Support Vector Machines with Weighted Regularization

In this paper, we propose a novel regularization criterion for robust classifiers. The criterion can produce many types of regularization terms by selecting an appropriate weighting function. L2 regularization terms, which are used for support vector machines (SVMs), can be produced with this criterion when the norm of patterns is normalized. In this regard, we propose two novel regularization terms based on the new criterion for a variety of applications. Furthermore, we propose new classifiers by applying these regularization terms to conventional SVMs. Finally, we conduct an experiment to demonstrate the advantages of these novel classifiers.