Consensus Learning with Complete Graph Regularization for Incomplete Multi-view Clustering

Jie Zhang, Lunke Fei, Shaohua Teng, Qinghua Zhu, Imad Rida, Jie Wen

Abstract: Incomplete multi-view clustering (IMC) is a long-standing and challenging problem due to the fact that multi-view dataset are often incomplete in practical applications. In this paper, we propose a new IMC method by learning the consensus representation of incomplete multi-view with embedding a complete graph regularization. Specifically, we first exploit the common representation of multiple views by embedding missing-view indicator matrix into the matrix factorization operation, such that the learned new representation is well shared among all views. Then, we learn the combination relationship between similarity matrices of each view to obtain the complete similarity graph. By combining the consensus representation with the complete graph regularization, our proposed method can simultaneously learn the consensus representation and preserve the local geometric information of both the available and missing instances of multiple views. Experimental results on several widely used missing-view datasets clearly demonstrate the promising clustering performance of the proposed method in comparison with most state-of-the-art IMC methods.