An New Algorithm on Feature Selection with L-Norm PCA

摘要

Traditional dimension reduction methods reduces noises by explicit rank reduction and dimension reduction simultaneously. In this paper, we propose a method by a robust formulation using L2, 1 norm together with rank reduction without dimension reduction using trace norm regularization. We derive an efficient algorithm for the nonlinear optimizations of proposed objective function. Extensive experiments on ten datasets show the effectiveness of the proposed methods.