Multiple kernel subspace representation and graph construction learning for multi-view clustering

Considering that multi-view descriptions naturally contain comprehensive information of the target object compared with single-view data, leading multi-view data analysis a hot topic for researchers. In this field, subspace representation based multi-view clustering has the advantage of capturing global information contained in the original data set. However, it can not effectively deal with the nonlinear problem. Although kernel technique has been adopted to deal with this problem, a proper kernel is difficult to select. In this paper, we propose the multiple kernel subspace representation to remark the dependence between different multi-view data sets and recover the consensus representation. Notably, our model can get rid of the difficulty of kernel selections. Furthermore, the ability of subspace representation can be enhanced by integrating an effective information fusion of graph construction to capture graphic information in the original data set. Then, we develop an iterative optimization approach for our proposed model. Finally, we conduct experiments on several benchmark multi-view data sets in terms of five evaluation metrics to valuate the effectiveness of our method compared with related multi-view learning methods.