简介:BasedonKKTcomplementaryconditioninoptimizationtheory,anunconstrainednon-differentialoptimizationmodelforsupportvectormachineisproposed.AnadjustableentropyfunctionmethodisgiventodealwiththeproposedoptimizationproblemandtheNewtonalgorithmisusedtofigureouttheoptimalsolution.Theproposedmethodcanfindanoptimalsolutionwitharelativelysmallparameterp,whichavoidsthenumericaloverflowinthetraditionalentropyfunctionmethods.Itisanewapproachtosolvesupportvectormachine.Thetheoreticalanalysisandexperimentalresultsillustratethefeasibilityandeffciencyoftheproposedalgorithm.
简介:Thispaperintroducesthemethodofsupportvectormachine(SVM)intothefieldofsyntheticearthquakepre-diction,whichisanon-linearandcomplexseismogenicsystem.Asanexample,weapplythismethodtopredictthelargestannualmagnitudefortheNorthChinaarea(30°E-42°E,108°N-125°N)andthecapitalregion(38°E-41.5°E,114°N-120°N)onthebasisofseismicityparametersandobservedprecursorydata.ThecorrespondingpredictionratesfortheNorthChinaareaandthecapitalregionare64.1%and75%,respectively,whichshowsthatthemethodisfeasible.
简介:Supportvectormachines(SVMs)haveshownremarkablesuccessinmanyapplications.However,thenon-smoothfeatureofobjectivefunctionisalimitationinpracticalapplicationofSVMs.Toovercomethisdisadvantage,atwicecontinuouslydifferentiablepiecewise-smoothfunctionisconstructedtosmooththeobjectivefunctionofunconstrainedsupportvectormachine(SVM),anditissuesapiecewise-smoothsupportvectormachine(PWESSVM).Comparingtotheothersmoothapproximationfunctions,thesmoothprecisionhasanobviousimprovement.ThetheoreticalanalysisshowsPWESSVMisgloballyconvergent.Numericalresultsandcomparisonsdemonstratetheclassificationperformanceofouralgorithmisbetterthanothercompetitivebaselines.
简介:MicroRNAs(miRNAs)是一个家庭(2123nt)突然,规章的非编码的RNA处理了从长(70110nt)miRNA先锋(pre-miRNAs)。识别真、假的先锋在miRNAs的计算鉴定起一个重要作用。一些数字特征从先锋序列和他们的第二等的结构被提取了适合一些分类方法;然而,他们可以失去在序列和结构隐藏的一些有用地歧视的信息。在这研究,pre-miRNA序列和他们的第二等的结构直接被用来基于在二个序列之间的加权的Levenshtein距离构造一个指数的核。这个字符串内核然后为检测真、假的pre-miRNAs与支持向量机器(SVM)被相结合。在331上基于训练真、假的人的pre-miRNAs的样品,在SVM的2个关键参数被5褶层选择有不同5褶层分区的十字确认和格子搜索,和5条认识被执行。在16独立人士之中,测试从3人,8动物,2工厂,1个病毒,和2人工地假的人设定pre-miRNAs,我们的方法统计上在11个集合上超过以前的基于SVM的技术包括3人,7动物,和1假人的pre-miRNAs。特别地,有通常在以前的工作被排除的多重环的premiRNAs正确地与92.66%的精确性在这研究被识别。
简介:Supportvectormachines(SVMs)havebeenintensivelyappliedinthedomainsofspeechrecognition,textcategorization,andfaultsdetection.However,thepracticalapplicationofSVMsislimitedbythenon-smoothfeatureofobjectivefunction.Toovercomethisproblem,anovelsmoothfunctionbasedonthegeometryofcircletangentisconstructed.Itsmoothesthenon-differentiabletermofunconstrainedSVM,andalsoproposesacircletangentsmoothSVM(CTSSVM).Comparedwithothersmoothapproachingfunctions,itssmoothprecisionhadanobviousimprovement.TheoreticalanalysisprovedtheglobalconvergenceofCTSSVM.NumericalexperimentsandcomparisonsshowedCTSSVMhadbetterclassificationandlearningefficiencythancompetitivebaselines.
简介:Licenseplaterecognition(LPR)isanimageprocessingtechnologythatisusedtoidentifyvehiclesbytheirlicenseplates.ThispaperpresentsalicenseplaterecognitionalgorithmforSaudicarplatesbasedonthesupportvectormachine(SVM)algorithm.ThenewalgorithmisefficientinrecognizingthevehiclesfromtheArabicpartoftheplate.Theperformanceofthesystemhasbeeninvestigatedandanalyzed.Therecognitionaccuracyofthealgorithmisabout93.3%.
