Multi-view Clustering And Semi-supervised Model Based On Tensor

With the development of information acquisition technology,the data collected from diverse fields with various features,forming multi-view data.Due to the correlation between multiple views,the multi-view data is approximately distributed in a low-rank subspace.Based on this assumption,this master's thesis has completed the following work for multi-view clustering and semi-supervised classification:1.In the multi-view clustering task,we propose a tensor Schatten-p norm based on tensor Singular Value Decomposition,and introduce it into subspace clustering framework,so as to propose a novel multi-view clustering framework,called multi-view subspace clustering based on tensor Schatten-p norm(MVSC-TCP).As a non-convex approximation of tensor multi-rank function,Schatten-p norm improves the clustering performance of the algorithm.2.Basing on the multi-view semi-supervised classification model,we effectively integrate the two independent stages of affinity construction and label propagation into a unified framework and propose self-taught multi-view semi-supervised learning(ST-MVSSL)algorithm.In this model,by using the ‘feedback' mechanism,the prediction labels of unlabeled samples is used in the semi-supervised learning process,which provides more supervision information for classification,effectively improved the classification accuracy.3.We introduce the non-convex surrogate function of-norm into self-taught framework,and propose a self-taught semi-supervised learning model based on non-convex approximation(NC-STMVSSL).Experiments on real datasets verify the superiority of the proposed methods.