A training algorithm for optimal margin classifiers-英文文献.pdf

A Training Algorithm for Optimal Margin Classiers Bernhard E Boser Isab elle M Guyon Vladimir N Vapnik EECS Department ATT Bell Laboratories ATT Bell Laboratories University of California Fremont Street th Floor Crawford Corner Road Berkeley CA San Francisco CA Holmdel NJ bosereecsberkeleyedu isabelleneuralattcom vladneuralattcom Abstract MooGVB VapBHTLSMaclinkthe generalization of a classier to the error on the training A training algorithmthat maximizes the mar examples and the complexity of the classier Meth gin between the training patterns and the de ods such as structural risk minimization Vap vary cision boundary is presented The technique the complexity of the classication function in order to is applicable to a wide variety of classiac optimize the generalization tion functions including Perceptrons polyno In this paper we describe a training algorithm that au mials and Radial Basis Functions The ef tomaticallytunes the capacity of the classication func fective number of parameters is adjusted auto tion by maximizingthe margin between training exam maticallyto matchthe complexityofthe prob ples and class boundary KM optionally after re lem The solution is expressed as a linear com movingsomeatypicalor meaninglessexamplesfromthe binatio