基于殼向量的支持向量機快速學(xué)習(xí)算法研究.pdf

1、浙江大學(xué)碩士學(xué)位論文基于殼向量的支持向量機快速學(xué)習(xí)算法研究姓名：李東暉申請學(xué)位級別：碩士專業(yè)：控制科學(xué)與工程指導(dǎo)教師：吳鐵軍杜樹新20060515塑堊查堂堡圭堂垡墮蘭AbstractSupportvectormachine(SVM)methodbasedOnstatistieslearningtheoryhasbeenwidelyresearchedandusedforitsgoodgeneralizedabilityandhaslot

2、sofimportantresultsButSVMmethodstillhassomeshortcomings，suchasthetrainingspeedisveryslowforlarge—scaledataetcInthispapersupportvectormachinemethodisanalyzedindetail，andsomefastlearningalgorithmsforsupportvectormachinesba

3、sedonhullvetorareproposedThemainresearchinthispaperCanbeshownasfollows：1)Anewgeometricfastlearningalgorithmforsupportvectormachine(SVM)isproposedAsetofhullvectorswhicharemostlikelytobecomethesupportvectorsareextractedfro

4、mtrainingsamples、Ⅳitllthehelpofgeometricinformationinthesesamples，theobtainedhullvectorsetiSconsideredasthenewtrainingsampleset，whichgreatlyreduces“thetimeconsumedinsolvingsequentialquadraticoptimizationproblemsinSVMtrai

5、ning，speedingupthetrainingprocessSimulationsshowedthealgorithmismuchfasterthanthenormalSVMmethodandtheclassificationaccuracyiSthesame2)OnthebasisofhullvectorSVM，anewgeometricfastincrementallearningalgorithmforsupportvect

6、ormachineisproposedAsetofhullvectorswhicharemostlikelytobecomethesupportvectorareextracted丹omtrainingsamples慚ththehelpofgeometricinformationinthesesamplesIntheincrementallearningprocess，theobtainedhullvectorsetandanewsam

7、plesetareconjoined粥theupdatedtrainingsampleset，whichgreatlyreducesthetimeconsumedinsolvingsequentialquadraticoptimizationproblemsinincrementalSVMtraining，speedingupthetrainingprocessInaddition，comparedwiththeexistingincr

8、ementalSVMlearningalgorithmsinwhichonlysupportvectorsareusedtorepresentoriginalsampleset，theproposedalgorithmimprovestheclassificationprecisionSimulationsshowedtheeffectivenessofthealgorithm3)Thealgorithmsareappliedinthe