Consistent Anchor Induced Multi-View Deep Matrix Factorization

Deep Matrix Factorization (DMF)-based multi-view learning approaches have demonstrated great potential in learning hierarchical semantics of multi-view data, but have paid limited attention to effectively exploiting the consensus attribute of different views. To further enhance their ability for multi-view data analysis, a Consistent Anchor induced Multi-view DMF method is proposed in this paper, which facilitates integrating information from different views by selecting consistent anchors on the data to reconstruct the relationship between instances. An effective graph regularization is also introduced to learn the optimal manifold of the multi-view data. Experimental results on five benchmark datasets show that the proposed method outperforms the state-of-the-art multi-view learning approaches.