Metric Multi-View Graph Clustering

Graph-based methods have hitherto been used to pursue the coherent patterns of data due its ease implementation and efficiency. These increasingly applied in multi-view learning achieved promising performance various clustering tasks. However, despite their noticeable empirical success, existing graph-based may still suffer suboptimal solution considering that can be very complicated raw feature space. Moreover, usually adopt similarity metric by an ad hoc approach, which largely simplifies relationship among real-world results inaccurate output. To address these issues, we propose seamlessly integrates graph for clustering. Specifically, employ a useful depict inherent structure with linearity-aware affinity representation learned based on self-expressiveness property. Furthermore, instead directly utilizing features, prefer recover smooth such geometric original retained. We model above concerns into unified framework, hence complements each subtask mutual reinforcement manner. The studies corroborate our theoretical findings, demonstrate proposed method is able boost performance.