1 Introduction
In recent years, more and more heterogeneous features in the form of multiple modalities (views) are accessible in many real-world applications. For instance, a document is easily described in different languages; a loan scoring system can be constructed based on the customers’ age, income and credit score [1], [2]. It might be feasible to simply concatenate the multi-view features for obtaining multi-view information [3], [4]. However, the view diversity which might influence the final learning result is often ignored in these algorithms. Therefore, exploring the complementary and consistent information among multiple views is a reasonable choice. In the unsupervised learning domain, multi-view clustering makes a contribution to achieving this target. Among them, multi-view subspace clustering methods are conducive to obtaining a prior clustering performance in myriads of tasks [5], [6], [7].