Consistent auto-weighted multi-view subspace clustering

Because the data in practical applications usually satisfy the assumption of mixing subspaces and contain multiple features, multi-view subspace clustering has attracted extensive attention in recent years. In previous work, multi-view information can be utilized comprehensively by considering consistency. However, they often treat the information from different views equally. Because it is difficult to ensure that the information of all views is well mined, difficult view will reduce the performance. In this paper, we propose a novel multi-view subspace clustering method named consistent auto-weighted multi-view subspace clustering (CAMVSC) to overcome the above limitation by weighting automatically the representation matrices of each view. In our model, the density and sparsity are both considered to ensure the learning effect of each view. Although simultaneously using the self-representation and the auto-weighting strategy will bring difficulties to solve the model, we successfully design a special updating scheme to obtain the numerical algorithm, and prove its convergence theoretically. Extensive experimental results demonstrate the effectiveness of our proposed method.