Beyond Missing Rates: Rethinking Incomplete Multi-View Clustering with Protocol Divergence
The paper introduces CRAFT, a new architecture for Incomplete Multi-View Clustering that solves robustness issues by changing model structure instead of relying on loss functions. It formalizes 'incompleteness divergence,' showing that simple missing rates are misleading. CRAFT achieves this by training once and deploying across many missing patterns, significantly reducing training time while maintaining high performance.