Robust high-order graph learning for incomplete multi-view clustering

Incomplete multi-view clustering is a challenging problem in unsupervised learning that involves grouping multi-view data with missing information into distinct clusters. In this study, we propose a novel robust high-order graph learning (RHGL) method for incomplete multi-view clustering. Specifically, a robust principal component analysis is first employed to gain error-free matrices based on the original available data and then the obtained robust representations are utilized for local manifold learning. Next, we align the view-specific partitions generated from the local similarity structures and perform complete similarity graph learning to explore the global similarity structure. To capture the high-order relationships among views, all complete similarity graphs are stacked into a third-order tensor with the low-rank constraint. As a result, our model can simultaneously explore both local and global similarity structures and mine high-order inter-view correlations. An effective iterative optimization strategy is developed to solve the proposed approach. Experiments on six benchmark datasets show that our model outperforms the competing methods.