摘要
Supportvectormachines(SVMs)havebeenintensivelyappliedinthedomainsofspeechrecognition,textcategorization,andfaultsdetection.However,thepracticalapplicationofSVMsislimitedbythenon-smoothfeatureofobjectivefunction.Toovercomethisproblem,anovelsmoothfunctionbasedonthegeometryofcircletangentisconstructed.Itsmoothesthenon-differentiabletermofunconstrainedSVM,andalsoproposesacircletangentsmoothSVM(CTSSVM).Comparedwithothersmoothapproachingfunctions,itssmoothprecisionhadanobviousimprovement.TheoreticalanalysisprovedtheglobalconvergenceofCTSSVM.NumericalexperimentsandcomparisonsshowedCTSSVMhadbetterclassificationandlearningefficiencythancompetitivebaselines.
出版日期
2016年01月11日(中国Betway体育网页登陆平台首次上网日期,不代表论文的发表时间)