Centerless Multi-View K-means Based on the Adjacency Matrix

Although K-Means clustering has been widely studied due to its simplicity, these methods still have the following fatal drawbacks. Firstly, they need initialize cluster centers, which causes unstable performance. Secondly, poor performance on non-Gaussian datasets. Inspired by affinity matrix, we propose a novel multi-view based adjacency matrix. It maps matrix distance according principle that every sample small from points in neighborhood and large outside of neighborhood. Moreover, this method well exploits complementary information embedded different views minimizing tensor Schatten p-norm regularize third-order consists assignment matrices views. Additionally, avoids initializing centroids obtain stable And there is no compute means clusters so our model not sensitive outliers. Experiment toy dataset shows excellent other experiments several benchmark datasets demonstrate superiority proposed method.