SGL-RFS: Semi-Supervised Graph Learning Robust Feature Selection

Abstract

Feature selection has obtained dramatic attentions in the recent years. In this paper, we propose a semi-supervised graph learning robust feature selection model (SGL-RFS). Our method can merge the procedures of sparse regression and graph construction as a whole to learn an optimal sparse regression matrix for feature selection. To solve our propose method, we also develop an effective alternating optimization algorithm. Experimental results on face and digit databases confirm the effectiveness of our proposed method.