Consensus Graph Learning for Incomplete Multi-view Clustering

Abstract

Multi-view data clustering is a fundamental task in current machine learning, known as multi-view clustering. Existing multi-view clustering methods mostly assume that each data instance is sampled in all views. However, in real-world applications, it is common that certain views miss number of data instances, resulting in incomplete multi-view data. This paper concerns the task of clustering of incomplete multi-view data. We propose a novel Graph-based Incomplete Multi-view Clustering (GIMC) to perform this task. GIMC can effectively construct a complete graph for each view with the help of other view(s), and automatically weight each constructed graph to learn a consensus graph, which gives the final clusters. An alternating iterative optimization algorithm is proposed to optimize the objective function. Experimental results on real-world datasets show that the proposed method outperforms state-of-the-art baseline methods markedly.