Supervised Multi-view Feature Selection Via Maximum Margin Criterion Joint Distributed Optimization

Abstract

This paper proposes a novel supervised multi-view feature selection method via maximum margin criterion (MMC) joint distributed optimization. Firstly, the proposed method integrates the common loss and the local loss of views to establish the global loss. Further, the view-based MMC regularizer and sparse regularizer are constructed, which can give the selected features better class separability. Then the proposed method combines the distributed alternating direction method of multipliers (DADMM) to design a new algorithm to realize the block-based computation. Numerical experiments verify the effectiveness of the proposed method.