Multi-view clustering with interactive mechanism

Existing multi-view clustering methods either seek to directly learn a consistent spectral embedding, or to learn a consistent graph. This work presents a novel model, called Multi-view Clustering with Interactive Mechanism (MCIM). Using the interactive mechanism, the uniform graph and spectral embedding can be learned alternatively and promote to each other. Furthermore, we perform spectral embedding learning on Grassmann manifold via an implicitly weighted-learning scheme and reveal the clustering result via graph learning. To solve the proposed model, we propose an efficient optimization method and provide the corresponding convergence analysis. The experimental results on real image datasets demonstrate the superiorities of MCIM compared to several SOTA methods.