Self-Representation Based Multiview Subspace Clustering

The key issue of multiview clustering is how to integrate the multiple views and discover the underlying structures.Although existing approaches of learning from multi-view data have been proposed.However,most of them concentrate on supervised or semi-supervised learning,in which a validation set is required o guide the learning process.Unlabeled data are often relatively plentiful compared to labeled data.Especially,with the significant progress of the ability of computation and memory,utilization of the large scale unlabeled data becomes more and more important.However,unsupervised learning is more challenging due to lack of guiding information with labels.In this paper,we focus on unsupervised and weak supervised multiview learning techniques.Specifically,there are three methods as follows:1)Complementarity Enhanced Multiview ClusteringThe main limitation of existing multiview clustering methods is that they ignore the complementarity while constructing the similarity matrices of different views.To this end,we propose a novel multiview clustering method,called Diversity-induced Multiview Subspace Clustering(DiMSC).We utilize the Hilbert Schmidt Independence Criterion(HSIC)as a diversity term to explore the complementarity of multi-view representations.2)High Order Correlation Multiview ClusteringMost existing methods only capture the correlations between pairwise views,but essentially ignore the high order correlations underlying the multiple views.To this end,we establish a novel method called Low-rank Tensor constrained Multiview Subspace Clustering(LT-MSC),which explores the complementary information from multiple views with a low-rank tensor constraint.3)Constrained Multiview ClusteringTo incorporate the prior information in clustering,we propose a constrained multiview clustering method,which considers both the pairwise constraints as well as the multi-view consistence simultaneously under a unified graph-based model.Moreover,take the videos face clustering for example,we propose an constrained multiview video face clustering framework.