Anchor-based fast balanced multi-view clustering.

Abstract

Multi-view clustering has garnered significant research interest due to its ability to integrate complementary information from diverse perspectives. Graph-based multi-view clustering has emerged as a prominent research area in recent years, owing to its superior performance in handling non-convex datasets. Despite the progress made, graph-based multi-view clustering methods still face three key challenges: (1) high computational complexity due to graph construction and spectral analysis; (2) insufficient attention to balanced cluster sizes, which are crucial for practical applications such as resource allocation and load balancing; (3) reliance on a separate post-processing step (e.g., k-means) to discretize the continuous spectral embedding, which introduces bias. To address these issues, we propose Fast Balanced Multi-View Clustering (FBMVC), a fast and balanced graph-based multi-view clustering framework. FBMVC employs anchor and label transmission strategies, significantly reducing computational costs while maintaining performance. The adaptive view-weighting scheme autonomously optimizes view-specific weights to achieve effective multi-view integration. The resulting problem implicitly ensures a balanced clustering distribution. In particular, FBMVC obtains the discrete cluster labels within its optimization framework, thereby avoiding the need for a separate post-processing stage. Experiments on six benchmark datasets demonstrate that FBMVC outperforms state-of-the-art methods in terms of clustering performance and computational efficiency.