An Ideal Local Structure Learning for Unsupervised Feature Selection

In this paper, we propose a novel Ideal Local Structure Learning (LSL) for unsupervised feature selection method, which performs local structure learning and feature selection simultaneously. To obtain more accurate information of data structure, an ideal local structure with block diagonal constraint is introduced. Furthermore, a simple yet effective iterative algorithm is presented to optimize the proposed problem. Experiments on various benchmark datasets demonstrate the superiority of LSL compared with the state-of-the-art algorithms.