Learning an enhanced consensus representation for multi-view clustering via latent representation correlation preserving

Abstract

Multi-view clustering has attracted significant attention due to its sufficiency of exploiting information available from multi-view data to generate a consensus representation across multiple views to improve clustering performance. However, most existing methods merely focus on learning the consistency or complementarity of multi-view data in the original space or latent embedding space alone, which limits their ability to learn an informative consensus representation. To solve this, we propose a novel multi-view clustering method, namely, learning an enhanced consensus representation via latent representation correlation preserving (LRCP). LRCP employs the adaptive neighbors graph approach to construct the initial affinity matrices as input. By introducing a latent representation correlation preserving model, the consensus representation of the initial affinity matrices in the original space and the shared latent representation of the initial affinity matrices in the latent embedding space are integrated into a unified optimization framework to learn an enhanced consensus representation with enriched information and improved representation capability. The learned consensus representation can not only incorporate consensus affinity structure information that is admitted by all views in the original space but also reconstruct the complementary affinity information that is learned by the shared latent representation in the latent space. An efficient optimization algorithm is developed to tackle the resultant optimization problem. Experiments on benchmark datasets have validated the efficacy of the proposed method.