核空間中的平分最近點法與模糊支持向量機(jī).pdf

1、華中科技大學(xué)碩士學(xué)位論文核空間中的平分最近點法與模糊支持向量機(jī)姓名：曹淑娟申請學(xué)位級別：碩士專業(yè)：概率論與數(shù)理統(tǒng)計指導(dǎo)教師：劉小茂20061021華中科技大學(xué)碩士學(xué)位論文IIABSTRACTSupptvectmachineisanewlearningmethoddevelopedinrecentyearsbasedonthefoundationsofStatisticalLearningThey(SLT).Itisgainingpopu

2、larityduetomanyattractivefeaturespromisingempiricalperfmanceinthefieldsofsmallsamplestatisticsnonlinearhighdimensionalpatternrecognition.Becauseoftheirexcellentlearningperfmancetheyhavesuccessfulapplicationsinmanyfieldss

3、uchas:facedetectionhwritingdigitrecognitiontextautocategizationetc.ButasanewtechniqueSVMalsohavemanyshtcomingsthatneedtoberesearched.InthispaperwehaveproposedseveraldevelopedSVMalgithmson(Extended)GeometricalBisectionMet

4、hod(GBM)FuzzySupptVectMachine(FSVM).Themainwksareasfollows:FirstlyGeometricanalysisisgivenoncompressedconvexhulldeterminedbytheEGBMinthispaperthenthecrespondingapproximatelylinearseparablesupptvectmachineisproposed.ThisS

5、VMcanchangetheapproximatelylinearseparableproblemtothestrictlinearseparableproblemcanbegeneralizedtosolvethenonlinearseparableproblembychoosingaproperkernelfunctionwhichdeducesthecrespondingnonlinearSVMbasedontheEGBM.Thu

6、sitshowsbettergeneralizationofthemethod.Secondlywedevelopedtwonewclassifiers:thekernelizedgeometricalbisectionmethoditsextendedversion.Thederivationofmethodsisbasedonthesocalled“kerneltrick”GBM(EGBM).Thealgithmscansolven

7、onlinearlyseparableproblemalsoimprovethetrainingaccuracythecomputationrobustnessevenfapproximatelylinearlyseparabledata.Computationalexperimentsshowthattheproposedalgithmsaremecompetitiveeffectivethanthewellknownconventi

8、onalmethods.FinallyanewFuzzySupptVectMachineofDismissingMargin(DMFSVM)basedontheideaofclasscenterisproposedaimingattheoutliersnoisesappearedinthelargequantitysampleswithfuzzymembership.Thenewalgithmhasweededoutsometraini