A novel consensus learning approach to incomplete multi-view clustering. (July 2021)

• Record Type: Journal Article
• Authors: Liu, Jianlun; Teng, Shaohua; Fei, Lunke; Zhang, Wei; Fang, Xiaozhao; Zhang, Zhuxiu; Wu, Naiqi
• Is Part Of: Pattern recognition. Volume 115(2021)
• Publication Date: 2021-07
• Subjects: Multi-view clustering -- Incomplete multi-view clustering -- Consensus representation -- Consensus similarity graph; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2021.107890
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: A framework of jointly exploiting the complementary multi-view information of original data representation and the underlying cross-view relations among data points is proposed. The proposed method can partition unlabeled data with or without negative entries and handle complete as well as various incomplete multi-view scenarios with missing instances. An iterative optimization algorithm is proposed for solving the objective function. Extensive experiments on eight multi-view datasets demonstrate that the proposed method outperforms eight state-of-the-art methods. Abstract: Multi-view data may lose some instances in real applications. Most existing methods for clustering such incomplete multi-view data still have at least one of the following limitations: 1) The common relations among data points across all views are ignored. 2) The complementary multi-view information of original data representation is not well exploited. 3) Arbitrary incomplete scenarios or data with negative entries cannot be handled. To address these limitations, in this paper, we propose a novel Consensus Learning approach to Incomplete Multi-view Clustering (CLIMC). Specifically, a low-dimensional consensus representation is introduced to exploit complementary multi-view information from the original feature representation of available instances by integrating index matrices into matrix factorization. In addition, by combining self-representation, index matrices, and consensus term, a consensus similarity graph is leveraged to explore the underlying cross-view relations among data points. Further, the key of the proposed CLIMC is that the consensus representation is correlated with the similarity graph by a graph Laplacian regularization. Consequently, the compactness of the low-dimensional representation and the accuracy of similarity degree of the graph are reciprocally promoted. Extensive experiments on several multi-view datasets demonstrate the effectiveness of CLIMC over state-of-the-arts.