A co-training approach for multi-view density peak clustering

Abstract

In this paper, we propose a multi-view clustering algorithm based on fast search and find of density peaks. We combined the original clustering algorithm with co-training to handle multi-view data and implement self-adapting cluster center selecting through cluster fusion. Based on the assumption that a point would be assigned to the same cluster in all views, we search for the clustering result that agree across the views by continually modifying one view with the clustering from another view. We demonstrate the efficacy of the proposed algorithm on several test cases.