A Novel Nonnegative Subspace Learning Approach for Unsupervised Feature Selection

* School of Computer Engineering, Jinling Institute of Technology Nanjing 211169, China, ([email protected]) Abstract Sparse subspace learning has been proven to be effective in data mining and machine learning. In this paper, we propose a novel scheme which performs robust feature selection with non-negative constraint and sparse subspace learning simultaneously. This work emphasizes joint l2, 1-norm and 1l minimization, where the former characterizes the weight matrix and the latter handles residual matrix to improve robustness. The Inexact Augmented Lagrange Multiplier framework has been adopted to solve our object function efficiently and extensive experimental results on original datasets with and without malicious pollutions have demonstrated the superiority of our new method of feature selection.