Abstract:Multi-view clustering is a powerful tool for improving clustering results via integrating the heterogeneous features from different views. In this work, we develop a novel invisible weighted method to this issue. Specifically, our proposed method first considers the diversities among different features of each view with kernel function, and then fuses the multiple pre-embeddings of all the views attentively. As a result, a consensus embedding can be obtained with the adaptive invisible weighted model. Besides, we provide an efficient optimization approach to solve the involved optimization problem and provide the corresponding convergence analysis. Extensive experimental results on benchmark datasets validate the superiorities of the proposed method to the state-of-the-art methods.

Multi-View Clustering Based On Invisible Weights