Multi-View Capsule Network

Abstract

Multi-view learning attempts to generate a model with a better performance by exploiting information among multi-view data. Most existing approaches only focus on either consistency or complementarity principle, and learn representations (or features) of the multi-view data. In this paper, to utilize both complementarity and consistency simultaneously, and explore the potential of deep learning in multi-view learning, we propose a novel supervised multi-view learning algorithm, called multi-view capsule network (MVCapsNet), which extracts a feature matrix of all views by a group of encoders, and obtains a classification matrix fusing common and special information of multiple views. Extensive experiments conducted on eight real-world datasets have demonstrated the effectiveness of our proposed method, and show its superiority over several state-of-the-art baseline methods.