We establish learning rates to the Bayes risk for support vector machines (SVMs) with hinge loss. In particular, for SVMs with Gaussian RBF kernels we propose a geometric condition for distributions which can be used to determine approximation properties of these kernels. Finally, we compare our methods with a recent paper of G. Blanchard et al..
I. Steinwart and C. Scovel, Fast Rates to Bayes for Kernel Methods. In Neural Information Processing Systems 17, pp. 1345-1352, (2005). Los Alamos National Laboratory Technical Report LA-UR-04-3767. [ Abstract | PDF (92 KB) ]
