Multi-View Clustering

Multi-view clustering studies how to group the same objects when each is described by several sources or feature spaces, such as images, text, sensors, or graph relations. Methods commonly learn shared and view-specific representations with autoencoders or contrastive objectives, then form clusters through consensus assignments, spectral or graph-based structure, tensor models, or transport-based alignment. A major focus is handling incomplete, noisy, or mismatched views while retaining useful differences between them; related work also considers fairness, streaming assignment, and interpretable clusters. The central tension is how to align views enough to produce coherent clusters without erasing complementary information or relying on assumptions such as clean correspondences and similar relations across views.