Deep multi-view robust representation learning

Abstract

Multi-view representations are widely existed in practical applications, the quality of latent representation learned from multi-view observations often suffer from noise and outliers in original data. In this work, we propose an auto encoder based deep multi-view robust representation learning (DMRRL) algorithm, which can learn a shared representation from multi-view observations and the algorithm is robust to noise and outliers by using Cauchy estimator as loss function. When the label of the original data is available, the algorithm can gain better representation by adding an auxiliary loss. We validate our methods on the CMU PIE dataset (noise applied) with face recognition task and UCF101 dataset with human motion recognition task, demonstrating that DMRRL is an effective algorithm for practical applications.