简介:Inthispaper,wepresentanovelSupportVectorMachineactivelearningalgorithmforeffective3Dmodelretrievalusingtheconceptofrelevancefeedback.Theproposedmethodlearnsfromthemostinformativeobjectswhicharemarkedbytheuser,andthencreatesaboundaryseparatingtherelevantmodelsfromirrelevantones.Whatitneedsisonlyasmallnumberof3Dmodelslabelledbytheuser.Itcangrasptheuser'ssemanticknowledgerapidlyandaccurately.Experimentalresultsshowedthattheproposedalgorithmsignificantlyimprovestheretrievaleffectiveness.Comparedwithfourstate-of-the-artqueryrefinementschemesfor3Dmodelretrieval,itprovidessuperiorretrievalperformanceafternomorethantworoundsofrelevance
简介:Inthisstudy,wepresentaconstructivealgorithmfortrainingcooperativesupportvectormachineensembles(CSVMEs).CSVMEcombinesensemblearchitecturedesignwithcooperativetrainingforindividualSVMsinensembles.Unlikemostpreviousstudiesontrainingensembles,CSVMEputsemphasisonbothaccuracyandcollaborationamongindividualSVMsinanensemble.AgroupofSVMsselectedonthebasisofrecursiveclassifiereliminationisusedinCSVME,andthenumberoftheindividualSVMsselectedtoconstructCSVMEisdeterminedby10-foldcross-validation.ThiskindofSVMEhasbeentestedontwoovariancancerdatasetspreviouslyobtainedbyproteomicmassspectrometry.BycombiningseveralindividualSVMs,theproposedmethodachievesbetterperformancethantheSVMEofallbaseSVMs.
简介:
简介:Supportvectormachines(SVM)receivedwideattentionforitsexcellentabilitytolearn,ithasbeenappliedinmanyfields.AreviewoftheapplicationofSVMinwelddefectdetectionandrecognitionofX-rayimageisbeenpresented.WewillshowsomecommonlyusedmethodsofwelddefectdetectionandrecognitionusingSVM,andtheadvantagesanddisadvantagesofeachmethodwillbediscussed.SVMappearstobepromisinginwelddefectdetectionandrecognition,butfutureresearchisneededbeforeitfullymatureinthisfiled.
简介:Afaultdiagnosismodelisproposedbasedonfuzzysupportvectormachine(FSVM)combinedwithfuzzyclustering(FC).Consideringtherelationshipbetweenthesamplepointandnon-selfclass,FCalgorithmisappliedtogeneratefuzzymemberships.Inthealgorithm,sampleweightsbasedonadistributiondensityfunctionofdatapointandgeneticalgorithm(GA)areintroducedtoenhancetheperformanceofFC.Thenamulti-classFSVMwithradialbasisfunctionkernelisestablishedaccordingtodirectedacyclicgraphalgorithm,thepenaltyfactorandkernelparameterofwhichareoptimizedbyGA.Finally,themodelisexecutedformulti-classfaultdiagnosisofrollingelementbearings.Theresultsshowthatthepresentedmodelachieveshighperformancesbothinidentifyingfaulttypesandfaultdegrees.TheperformancecomparisonsofthepresentedmodelwithSVManddistance-basedFSVMfornoisycasedemonstratethecapacityofdealingwithnoiseandgeneralization.
简介:基于小浪包转变(WPT),基因算法(GA),神经网络(BPNN)和支持向量用机器制造的背繁殖(SVM),柴油机引擎阀门清理的一个差错诊断方法被介绍。与力量光谱密度分析,与运用条件的引擎有关的典型频率能从颤动信号被提取。小浪系数和根的最大的单个值(BSV)意味着在典型频率亚乐队的颤动的平方(RMS)值在颤动信号的第三水平分解的目的被提取,并且他们被用作BPNN或SVM的输入向量。为了避免,在本地最小被套住,GA被采用。正常和处于不同阀门清理条件测量的差错颤动信号被分析。BPNN,GA背繁殖神经网络(GA-BPNN),SVM和GA-SVM被用于为不同特征的抽取训练并且测试,并且分类精确性和训练时间与相比决定最佳差错分类器和特征选择。试验性的结果证明建议特征和分类算法给100%的分类精确性。
简介:Thispaperproposesanewmethodtopredictthecoronaonsetvoltageforarodplaneairgap,basedonthesupportvectormachine(SVM).BecausetheSVMisnotlimitedbythesize,dimensionandnonlinearityofthesamples,thismethodcanrealizeaccuratepredictionwithfewtrainingdata.Onlyelectricfieldfeaturesarechosenastheinput;nogeometricparameterisincluded.Therefore,theexperimentdataofonekindofelectrodecanbeusedtopredictthecoronaonsetvoltagesofotherelectrodeswithdifferentsizes.Withtheexperimentaldataobtainedbyozonedetectiontechnology,andexperimentaldataprovidedbythereference,theefficiencyoftheproposedmethodisvalidated.Accuratepredictedresultswithanaveragerelativelessthan3%areobtainedwithonly6experimentaldata.