Large-scale multi-view clustering based on anchor strategy and tensor collaborative learning

The ability of anchor strategy based multi-view subspace clustering methods for effectively tackling large-scale data has led to its widespread attention in the research community. However, most of the existing algorithms usually focus on the problem of reducing complexity, and neglect to explore complementary information in the clustering process. Therefore, we propose a large-scale multi-view clustering method with anchor strategy and tensor collaborative learning (ASTCL). Specifically, we learn both consensus anchor and each view’s anchor graph directly from the original data to make the learned anchors more representative. Then, we design a new structure of tensor low-rank norm to probe higher-order associations between anchor graphs. At the same time, in an attempt optimize the local structure of the tensor, symmetric constraints are applied to the front slice of the tensor. Finally, the experimental results conducted on datasets of varying scales validate that the methods outlined are relatively valid in terms of improving clustering performance.