Hierarchical Multiple Kernel Clustering

Current multiple kernel clustering algorithms compute a partition with the consensus or graph learned from pre-specified ones, while emerging late fusion methods firstly construct partitions each separately, and then obtain one them. However, both of them directly distill information kernels graphs to matrices, where sudden dimension drop would result in loss advantageous details for clustering. In this paper, we provide brief insight aforementioned issue propose hierarchical approach perform preserving maximumly. Specifically, gradually group samples into fewer clusters, together generating sequence intermediary matrices descending sizes. The is simultaneously conversely guides construction matrices. Nevertheless, cyclic process modeled an unified objective alternative algorithm designed solve it. addition, proposed method validated compared other representative on benchmark datasets, demonstrating state-of-the-art performance by large margin.