网站大量收购闲置独家精品文档,联系QQ:2885784924

人工神经网络ANN.ppt

  1. 1、本文档共10页,可阅读全部内容。
  2. 2、有哪些信誉好的足球投注网站(book118)网站文档一经付费(服务费),不意味着购买了该文档的版权,仅供个人/单位学习、研究之用,不得用于商业用途,未经授权,严禁复制、发行、汇编、翻译或者网络传播等,侵权必究。
  3. 3、本站所有内容均由合作方或网友上传,本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺!文档内容仅供研究参考,付费前请自行鉴别。如您付费,意味着您自己接受本站规则且自行承担风险,本站不退款、不进行额外附加服务;查看《如何避免下载的几个坑》。如果您已付费下载过本站文档,您可以点击 这里二次下载
  4. 4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等,请点击“版权申诉”(推荐),也可以打举报电话:400-050-0827(电话支持时间:9:00-18:30)。
查看更多

*ArtificialNeuralNetworks-I*MainProblemswithANNKnowledgebasenottransparent(blackbox)(Partiallyresolved)Learningsometimesdifficult/slowLimitedstoragecapability*ArtificialNeuralNetworks-I*ANNLearningParadigmsSupervisedlearningClassificationControlFunctionapproximationAssociativememoryUnsupervisedlearningClusteringReinforcementlearningControl*ArtificialNeuralNetworks-I*SupervisedLearningTeacherpresentsANNinput-outputpairs01ANNweightsadjustedaccordingtoerror02Iterativealgorithms(e.g.Deltarule,BPrule)03shotlearning(Hopfield)04Qualityoftrainingexamplesiscritical05*ArtificialNeuralNetworks-I*LinearSeparabilityinPerceptronsPresentedbyMartinHo,EddyLi,EricWongandKittyWong-Copyright?2000*ArtificialNeuralNetworks-I*PresentedbyMartinHo,EddyLi,EricWongandKittyWong-Copyright?2000LearningLinearlySeparableFunctions(1)Whatcanthesefunctionslearn?Badnews: -Therearenotmanylinearlyseparablefunctions.Goodnews: -Thereisaperceptronalgorithmthatwilllearn anylinearlyseparablefunction,givenenough trainingexamples.*ArtificialNeuralNetworks-I*DeltaRulea.k.a.LeastMeanSquaresWidrow-Hoffiterativedeltarule(1960)GradientdescentoftheerrorsurfaceGuaranteedtofindminimumerrorconfigurationinsinglelayerANNsStochasticapproximationofdesiredbehaviourl=learningcoefficientwij=connectionfromneuronxjtoyix=(x1,x2,...,xn)ANNinputy=(y1,y2,...,yn)ANNoutputd=(d1,d2,...,dn)desiredoutput(x,d)trainingexamplee=ANNerrorw11w12w13w14y1y2y3x1x2x3x4*ArtificialNeuralNetworks-I*UnsupervisedLearningANNadaptsweightstoclusterinputdataHebbianlearningConnectionstimulus-responsestrengthened(hebbian)CompetitivelearningalgorithmsKohonenARTInputweightsadjustedtoresemblestimulus*ArtificialNeuralNetworks-I*HebbianLearningHebbpostulate(1948)

文档评论(0)

junjun37473 + 关注
实名认证
内容提供者

该用户很懒,什么也没介绍

1亿VIP精品文档

相关文档