Generalization performance of support vector machines and other pattern classifiers

The aim of this chapter is to summarise results that have been obtained for high con dence generalization error bounds for the Support Vector Machine (SVM) and other pattern classi ers related to the SVM. As a by-product of the analysis we argue that the margin and number of support vectors are both estimators of the degree to which the distribution generating the inputs assists identi cation of the target hyperplane.