独立成分分析Independent+Component+Analysis+Algorithms+and+Applications
IndependentComponentAnalysis:
AlgorithmsandApplications
AapoHyvärinenandErkkiOja
NeuralNetworksResearchCentre
HelsinkiUniversityofTechnology
P.O.Box5400,FIN-02015HUT,Finland
NeuralNetworks,13(4-5):411-430,2000
Abstract
Afundamentalprobleminneuralnetworkresearch,aswellasinmanyotherdisciplines,is ndingasuitablerepresentationofmultivariatedata,i.e.randomvectors.Forreasonsofcomputationalandconceptualsimplicity,therepresentationisoftensoughtasalineartransformationoftheoriginaldata.Inotherwords,eachcomponentoftherepresentationisalinearcombinationoftheoriginalvariables.Well-knownlineartransformationmethodsincludeprincipalcomponentanalysis,factoranalysis,andprojectionpursuit.Independentcomponentanalysis(ICA)isarecentlydevelopedmethodinwhichthegoalisto ndalinearrepresentationofnongaussiandatasothatthecomponentsarestatisticallyindependent,orasindependentaspossible.Sucharepresentationseemstocapturetheessentialstructureofthedatainmanyapplications,includingfeatureextractionandsignalseparation.Inthispaper,wepresentthebasictheoryandapplicationsofICA,andourrecentworkonthesubject.
Keywords:Independentcomponentanalysis,projectionpursuit,blindsignalseparation,sourceseparation,factoranalysis,representation
1Motivation
Imaginethatyouareinaroomwheretwopeoplearespeakingsimultaneously.Youhavetwomicrophones,whichyouholdindifferentlocations.Themicrophonesgiveyoutworecordedtimesignals,whichwecoulddenotebyx1(t)andx2(t),withx1andx2theamplitudes,andtthetimeindex.Eachoftheserecordedsignalsisaweightedsumofthespeechsignalsemittedbythetwospeakers,whichwedenotebys1(t)ands2(t).Wecouldexpressthisasalinearequation:
x1(t)=a11s1+a12s2
x2(t)=a21s1+a22s2(1)(2)
wherea11,a12,a21,anda22aresomeparametersthatdependonthedistancesofthemicrophonesfromthespeakers.Itwouldbeveryusefulifyoucouldnowestimatethetwooriginalspeechsignalss1(t)ands2(t),usingonlytherecordedsignalsx1(t)andx2(t).Thisiscalledthecocktail-partyproblem.Forthetimebeing,weomitanytimedelaysorotherextrafactorsfromoursimpli edmixingmodel.
Asanillustration,considerthewaveformsinFig.1andFig.2.Theseare,ofcourse,notrealisticspeechsignals,butsuf ceforthisillustration.TheoriginalspeechsignalscouldlooksomethinglikethoseinFig.1andthemixedsignalscouldlooklikethoseinFig.2.TheproblemistorecoverthedatainFig.1usingonlythedatainFig.2.Actually,ifweknewtheparametersaij,wecouldsolvethelinearequationin(1)byclassicalmethods.Thepointis,however,thatifyoudon’tknowtheaij,theproblemisconsiderablymoredif cult.
Oneapproachtosolvingthisproblemwouldbetousesomeinformationonthestatisticalpropertiesofthesignalssi(t)toestimatetheaii.Actually,andperhapssurprisingly,itturnsoutthatitisenoughtoassumethats1(t)ands2(t),ateachtimeinstantt,arestatisticallyindependent.Thisisnotanunrealisticassumptioninmanycases,
1
独立成分分析 Independent Component Analysis (ICA)齐娟 2007-5-29 主要内容 ?...论文: Independent component analysis: algorithms and applications Survey on ...
A., Oja的英文著作independent component analysis 的中文翻译版.《独立成分分析...applications of ICA will find useful guidelines and theoretical understanding ...
独立成分分析(ICA)_信息与通信_工程科技_专业资料。模式识别独Independent ...论: Independent component analysis: algorithms and applications Survey on ...
独立成分分析法_数学_自然科学_专业资料 暂无评价
立成分分析 (ICA, Independent Component Analysis)是上世纪 90 年代为解决盲... Algorithms and Applications[J].Neural Networks, 2000,13(4-5):411-430. ...
独立成分分析(Independent Component Analysis)_信息与通信_工程科技_专业资料。1...Component Analysis: Algorithms and Applications Applications》(AapoHyv?rinen and...
erps-using-independent-component-analysisica/ 的每个分量都由 的分量线性表示...《Independent Component Analysis: Algorithms and Applications》(Aapo Hyv?rinen ...
更多>>介绍了独立成分分析(ICA)的基本模型及其假设、含混性、非高斯性度量和通用求解过程。讨讨了目前ICA的几个研究方向的发展现状和面临的问题,分析了ICA基本模型...
Independent Component Analysis独立成分分析 PPT_理学_高等教育_教育专区。...… and most applications where Factor Analysis and PCA is currently used. ...
rce-separation-bss-of-erps-using-independent-component-analysis-ica/ (??)...《Independent Component Analysis: Algorithms and Applications》 (Aapo Hyv?rinen...
热门文档