A scalable Consensus Fast Graph Filtering approach for late fusion multi-view clustering

The rapid growth of multi-view data presents significant challenges for clustering algorithms due to its complexity and high dimensionality. Late fusion multi-view clustering (LFMVC) often suffer from low-quality base partitions. To address these challenges, we propose a scalable method called Consensus Fast Graph Filtering for late fusion (CFGFLF) multi-view clustering. This approach integrates multi-view consensus graph filtering with discrete clustering into a unified optimization framework, enhancing clustering accuracy and keeping efficiency. CFGFLF constructs bipartite graphs for each view to capture local relationships, applies higher-order graph diffusion to model global relationships, and refines base partitions through the low-pass filtering property of graph filters. By avoiding costly operations like matrix inversions and utilizing low-rank bipartite graph structures, CFGFLF achieves linear complexity for base partition filtering. Experimental results show that CFGFLF outperforms state-of-the-art methods in clustering accuracy, particularly for large-scale datasets and noisy environments, without sacrificing computational efficiency. The codes of this paper are released in https://github.com/GuoYiqing1/CFGFLF.