半監(jiān)督非平行超平面分類機(jī).pdf

1、碩士研究生學(xué)位論文碩士研究生學(xué)位論文新疆大學(xué)論文題目（中文）：論文題目（中文）：半監(jiān)督非平行超平面分類機(jī)半監(jiān)督非平行超平面分類機(jī)論文題目（英文）論文題目（英文）：nonparallelhyperplanesclassifierfsemisupervisedclassification研究生姓名：名：閆金花閆金花專業(yè)、學(xué)業(yè)、學(xué)科：數(shù)學(xué)科：數(shù)學(xué)研究方向：向：最優(yōu)化最優(yōu)化導(dǎo)師導(dǎo)師姓名姓名職稱：楊志霞職稱：楊志霞教授教授論文答辯日期論文答辯日期

2、2015年5月27日學(xué)位授予日期學(xué)位授予日期2015年月日IIAbstractMachinelearninghasveryimptantpositioninartificialintelligencepatternrecognitionresearch.Supptvectmachine(SVM)isamachinelearningmethodbasedonstatisticallearningtheydevelopedinviewoft

3、hesmallsamplebyVapnik.Thismethodisbecomingmemeextensiveresearchapplicationbecauseithasgoodgeneralizationabilityiseasytooperatefhighdimensionaldata.Traditionalsupervisedclassificationmethodcouldsolvevariouspracticalproble

4、msefficientlybutitneedstohavealargenumberofsamplesinmarkersindertogetenoughtrainingsamples.Inawdthepriceishightheefficiencyislow.Therefesomesemisupervisedsupptvectmachine(SVM)classificationmethodsareproposedaccdingtothea

5、ctualneeds.Thetraditionalclassificationmethodofthesupervisionproblemscansolvevariouspracticalproblemseffectively.Howevertheboundaryhyperpanesofthedecisionfunctionareparallelsothatitspopularizationisrestricted.Soweputfwar

6、dnonparallelhyperplanesclassifierfsemisupervisedclassification.Itputslabeleddatapointsunlabeleddatespointcodinginaconnectiondiagrameachnoderepresentsadatapoint.Ifthereisalotofsimilaritiesbetweenthetwodatapointswithanedge

7、connectingtheircrespondingnode.weshouldfindtherightcategytomakethemwithexistinglabeleddatapotentialgraphstructurefunlabeleddataminimizetheinconsistencytoimprovethepredictingaccuracyofthemodel.Atthesametimetheperfmanceofo

8、urproposedmethodisbetterthanthatofLaplaciansupptvectmachine(SVM)Laplaciantwinsupptvectmachine(TSVM)innumericalexperiments.Furthermeweintroducethefeatureiontechniqueintoourmethodimprovetheaccuracyofthemodel.Keywds：Supptve