Visible-Hidden Multi-view Collaborative Fuzzy Clustering Based on Graph Constraints

Among many multi-view clustering methods, collaborative multi-view clustering improves the clustering effect by utilizing collaborative learning among different views. However, most of the existing methods only consider visible views or only consider the hidden information between views. Although some methods achieve joint learning of visible and hidden views, they ignore the utilization of topology. To solve these problems, this paper proposes a visible and hidden multi-view collaborative fuzzy clustering algorithm based on graph constraints. Firstly, the non-negative matrix decomposition is utilized to obtain the share hidden view containing the information of each visible view, and then the collaborative learning of visible and hidden views is realized in the fuzzy clustering process. Secondly, we use the hidden view to build an affinity matrix to describe the topology and impose constraints on membership. In addition, the introduction of graph constraints makes the traditional Lagrange multiplier method unable to be used to optimize the membership, so we use Adam gradient descent method to update the membership. Experimental results on seven benchmark multi-view datasets show that the GMVFC presented in this paper has better performance than other mainstream algorithms.