Re-weighting regression and sparsity regularization for multi-view classification

摘要

Multi-view data widely exists in real life, which contains rich and comprehensive information. Multi-view learning aims to make full use of the information of multiple views to improve the performance of the learner. Most traditional supervised methods learn an entire transformation matrix by concatenating multiple views into a long vector, thus they often ignore the relationship between views. To tackle this problem, in this paper, a novel re-weighting regression and sparsity regularization method for multi-view classification is proposed. The proposed method adopts view-based joint sparsity-inducing norm regularization to reduce the impact of redundant features via exploring the correlation between the features of different views and the predicted categories. Moreover, the model adopts the re-weighting strategy to weigh the importance of different views by adding a proper weight for each view. Benefited from the clustering idea, the proposed model learns several sub-matrices in different view subspaces independently and integrates them into the final decision classifier with different weights to improve the classification performance. Extensive experimental results show that the proposed model obtains better classification performance compared to several state-of-the-art classification methods.