Multi-view subspace clustering networks with local and global graph information

This study investigates the problem of multi-view subspace clustering, goal which is to explore underlying grouping structure data collected from different fields or measurements. Since do not always comply with linear models in many real-world applications, most existing clustering methods based on shallow may fail practice. Furthermore, graph information usually ignored methods. To address aforementioned limitations, we proposed novel networks local and global information, termed MSCNLG, this paper. Specifically, autoencoder are employed multiple views achieve latent smooth representations that suitable for assumption. Simultaneously, by integrating fused into self-expressive layers, MSCNLG obtains common shared representation, can be used get results employing standard spectral algorithm. As an end-to-end trainable framework, method fully valuable views. Comprehensive experiments six benchmark datasets validate effectiveness superiority MSCNLG.