半監(jiān)督流形學習算法研究和應用.pdf

1、西安電子科技大學碩士學位論文半監(jiān)督流形學習算法研究和應用姓名：李昱申請學位級別：碩士專業(yè)：測試計量技術及儀器指導教師：任獲榮20100101ABSTRACTAsanonlineardimensionalityreductionmethod，manifoldlearningcRndiscovertheintrinsicconstructionofcomplexdataforfurtherprocessingSeveralmanifoldl

2、earningalgorithmshavebeendevelopedandwerewidelyusedinpaRemrecognitionandmachinevisionetcareaMostoftheexistingmanifoldlearningalgorithmsaleunsupervisedmethodwithoutusingpriorinformationIfweCallobtainsomepriorinformationof

3、trainingsamples，wecanusethisinformationtohelpthestudysteptoincreasetheclassificationabilitySOimprovingtheoriginalmethodtosemisupervisedmethodTheobjectiveofthispaperistoresearchaboutthesemisupervisedmanifoldlearningAftera

4、nalyzingexistingmethods，WeproposeasemisupervisedmanifoldlearningalgorithmbasedonLaplacianEigenmaps，namedsemisupervisedLaplacianEigenmaps(ss—LE)SSLEusespriorinformationofveryfewsamplestocalculatemoreaccuratelowdimensional

5、embeddingcoordinatesWealSOanalyzeitssuperiorinaccuracyandcomputationcomplexitycomparedwinlothermethodsCompared、析tllSSLLE，thecomputationtimeofSSLEisdramaticallyreducedandthedropoftheaccuracyiswithinallacceptablerangeBesid

6、es，theaccuracyofourmethodCallachievenearlybestresultofSSLLEonlyneedtosetK(thenumberofneighbors)toarelativesmallvalueWedemonstratedtheusefulnessofouralgorithmbysyntheticandrealworldproblems，especiallyitsefficiencyinobject