A majorization penalty method for SVM with sparse constraint

Support vector machine (SVM) is an important and fundamental technique in learning. Soft-margin SVM models have stronger generalization performance compared with the hard-margin SVM. Most existing works use hinge-loss function which can be regarded as upper bound of 0–1 loss function. However, it cannot explicitly control number misclassified samples. In this paper, we idea soft-margin propose a new model sparse constraint. Our strictly limit samples, expressing constraint By constructing majorization function, penalty method used to solve sparse-constrained optimization problem. We apply Conjugate-Gradient (CG) resulting subproblem. Extensive numerical results demonstrate impressive proposed method.