RBF神经网络英文课件.pdf

RBF神经网络英文课件.pdf

ID:53006754

大小:28.57 KB

页数:12页

时间:2020-04-11

RBF神经网络英文课件.pdf_第1页
RBF神经网络英文课件.pdf_第2页
RBF神经网络英文课件.pdf_第3页
RBF神经网络英文课件.pdf_第4页
RBF神经网络英文课件.pdf_第5页
资源描述:

《RBF神经网络英文课件.pdf》由会员上传分享,免费在线阅读,更多相关内容在教育资源-天天文库

1、RadialBasisFunctionNetworks:AlgorithmsIntroductiontoNeuralNetworks:Lecture13©JohnA.Bullinaria,20041.TheRBFMapping2.TheRBFNetworkArchitecture3.ComputationalPowerofRBFNetworks4.TraininganRBFNetwork5.UnsupervisedOptimizationoftheBasisFunctions6.FindingtheOutputWeightsTheRadialBasisFunction

2、(RBF)MappingWeareworkingwithinthestandardframeworkoffunctionapproximation.WehaveasetofNdatapointsinamulti-dimensionalspacesuchthateveryDdimensionalinputvectorppppx={x:i=1,...,D}hasacorrespondingKdimensionaltargetoutputt={t:k=1,...,K}.ikThetargetoutputswillgenerallybegeneratedbysomeunder

3、lyingfunctionsg(x)pluskrandomnoise.Thegoalistoapproximatetheg(x)withfunctionsy(x)oftheformkkMyk(x)=åwkjfj(x)j=0WeshallconcentrateonthecaseofGaussianbasisfunctions2æx-möjfj(x)=expçç-2÷÷2sèjøinwhichwehavebasiscentres{mj}andwidths{sj}.Naturally,thewaytoproceedistodevelopaprocessforfindingt

4、heappropriatevaluesforM,{wkj},{mij}and{sj}.L13-2TheRBFNetworkArchitectureWecancasttheRBFMappingintoaformthatresemblesaneuralnetwork:1•••Koutputsykweightswkj1•••j•••Mbasisfunctionsfj(xi,mij,sj)weightsmij1•••DinputsxiThehiddentooutputlayerpartoperateslikeastandardfeed-forwardMLPnetwork,wi

5、ththesumoftheweightedhiddenunitactivationsgivingtheoutputunitactivations.Thehiddenunitactivationsaregivenbythebasisfunctionsfj(x,mj,sj),whichdependonthe“weights”{mij,sj}andinputactivations{xi}inanon-standardmanner.L13-3ComputationalPowerofRBFNetworksIntuitively,wecaneasilyunderstandwhyl

6、inearsuperpositionsoflocalisedbasisfunctionsarecapableofuniversalapproximation.Moreformally:Hartman,Keeler&Kowalski(1990,NeuralComputation,vol.2,pp.210-215)providedaformalproofofthispropertyfornetworkswithGaussianbasisfunctionsinwhichthewidths{sj}aretreatedasadjustableparameters.Park&Sa

7、ndberg(1991,NeuralComputation,vol.3,pp.246-257;and1993,NeuralComputation,vol.5,pp.305-316)showedthatwithonlymildrestrictionsonthebasisfunctions,theuniversalfunctionapproximationpropertystillholds.AswiththecorrespondingproofsforMLPs,theseareexistenceproofswhichrelyontheavailabil

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。