简介:这份报纸建议一些整齐条件,它在伪可能性的导致存在,强壮的一致性和最大的伪可能性的评估者(MQLE)的asymptotic规度有随机的regressors的非线性的模型(QLNM)。有随机的regressors的概括线性模型(GLM)的asymptotic结果与随机的regressors被概括到QLNM。关键词Asymptotic规度-一致性-最大的伪可能性的评估者-伪可能性的有随机的regressors的非线性的模型2000苏布杰克特先生分类62F12-62J02由中国的国家自然科学基础支持了(号码10761011,10671139,10901135),云南省(号码2008CD081)的自然科学基础和为云南大学的中间、年轻的优秀教师的特殊基础。
简介:Inthispaper,theso-calledlocallikelihoodmethodissuggestedforsolvingthechangepointproblemswhenthedataaredistributedasmultivariatenormal.Thedetectionproceduresproposednotonlyprovidestronglyeonsistentestimatesforthenumberandlocationsofthechangepoints,butalsosimplifysignificantlythecomputation.
简介:Empiricallikelihood(EL)combinedwithestimatingequations(EE)providesamodernsemi-parametricalternativetoclassicalestimationtechniquessuchasmaximumlikelihoodestimation(MLE).ThispapernotonlyusesclosedformofconditionalexpectationandconditionalvarianceofLogisticequationwithrandomperturbationtoperformmaximumempiricallikelihoodestimation(MELE)forthemodelparameters,butalsoproposesanempiricallikelihoodratiostatistic(ELRS)forhypothesesconcerningtheinterestingparameter.MonteCarlosimulationresultsshowthatMELEandELRSprovidecompetitiveperformancetoparametricalternatives.
简介:Inthispaper,twokindsofKullback-Leiblercriteriawithappropriateconstraintsareproposedtoconstructempiricallikelihoodconfidenceintervalsforthemeanofrightcensoreddata.ItisshownthatoneofthecriteriaisequivalenttoAdimari's(1997)procedure,andtheothersharesthesameasymptoticbehavior.
简介:线性假设是最大的可能性的主要劣势线性re-gressionMLLR。这份报纸使用多项式回归方法为改编建模并且为柔韧的语音识别用最大的可能性的多项式回归MLPR建立一个非线性的模型改编算法。在这个算法,在在每条Mel隧道训练并且测试Gaussian工具之间的非线性的关系被一套多项式接近,多项式系数用期望在测试envi-ronment从改编数据被估计--最大化他们算法和最大的可能性的ML标准。试验性的结果证明秒顺序多项式能接近实际非线性的函数更好并且在噪音赔偿和扬声器改编,MLPR的词错误率是比MLLR的那些显著地低的。建议MLPR算法克服好的线性假设和罐头减少的限制噪音,说话者和另外的因素的影响同时。它对说话者和噪音的联合改编特别合适。
简介:Inthispaperweconsidersomerelatednegativehypergeometricdistributionsarisingfromtheproblemofsamplingwithoutreplacementfromanurncontainingballsofdifferentcoloursandindifferentproportionsbutstoppingonlyaftersomespecificnumberofballsofdifferentcolourshavebeenobtained.Withtheaidofsomesimplerecurrencerelationsandidentitiesweobtaininthecaseoftwocoloursthemomentsforthemaximumnegativehypergeometricdistribution,theminimumnegativehypergeometricdistribution,thelikelihoodrationegativehypergeometricdistributionandconsequentlythelikelihoodproportionalnegativehypergeometricdistributiuon.TotheextentthatthesamplingschemeisapplicabletomodellingdataasillustratedwithabiologicalexampleandinfactmanysituationsofestimatingBernoulliparametersforbinarytraitswithinafinitepopulation,theseareimportantfirst-stepresults.
简介:AspecializedHungarianalgorithmwasdevelopedhereforthemaximumlikelihooddataassociationproblemwithtwoimplementationversionsduetopresenceoffalsealarmsandmisseddetections.Themaximumlikelihooddataassociationproblemisformulatedasabipartiteweightedmatchingproblem.Itsdualityandtheoptimalityconditionsaregiven.TheHungarianalgorithmwithitscomputationalsteps,datastructureandcomputationalcomplexityispresented.Thetwoimplementationversions,Hungarianforest(HF)algorithmandHungariantree(HT)algorithm,andtheircombinationwiththenaveauctioninitializationarediscussed.ThecomputationalresultsshowthatHTalgorithmisslightlyfasterthanHFalgorithmandtheyarebothsuperiortotheclassicMunkresalgorithm.
简介:1.IntroductionThemethodoflikelihoodintroducedbyFisheriscertainlyoneofthemostcommonlyusedtechniquesforparametricmodels.Recentlythelikelihoodhasalsobeenshowntobeveryusefulinnonparametriccontexts.O...II--31hasintroducedempiricallikelihoodratiostatistics...
简介:TheperformanceofthetraditionalVoiceActivityDetection(VAD)algorithmsdeclinessharplyinlowerSignal-to-NoiseRatio(SNR)environments.Inthispaper,afeatureweightinglikeli-hoodmethodisproposedfornoise-robustVAD.Thecontributionofdynamicfeaturestolikelihoodscorecanbeincreasedviathemethod,whichimprovesconsequentlythenoiserobustnessofVAD.Divergencebaseddimensionreductionmethodisproposedforsavingcomputation,whichreducesthesefeaturedimensionswithsmallerdivergencevalueatthecostofdegradingtheperformancealittle.ExperimentalresultsonAuroraIIdatabaseshowthatthedetectionperformanceinnoiseenvironmentscanremarkablybeimprovedbytheproposedmethodwhenthemodeltrainedincleandataisusedtodetectspeechendpoints.Usingweightinglikelihoodonthedimension-reducedfeaturesobtainscom-parable,evenbetter,performancecomparedtooriginalfull-dimensionalfeature.
简介:
简介:Inthispaperweuseprofileempiricallikelihoodtoconstructconfidenceregionsforregressioncoefficientsinpartiallylinearmodelwithlongitudinaldata.Themaincontributionisthatthewithin-subjectcorrelationisconsideredtoimproveestimationefficiency.Wesupposeasemi-parametricstructureforthecovariancesofobservationerrorsineachsubjectandemployboththefirstorderandthesecondordermomentconditionsoftheobservationerrorstoconstructtheestimatingequations.Althoughtherearenonparametricestimators,theempiricallog-likelihoodratiostatisticstilltendstoastandardχ2pvariableindistributionafterthenuisanceparametersareprofiledaway.Adatasimulationisalsoconducted.
简介:Inthispaperanewtext-independentspeakerverificationmethodGSMSVisproposedbasedonlikelihoodscorenormalization.Inthisnovelmethodaglobalspeakermodelisestablishedtorepresenttheuniversalfeaturesofspeechandnormalizethelikelihoodscore.Statisticalanalysisdemonstratesthatthisnormalizationmethodcanremovecommonfactorsofspeechandbringthedifferencesbetweenspeakersintoprominence.Asaresulttheequalerrorrateisdecreasedsignificantly,verificationprocedureisacceleratedandsystemadaptabilitytospeakingspeedisimproved.