Exact top-k feature selection via l 2,0 -norm constraint

摘要: In this paper, we propose a novel robust and pragmatic feature selection approach. Unlike those sparse learning based methods which tackle the approximate problem by imposing sparsity regularization in objective function, proposed method only has one l2,1-norm loss term with an explicit l2,0-Norm equality constraint. An efficient algorithm on augmented Lagrangian will be derived to solve above constrained optimization find out stable local solution. Extensive experiments four biological datasets show that although our model is not convex problem, it outperforms counterparts state-of-art evaluated terms of classification accuracy two popular classifiers. What more, since parameter meaning, i.e. number selected, avoids burden tuning parameter, making method.