are proposed based on backstepping methods, such that the system output tracks a desired trajectory. The proposed system has a similar evolution with a transform function that aggressively tracks the increasing power of different assets. As a result, the better performing assets will receive more investment. The proposed PPT objective can be formulated as a fast backpropagation algorithm, which is suitable for large-scale and time-limited applications, such as high-frequency trading. Extensive experiments on several benchmark data sets from diverse real financial markets show that PPT outperforms other state-of-the-art systems in computational time, cumulative wealth, and risk-adjusted metrics. It suggests that PPT is effective and even more robust than some defensive systems in PS.

Subspace learning for dimensionality reduction is an important topic in pattern analysis and machine learning, and it has extensive applications in feature representation and image classification. Linear discriminant analysis (LDA) is a well-known subspace learning approach for supervised dimensionality reduction due to its effectiveness and efficacy in discriminant analysis. However, LDA is not stable and suffers from the singularity problem when addressing small sample size and high-dimensional data. In this paper, we develop a novel subspace learning model, named sparse approximation to discriminant projection learning (SADPL), to learn the sparse projection matrix. Different from the traditional LDA-based methods, we learn the projection matrix based on a new objective function rather than the Fisher criterion, which avoids the matrix singularity problem. In order to distinguish which features play an important role in discriminant analysis, we embed a feature selection framework to the subspace learning model to select the informative features. Finally, we can attain a convex objective function which can be solved by an effective optimization algorithm, and theoretically prove the convergence of the proposed optimization algorithm. Extensive experiments on all sorts of image classification tasks, such as face recognition, palmprint recognition, object categorization and texture classification show that our SADPL achieves competitive performance compared to the state-of-the-art methods.

Person re-identification (re-ID) is to match different images of the same pedestrian. It has attracted increasing research interest in pattern recognition and machine learning. Traditionally, person re-ID is formulated as a metric learning problem with binary classification output. However, higher order relationship, such as triplet closeness among the instances, is ignored by such pair-wise based metric learning methods. Thus, the discriminative information hidden in these data is insufficiently explored. This paper proposes a new structured loss function to push the frontier of the person re-ID performance in realistic scenarios. The new loss function introduces two margin parameters. They operate as bounds to remove positive pairs of very small distances and negative pairs of large distances. A trade-off coefficient is assigned to the loss term of negative pairs to alleviate class-imbalance problem. By using a linear function with the margin-based objectives, the gradients w.r.t. weight matrices are no longer dependent on the iterative loss values in a multiplicative manner. This makes the weights update process robust to large iterative loss values. The new loss function is compatible with many deep learning architectures, thus, it induces new deep network with pair-pruning regularization for metric learning. To evaluate the performance of the proposed model, extensive experiments are conducted on benchmark datasets. The results indicate that the new loss together with the ResNet-50 backbone has excellent feature representation ability for person re-ID.

Dimension reduction is an important topic in pattern analysis and machine learning, and it has wide applications in feature representation and pattern classification. In the past two decades, sliced inverse regression (SIR) has attracted much research efforts due to its effectiveness and efficacy in dimension reduction. However, two drawbacks limit further applications of SIR. First, the computation complexity of SIR is usually high in the situation of high-dimensional data. Second, sparsity of projection subspace is not well mined for improving the feature selection and model interpretation abilities. This paper proposes to compute the SIR projection vectors in the spectral space, then an approximated regression solution can be obtained with a faster speed. Moreover, the adaptive lasso is used to attain a sparse and globally optimal solution, which is important in variable selection. To complete the robust pattern classification task with corruptions, a correntropy-based and class-wise regression model is designed in this paper. It takes a smooth penalty instead of sparsity constraint in the regression coefficients, and it can be conducted in class-wise, thus it is more flexible in practice. Extensive experiments are conducted by using some real and benchmark data sets, e.g., high-dimensional facial images and gene microarray data, to evaluate the new algorithms. The new proposals attain competitive results and are compared with other state-of-the-art methods.

Robust descriptor-based subspace learning with complex data is an active topic in pattern analysis and machine intelligence. A few researches concentrate the optimal design on feature representation and metric learning. However, traditionally used features of single-type, e.g., image gradient orientations (IGOs), are deficient to characterize the complete variations in robust and discriminant subspace learning. Meanwhile, discontinuity in edge alignment and feature match are not been carefully treated in the literature. In this paper, local order constrained IGOs are exploited to generate robust features. As the difference-based filters explicitly consider the local contrasts within neighboring pixel points, the proposed features enhance the local textures and the order-based coding ability, thus discover intrinsic structure of facial images further. The multimodal features are automatically fused in the most discriminant subspace. The utilization of adaptive interaction function suppresses outliers in each dimension for robust similarity measurement and discriminant analysis. The sparsity-driven regression model is modified to adapt the classification issue of the compact feature representation. Extensive experiments are conducted by using some benchmark face data sets, e.g., of controlled and uncontrolled environments, to evaluate our new algorithm.