Multiple kernel subspace clustering (MKSC), as an important extension for handling multi-view non-linear subspace data, has shown notable success in a wide variety of machine learning tasks. The key objective of MKSC is to build a flexible and appropriate graph for clustering from the kernel space. However, existing MKSC methods apply a mechanism utilizing the kernel trick to the traditional self-expressive principle, where the similarity graphs are built on the respective high-dimensional (or even infinite) reproducing kernel Hilbert space (RKHS). Regarding this strategy, we argue that the original high-dimensional spaces usually include noise and unreliable similarity measures and, therefore, output a low-quality graph matrix, which degrades clustering performance. In this paper, inspired by projective clustering, we propose the utilization of a complementary similarity graph by fusing the multiple kernel graphs constructed in the low-dimensional partition space, termed projective multiple kernel subspace clustering (PMKSC). By incorporating intrinsic structures with multi-view data, PMKSC alleviates the noise and redundancy in the original kernel space and obtains high-quality similarity to uncover the underlying clustering structures. Furthermore, we design a three-step alternate algorithm with proven convergence to solve the proposed optimization problem. The experimental results on ten multiple kernel benchmark datasets validate the effectiveness of our proposed PMKSC, compared to the state-of-the-art multiple kernel and kernel subspace clustering methods, by a large margin. Our code is available athttps://github.com/MengjingSun/PMKSC-code.

Taking the assumption that data samples are able to be reconstructed with the dictionary formed by themselves, recent multiview subspace clustering (MSC) algorithms aim to find a consensus reconstruction matrix via exploring complementary information across multiple views. Most of them directly operate on the original data observations without preprocessing, while others operate on the corresponding kernel matrices. However, they both ignore that the collected features may be designed arbitrarily and hard guaranteed to be independent and nonoverlapping. As a result, original data observations and kernel matrices would contain a large number of redundant details. To address this issue, we propose an MSC algorithm that groups samples and removes data redundancy concurrently. In specific, eigendecomposition is employed to obtain the robust data representation of low redundancy for later clustering. By utilizing the two processes into a unified model, clustering results will guide eigendecomposition to generate more discriminative data representation, which, as feedback, helps obtain better clustering results. In addition, an alternate and convergent algorithm is designed to solve the optimization problem. Extensive experiments are conducted on eight benchmarks, and the proposed algorithm outperforms comparative ones in recent literature by a large margin, verifying its superiority. At the same time, its effectiveness, computational efficiency, and robustness to noise are validated experimentally.

Deep clustering is a fundamental yet challenging task for data analysis. Recently we witness a strong tendency of combining autoencoder and graph neural networks to exploit structure information for clustering performance enhancement. However, we observe that existing literature 1) lacks a dynamic fusion mechanism to selectively integrate and refine the information of graph structure and node attributes for consensus representation learning; 2) fails to extract information from both sides for robust target distribution (i.e., “groundtruth” soft labels) generation. To tackle the above issues, we propose a Deep Fusion Clustering Network (DFCN). Specifically, in our network, an interdependency learning-based Structure and Attribute Information Fusion (SAIF) module is proposed to explicitly merge the representations learned by an autoencoder and a graph autoencoder for consensus representation learning. Also, a reliable target distribution generation measure and a triplet self-supervision strategy, which facilitate cross-modality information exploitation, are designed for network training. Extensive experiments on six benchmark datasets have demonstrated that the proposed DFCN consistently outperforms the state-of-the-art deep clustering methods.

Multi-view spectral clustering aims to group data into different categories by optimally exploring complementary information from multiple Laplacian matrices. However, existing methods usually linearly combine a group of pre-specified first-order Laplacian matrices to construct an optimal Laplacian matrix, which may result in limited representation capability and insufficient information exploitation. In this paper, we propose a novel optimal neighborhood multi-view spectral clustering (ONMSC) algorithm to address these issues. Specifically, the proposed algorithm generates an optimal Laplacian matrix by searching the neighborhood of both the linear combination of the first-order and high-order base Laplacian matrices simultaneously. This design enhances the representative capacity of the optimal Laplacian and better utilizes the hidden high-order connection information, leading to improved clustering performance. An efficient algorithm with proved convergence is designed to solve the resultant optimization problem. Extensive experimental results on 9 datasets demonstrate the superiority of our algorithm against state-of-the-art methods, which verifies the effectiveness and advantages of the proposed ONMSC.

Self-supervised learning is a thriving research direction since it can relieve the burden of human labeling for machine learning by seeking for supervision from data instead of human annotation. Although demonstrating promising performance in various applications, we observe that the existing methods usually model the auxiliary learning tasks as classification tasks with finite discrete labels, leading to insufficient supervisory signals, which in turn restricts the representation quality. In this paper, to solve the above problem and make full use of the supervision from data, we design a regression model to predict the continuous parameters of a group of transformations, i.e., image rotation, translation, and scaling. Surprisingly, this naive modification stimulates tremendous potential from data and the resulting supervisory signal has largely improved the performance of image representation learning. Extensive experiments on four image datasets, including CIFAR10, CIFAR100, STL10, and SVHN, indicate that our proposed algorithm outperforms the state-of-the-art unsupervised learning methods by a large margin in terms of classification accuracy. Crucially, we find that with our proposed training mechanism as an initialization, the performance of the existing state-of-the-art classification deep architectures can be preferably improved.

