Projection-based coupled tensor learning for robust multi-view clustering

Abstract

Multi-view clustering methods based on tensor learning have received extensive attention due to their ability to effectively mine high-order correlation information between views. However, the presence of noise and redundant information in multi-view data can seriously interfere with the performance of clustering tasks. To this end, we propose a projection-based coupled tensor learning method (PCTL). In particular, we first construct an orthogonal projection matrix to obtain the main characteristic information of the raw data of each view and learn the representation matrix in a clean embedding space. Then, we use tensor learning to couple the projection matrix and the representation matrix to mine the high-order information between views and construct a more suitable and optimal representation of the embedding space. A large number of experiments prove that PCTL can effectively suppress the interference of noise and redundant information, and the clustering performance is better than some existing excellent algorithms.