Constrained Multilinear Multi-view Subspace Representation Learning for Clustering based on Tensor Nuclear Norm

Recently, self-representation learning has received more and more attention. In this paper, we propose a new constrained multilinear multi-view subspace representation learning based on tensor nuclear norm approach, which is called CMMSRL. The CMMSRL approach simultaneously exploits multiple views and prior constraint to promote the performance of the classical subspace representation learning approaches. To utilize different views, a tensor is built by coalescing the representation matrices of different views. The tensor nuclear norm (TNN) based on tensor singular value decomposition (t-SVD) is imposed on the tensor to exploit the high-order correlation of multi-view data. Besides, a constraint matrix, which is elegantly unified into the subspace representation learning, is devised to incorporate prior information. Utilizing the augmented Lagrangian method to resolve the CMMSRL. To verify the performance of the CMMSRL approach, some experiments are executed on datasets and the experimental results exhibit that the CMMSRL outperforms some advanced multi-view clustering methods.