Deep clustering gains superior performance than conventional clustering by jointly performing feature learning and cluster assignment. Although numerous deep clustering algorithms have emerged in various applications, most of them fail to learn robust cluster-oriented features which in turn hurts the final clustering performance. To solve this problem, we propose a two-stage deep clustering algorithm by incorporating data augmentation and self-paced learning. Specifically, in the first stage, we learn robust features by training an autoencoder with examples that are augmented by random shifting and rotating the given clean examples. Then, in the second stage, we encourage the learned features to be cluster-oriented by alternatively finetuning the encoder with the augmented examples and updating the cluster assignments of the clean examples. During finetuning the encoder, the target of each augmented example in the loss function is the center of the cluster to which the clean example is assigned. The targets may be computed incorrectly, and the examples with incorrect targets could mislead the encoder network. To stabilize the network training, we select most confident examples in each iteration by utilizing the adaptive self-paced learning. Extensive experiments validate that our algorithm outperforms the state of the arts on four image datasets.

Existing deep neural networks mainly focus on learning transformation invariant features. However, it is the equivariant features that are more adequate for general purpose tasks. Unfortunately, few work has been devoted to learning equivariant features. To fill this gap, in this paper, we propose an affine equivariant autoencoder to learn features that are equivariant to the affine transformation in an unsupervised manner. The objective consists of the self-reconstruction of the original example and affine transformed example, and the approximation of the affine transformation function, where the reconstruction makes the encoder a valid feature extractor and the approximation encourages the equivariance. Extensive experiments are conducted to validate the equivariance and discriminative ability of the features learned by our affine equivariant autoencoder.

Determining the number of clusters is important but challenging, especially for data of high dimension. In this paper, we propose Deep Embedding Determination (DED), a method that can solve jointly for the unknown number of clusters and feature extraction. DED first combines the virtues of the convolutional autoencoder and the t-SNE technique to extract low dimensional embedded features. Then it determines the number of clusters using an improved density-based clustering algorithm. Our experimental evaluation on image datasets shows significant improvement over state-of-the-art methods and robustness with respect to hyperparameter settings.

Deep Embedded Clustering (DEC) surpasses traditional clustering algorithms by jointly performing feature learning and cluster assignment. Although a lot of variants have emerged, they all ignore a crucial ingredient, \emph{data augmentation}, which has been widely employed in supervised deep learning models to improve the generalization. To fill this gap, in this paper, we propose the framework of Deep Embedded Clustering with Data Augmentation (DEC-DA). Specifically, we first train an autoencoder with the augmented data to construct the initial feature space. Then we constrain the embedded features with a clustering loss to further learn clustering-oriented features. The clustering loss is composed of the target (pseudo label) and the actual output of the feature learning model, where the target is computed by using clean (non-augmented) data, and the output by augmented data. This is analogous to supervised training with data augmentation and expected to facilitate unsupervised clustering too. Finally, we instantiate five DEC-DA based algorithms. Extensive experiments validate that incorporating data augmentation can improve the clustering performance by a large margin. Our DEC-DA algorithms become the new state of the art on various datasets.

Deep clustering learns deep feature representations that favor clustering task using neural networks. Some pioneering work proposes to simultaneously learn embedded features and perform clustering by explicitly defining a clustering oriented loss. Though promising performance has been demonstrated in various applications, we observe that a vital ingredient has been overlooked by these work that the defined clustering loss may corrupt feature space, which leads to non-representative meaningless features and this in turn hurts clustering performance. To address this issue, in this paper, we propose the Improved Deep Embedded Clustering (IDEC) algorithm to take care of data structure preservation. Specifically, we manipulate feature space to scatter data points using a clustering loss as guidance. To constrain the manipulation and maintain the local structure of data generating distribution, an under-complete autoencoder is applied. By integrating the clustering loss and autoencoder's reconstruction loss, IDEC can jointly optimize cluster labels assignment and learn features that are suitable for clustering with local structure preservation. The resultant optimization problem can be effectively solved by mini-batch stochastic gradient descent and backpropagation. Experiments on image and text datasets empirically validate the importance of local structure preservation and the effectiveness of our algorithm.

Correlation tracker has made a huge success in visual object tracking. However, it is mainly because that the tracker cannot catch the occurrence of appearance changes, tracking based on correlation filters often drifts due to the unexpected appearance changes caused by occlusion, deformation and background clutter. In this paper, we propose a new method to detect the case when the tracker encountered the unexpected appearance changes. This method uses the following points: 1) Filter response curve would decreases dramatically when target suffers heavy appearance changes. 2) Features extracted from deeper layers of convolutional neural networks (CNNs) have more semantics information and features extracted from shadower layers have more spatial information. Extensive experimental results on several public benchmark datasets show that the proposed method can deal with the appearance changes effectively.

Domain adaptation (DA) aims to eliminate the difference between the distribution of labeled source domain on which a classifier is trained and that of unlabeled or partly labeled target domain to which the classifier is to be applied. Compared with the semi-supervised domain adaptation where some labeled data from target domain is utilized to help train the classifier, the unsupervised domain adaptation where no labels can be seen from the target domain is without doubt more challenging. Most published approaches suffer from high complexity of designment or implementation. In this paper, we propose a simple method for unsupervised domain adaptation which minimizes domain shift by projecting each instance from source and target domains into a common feature space using a linear kernel function. Our method is extremely simple without hyper-parameters (it can be implemented in two lines of Matlab code) but still outperforms the state-of-the-art domain adaptation approaches on standard benchmark datasets.