Fine Multi-View Feature Selection

With the rapid development of computer science, high-dimensional multi-view data has presented which contains more noise and redundant information and is not conducive to multi-view learning task. It is necessary to reduce the dimension of multi-view data. Multi-view feature selection is a common method of dimensionality reduction for multi-view data, which can effectively eliminate noise and redundant information. In addition, multi-view feature selection requires further mining consistency information and complementary information of multi-view data. To solve above problems, we propose a fine multi-view feature selection method based on supervised environment. The proposed method introduces $l_{2,0}$ -norm to achieve fine feature selection. Further, the proposed method introduces two adaptive weighting parameter learning schemes to mine the correlation information between views and categories. This paper designs an effective alternating iterative optimization algorithm. Experiments on six multi-view datasets demonstrate the effectiveness and superiority of the proposed method.