Contrastive Consistency and Attentive Complementarity for Deep Multi-View Subspace Clustering

Abstract

Deep multi-view subspace clustering (DMVSC) based on self-expression has attracted increasing attention due to its outstanding performance and nonlinear application. However, most existing methods neglect that viewprivate meaningless information or noise may interfere with the learning of self-expression, which may lead to the degeneration of clustering performance. In this paper, we propose a novel framework of Contrastive Consistency and Attentive Complementarity (CCAC) for DMVsSC. CCAC aligns all the self-expressions of multiple views and fuses them based on their discrimination, so that it can effectively explore consistent and complementary information for achieving precise clustering. Specifically, the view-specific self-expression is learned by a selfexpression layer embedded into the auto-encoder network for each view. To guarantee consistency across views and reduce the effect of view-private information or noise, we align all the view-specific self-expressions by contrastive learning. The aligned self-expressions are assigned adaptive weights by channel attention mechanism according to their discrimination. Then they are fused by convolution kernel to obtain consensus self-expression with maximum complementarity of multiple views. Extensive experimental results on four benchmark datasets and one large-scale dataset of the CCAC method outperform other state-of-the-art methods, demonstrating its clustering effectiveness.