An approach for constructing sparse Kernel classifier

This paper presents a new approach for constructing sparse kernel classifier with large margin. Firstly, we propose a kernel function pursuit strategy for selecting a small number of kernel functions which are used for expanding final classifier. And then an added constraint controls the sparseness of the final classifier and an approach is provided to solve the optimization problem with L2 loss function and complexity measure. The experiment results show that sparse kernel classifier can achieved higher efficiency for both training and testing without sacrificing prediction accuracy.