Multi-view Similarity Learning of Manifold Data

In recent years, multi-view learning methods have developed rapidly where graph-based approaches have achieved good performance. Usually, these learning methods construct information graph for each view or fuse different views into one graph. In this paper, a novel multi-view learning model that learns one similarity matrix for all views named Multi-view Similarity Learning (MSL) is proposed, where adaptive weights are learned for each view. The multi-view similarity learning method is further extended to kernel space. Experiments of classification, clustering and semi-supervised classification on different real-world datasets show the effectiveness of the proposed method.