Robust multi-view clustering via inter-and-intra-view low rank fusion

Abstract Multi-view Clustering has become a vital task with the rapidly growing amount of data in multiple representations. Under the context of unsupervised learning, the existing clustering methods show limited capability in leveraging the discriminative information from multiple views. View-independent noises and distribution differences among views are two main causes that may degrade the clustering performances. By introducing the inter-and-intra-view low rank decomposition, we propose Multi-view Clustering via Intra and Inter-view Low Rank Fusion (MCIIF). The proposed method facilitates clustering with the discriminative intra-view low rank structures that complement the shared inter-view transition. By recovering complementary low rank transitions from the differences between the shared inter-view transition and individual-view transitions, we pursue a more accurate transition probability matrix by combining the complementary intra-view transitions with the shared inter-view transition in an additive manner. To solve the corresponding optimization problem, we propose a procedure based on the Alternating Direction Method of Multipliers (ADMM) scheme with convergence guarantees. Experimental results on various real-world datasets verify the effectiveness of the proposed method over the state-of-art multi-view clustering methods.