Exclusivity and consistency induced NMF for multi-view representation learning

Many unsupervised multi-view representation learning (MRL) techniques have been devised as multi-view data becomes more common in real-world applications. However, most existing MRL approaches solely concentrate on the diversity or the consistency between different views. In this paper, we propose a Non-negative Matrix Factorization based MRL framework, seeking to jointly consider two components. To be specific, the exclusivity term is designed to exploit the intra-view diverse information, while the consistency term is employed to ensure the common representation among multiple views. Meanwhile, the local manifold is added to preserve the local geometric structure of the data. Finally, we present a multiplicative-based alternating optimization algorithm to solve the problem, along with proofs of convergence. In addition, the experiments conducted on five multi-view datasets demonstrate the superiority of the proposed approach against the state-of-the-art MRL methods.