Multi-view Subspace Clustering via Joint Latent Representations

Wenhua Dong, Xiaojun Wu, Tianyang Xu

Abstract: Multi-view data are usually collected from distinct sources or domains which lead to each view owning both specific physical attributes and shared attributes. How to make better use of the consistency and complementarity of multiple views to improve clustering performance is a challenging problem in multi-view subspace clustering task. In this paper, we propose a novel multi-view subspace clustering method which learns shared and specific latent representations, and corresponding self-representations with local structure preserving by projecting all views to their respective low-dimensional latent spaces, called multi-view subspace clustering via joint latent representations. Related optimization problem is effectively solved utilizing the alternating direction method of multipliers. The consistency and complementary information of multiple views is captured by the learned shared and specific representations, respectively, which strengthens the performance of our proposed approach. Experimental results on real datasets demonstrate the effectiveness of the proposed method.