ℓ2;1-norm based Regression for Classification

Abstract

We present a novel classification method formulating an objective model by ℓ2;1-norm based regression. The ℓ2;1-norm based loss function is robust to outliers or the large variations within given data, and the ℓ2;1-norm regularization term selects correlated samples across the whole training set with grouped sparsity. This constrained optimization problem can be efficiently solved by an iterative procedure. Several benchmark data sets including facial images and gene expression data are used for evaluating the robustness and effectiveness of the new proposed algorithm, and the results show the competitive performance.