Unsupervised Maximum Margin Incomplete Multi-view Clustering

Discarding incomplete multi-view instances in conventional multi-view algorithms leads to a severe loss of available information. To make up for this loss, learning from multi-view incomplete data has attracted much attention. With the goal of better clustering, the Unsupervised Maximum Margin Incomplete Multi-view Clustering (UMIMC) algorithm is proposed in this paper. Different from the existing works that simply project data into a common subspace, discriminative information is incorporated into the unified representation by applying the unsupervised maximum margin criterion. Thus, the margin between different classes is enlarged in the learned subspace, leading to improvement in the clustering performance. An alternating iterative algorithm with guaranteed convergence is developed for optimization. Experimental results on several datasets verify the effectiveness of the proposed method.