Multi-view Clustering Network Based on Contrastive Learning and CycleGAN

Recently multi-view clustering has become a research hotspot. Compared with the single view clustering, it can obtain more comprehensive sample information. However, the existing work only focuses on extracting more information in the multi-view data and always ignores their identification. Besides, multi-view data cannot achieve deeper fusion. To address these issues, in this paper, we propose a new network for multi-view clustering.Firstly, we design a contrastive learning filter for the raw data to remove useless private information in multiple views. Secondly, we design a cycle generative adversarial network (CycleGAN) for the filtered data, which makes the data among different views reach a deeper integration. Finally, new clustering loss functions are designed to obtain a better clustering structure. The experimental results on four databases with different view numbers prove the superiority of our method.