Robust multi-view clustering with hyper-Laplacian regularization

Multi-view clustering has attracted much attention in diverse fields of study for its excellent performance in clustering. However, there are still several issues to be solved. First, many existing methods struggle to handle the presence of noise, which is often encountered in real-world datasets. Second, some methods assume that each view is equally important in the clustering process, overlooking the diversity that multiple views can bring to the table. Third, traditional methods tend to rely on pairwise similarity relationships, which may not fully explore the underlying clustering structures among samples. In this paper, we propose a multi-view clustering method named Robust Multi-view Clustering with Hyper-Laplacian Regularization (RICHIE) to solve these problems. RICHIE indirectly learns the unified representation matrix and adopts a matrix norm to enhance the robustness of the algorithm. Besides, RICHIE can automatically assign the optimal weights to each view, allowing for effective utilization of the diverse multi-view data. Furthermore, RICHIE utilizes hyper-Laplacian regularization on the unified representation matrix to fully exploit the similarity relationship among the data points. In addition, an optimization algorithm is proposed to solve the problem. Experimental evaluations on eight different datasets using 10 baseline methods demonstrated the effectiveness of our algorithm.