简介:Supportvectormachines(SVMs)havebeenintensivelyappliedinthedomainsofspeechrecognition,textcategorization,andfaultsdetection.However,thepracticalapplicationofSVMsislimitedbythenon-smoothfeatureofobjectivefunction.Toovercomethisproblem,anovelsmoothfunctionbasedonthegeometryofcircletangentisconstructed.Itsmoothesthenon-differentiabletermofunconstrainedSVM,andalsoproposesacircletangentsmoothSVM(CTSSVM).Comparedwithothersmoothapproachingfunctions,itssmoothprecisionhadanobviousimprovement.TheoreticalanalysisprovedtheglobalconvergenceofCTSSVM.NumericalexperimentsandcomparisonsshowedCTSSVMhadbetterclassificationandlearningefficiencythancompetitivebaselines.