Robust Rank-Constrained Sparse Learning A Graph-Based Framework for Single View and Multiview Clustering
Abstract : Graph-based clustering aims to partition the data according to a similarity graph, which has shown impressive performance on various kinds of tasks. The quality of similarity graph largely determines the clustering results, but it is difficult to produce a high-quality one, especially when data contain noises and outliers. To solve this problem, we propose a robust rank constrained sparse learning (RRCSL) method in this article. The L2,1-norm is adopted into the objective function of sparse representation to learn the optimal graph with robustness. To preserve the data structure, we construct an initial graph and search the graph within its neighborhood. By incorporating a rank constraint, the learned graph can be directly used as the cluster indicator, and the final results are obtained without additional postprocessing. In addition, the proposed method cannot only be applied to single-view clustering but also extended to multiview clustering. Plenty of experiments on synthetic and real-world datasets have demonstrated the superiority and robustness of the proposed framework.
? When the data are contaminated with noise or outlier, the performance of existing methods will decline greatly because the original true data distribution is destroyed.
? The most existing multiview graph learning approaches, they generally focus on the elegant graphs construction model and their corresponding multiview learning mechanism but ignore the data themselves, especially for uncertain noisy data.
? However, the data quality restoration are not sufficiently and especially considered in the most existing multiview graph learning approaches.
? Extensive experiments on four visual datasets demonstrate the superior and robustness of RMvSL compared with various existing excellent similarity graph learning approaches.
? To alleviate this problem, some methods update the graph during the clustering procedure, such as clustering with adaptive neighbors (CANs), constrained Laplacian rank (CLR), and simplex sparse representation (SSR).
? In this article, a robust rank constrained sparse learning (RRCSL) method is proposed to solve the above-mentioned problems.
? There are various existing algorithms that can solve this problem effectively , such as the Lagrangian method.
? This is because the RRCSL method has stronger ability on data representation compared with CAN and CLR, and it avoids the problem of SSR that the similarity graph is not unique.
• Many multiview graph learning approaches are proposed and applied to clustering problem.
• To address the two problems, a Robust Multiview Similarity Learning(RMvSL) method is proposed in this paper.
• A multi-graph fusion scheme is proposed in for multiview clustering, which enforces the fusion graph to be approximated to original graph from each view but with an explicit cluster structure.
• Multiview graph learning aims to learn a fusion graph from the different views, which can cover all the useful cues from all views and obtain further performance improvement.
• With the proposed unified framework, each variable is not optimized in isolation during iteration.
? All these ensure that the RRCSL method achieves the best performance on all datasets.
? Among numerous clustering techniques, graph-based clustering methods focus on the internal data structure, and have shown better performance.
? Therefore, the clustering performance relies highly on the graph construction procedure.
? Two widely used clustering performance measures are adopted to evaluate the clustering results, namely, accuracy (ACC) and normalized mutual information (NMI).
? In this way, data can be classified directly and, therefore, the quality and efficiency of clustering have been improved.
|