分层随机采样支持向量机.pdf

《分层随机采样支持向量机.pdf》由会员上传分享，免费在线阅读，更多相关内容在工程资料-天天文库。

1、©©©ÅÅÅæææ

2、

3、

4、±±±þþþÅÅÅStratifiedRandomSamplingforSupportVectorMachinesÆ;µOÅÆEâïÄ)µÍµ¬¥ÇU9ÆOÅÆEâÆocÕM5(²<(²¤¥ÆØ©´<3e?1ïÄóÚïĤJ§Ø©¥AO±I5Ú?§Ø©¥Ø¹Ù¦<®²uL½>LïĤJ§Ø¹¼UUU999ÆÆƽ٦ÅƽyÖ¦^Lá"·ÓóÓéïĤ?Ûzþ®3Ø©¥²(`²¿L«¿"ÆØ©ö¶µiFϵ

5、'uÆØ©¦^Ç`²ÆØ©ö)UUU999ÆÆÆk'3!¦^ÆØ©5½"AÇUUU999ÆÆƱòÆØ©ܽܩSN?k'êâ¥?1u¢§¿æ^K

6、±þůÅCq{§{üÅæ{S

7、ASVM©Åæ{SS-SVM"òêâ8Åy©ü°§3z°f8þÔöÑ5

8、±þkU¤êâ8

9、±þ"Ïd§rlüf8þÔöÑ5

10、±þ

11、Üå5êâ8æ§{SASVM"TæL§±48¢y§Äud{SS-SVM"ÌSN)µ1.JÑ{üÅæ

12、±þÅ{SASVM§©ÛÙCqØ.!OE,5§`²TÅæg15"2.ÄuSASVM{§JÑ©Åæ

13、±þÅSS-SVM§¿©ÛÙØ.OE,5"3.OIOêâ8þ¢§¿®kÅCqSVM{ÚIOSVM{é'§¢ySASVMSS-SV

14、M{15k5"'cµÅ{¶Åæ¶

15、±þŶ©aAbstractAnyefficientlearningalgorithmshouldatleasttakeabrieflookateachexam-ple.Butalltheexamplesshouldn'tbegivenequalattention.Thispaperproposestwonewalgorithmsfortrainingsupportvectormachinesbasedonrandomizedsam-plingidea,oneisasim

16、plerandomizedsamplingalgorithm(SASVM)andtheotherisastratifiedrandomizedsamplingalgorithm(SS-SVM).Whentheoriginaldatasetisrandomlydividedintotwoequalparts,thesupportvectorsfromtrainingonbothsubsetareprobablymoretobesupportvectorsoftheoriginaldataset.Therefore,

17、wegetanewalgorithmSASVMbycombiningthetwosubsets'ssupportvectorsasasamplesetoftheoriginaldataset.Whiletheaboveprocesscanberealizedrecursively,sowecangetanotheralgorithmSS-SVM.Ourmainworkis:1.WedevelopasimplerandomizedsamplingSVMalgorithmSASVMandanal-ysisitsap

18、proximateerrorboundandcomputingcomplexity,whichtestifythefeasibilityofthissamplingidea.2.BasedonSASVMalgorithm,wedevelopastratifiedrandomizedsamplingSVMalgorithmSS-SVMandanalysisitserrorboundandcomplexity.3.ExperimentsdoneonstandarddatasetsshowthatSASVMandSS-

19、SVMarebothfeasibleandeffectivewhencomparingtheresultswithstandardSVMandotherrandomizedSVMalgorithms.Keywords:Randomizedalgorithm;RandomizedSampling;SupportVectorMa-chines;BinaryClassification8¹1Ù