




已阅读5页,还剩26页未读, 继续免费阅读
版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领
文档简介
DeepLearning,Deeplearningattractslotsofattention.,Ibelieveyouhaveseenlotsofexcitingresultsbefore.,DeeplearningtrendsatGoogle.Source:SIGMOD/JeffDean,UpsanddownsofDeepLearning,1958:Perceptron(linearmodel)1969:Perceptronhaslimitation1980s:Multi-layerperceptronDonothavesignificantdifferencefromDNNtoday1986:BackpropagationUsuallymorethan3hiddenlayersisnothelpful1989:1hiddenlayeris“goodenough”,whydeep?2006:RBMinitialization(breakthrough)2009:GPU2011:Starttobepopularinspeechrecognition2012:winILSVRCimagecompetition,ThreeStepsforDeepLearning,DeepLearningissosimple,NeuralNetwork,NeuralNetwork,“Neuron”,Differentconnectionleadstodifferentnetworkstructures,NeuralNetwork,Networkparameter:alltheweightsandbiasesinthe“neurons”,FullyConnectFeedforwardNetwork,1,-1,1,-2,1,-1,1,0,4,-2,0.98,0.12,FullyConnectFeedforwardNetwork,1,-2,1,-1,4,-2,0.98,0.12,2,-1,-1,-2,3,-1,4,-1,0.86,0.11,0.62,0.83,1,-1,FullyConnectFeedforwardNetwork,1,-2,1,-1,1,0,0.73,0.5,2,-1,-1,-2,3,-1,4,-1,0.72,0.12,0.51,0.85,0,0,-2,2,00=0.510.85,11=0.620.83,0,0,Thisisafunction.,Inputvector,outputvector,Givennetworkstructure,defineafunctionset,OutputLayer,HiddenLayers,InputLayer,FullyConnectFeedforwardNetwork,Input,Output,y1,y2,yM,neuron,8layers,19layers,22layers,AlexNet(2012),VGG(2014),GoogleNet(2014),16.4%,7.3%,6.7%,/slides/winter1516_lecture8.pdf,Deep=Manyhiddenlayers,AlexNet(2012),VGG(2014),GoogleNet(2014),152layers,3.57%,ResidualNet(2015),Taipei101,101layers,16.4%,7.3%,6.7%,Deep=Manyhiddenlayers,Specialstructure,MatrixOperation,1,-2,1,-1,1,0,4,-2,0.98,0.12,11,1211,+,10,0.980.12,=,1,-1,42,y1,y2,yM,NeuralNetwork,W1,W2,WL,b2,bL,x,a1,a2,y,aL-1,b1,=,y1,y2,yM,NeuralNetwork,W1,W2,WL,b2,bL,x,a1,a2,y,y,b1,W1,x,+,b2,W2,+,bL,WL,x,+,b1,Usingparallelcomputingtechniquestospeedupmatrixoperation,OutputLayer,y1,y2,yM,OutputLayer,HiddenLayers,InputLayer,Featureextractorreplacingfeatureengineering,=Multi-classClassifier,Softmax,ExampleApplication,Input,Output,16x16=256,Ink1Noink0,Eachdimensionrepresentstheconfidenceofadigit.,is1,is2,is0,0.1,0.7,0.2,Theimageis“2”,ExampleApplication,HandwritingDigitRecognition,Machine,“2”,is1,is2,is0,Whatisneededisafunction,Input:256-dimvector,output:10-dimvector,NeuralNetwork,OutputLayer,HiddenLayers,InputLayer,ExampleApplication,Input,Output,“2”,is1,is2,is0,AfunctionsetcontainingthecandidatesforHandwritingDigitRecognition,Youneedtodecidethenetworkstructuretoletagoodfunctioninyourfunctionset.,FAQ,Q:Howmanylayers?Howmanyneuronsforeachlayer?Q:Canthestructurebeautomaticallydetermined?E.g.EvolutionaryArtificialNeuralNetworksQ:Canwedesignthenetworkstructure?,ConvolutionalNeuralNetwork(CNN),ThreeStepsforDeepLearning,DeepLearningissosimple,NeuralNetwork,LossforanExample,y1,y2,y10,CrossEntropy,“1”,1,0,0,target,Softmax,=110,1,2,10,Givenasetofparameters,TotalLoss,NN,NN,NN,1,2,1,NN,3,Foralltrainingdata,=1,FindthenetworkparametersthatminimizetotallossL,TotalLoss:,2,3,FindafunctioninfunctionsetthatminimizestotallossL,ThreeStepsforDeepLearning,DeepLearningissosimple,NeuralNetwork,GradientDescent,1,Compute1,1,0.15,2,Compute2,2,0.05,1,Compute1,1,0.2,0.2,-0.1,0.3,121,=,gradient,GradientDescent,1,Compute1,1,0.15,1,Compute1,0.09,2,Compute2,2,0.05,2,Compute2,0.15,1,Compute1,1,0.2,1,Compute1,0.10,0.2,-0.1,0.3,GradientDescent,Thisisthe“learning”ofmachinesindeeplearning,Evenalphagousingthisapproach.,Ihopeyouarenottoodisappointed:p,Peopleimage,Actually.,Backpropagation,Backpropagation:anefficientwaytocomputeinneuralnetwork,Ref:.tw/tlkagk/courses/MLDS_2015_2/Lecture/DNN%20backprop.ecm.mp4/index.html,ConcludingRemarks,NeuralNetwork,Whatarethebenefitsofdeeparchitecture?,DeeperisBetter?,Seide,Frank,GangLi,andDongYu.ConversationalSpeechTranscriptionUsingContext-DependentDeepNeuralNetworks.Interspeech.2011.,Notsurprised,moreparameters,betterperformance,UniversalityTheorem,Referenceforthereason:,Anycontinuousfunctionf,Canberealizedbyanetworkwithonehiddenlayer,(givenenough
温馨提示
- 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
- 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
- 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
- 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
- 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
- 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
- 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
最新文档
- 物流与仓储解决方案创新创业项目商业计划书
- 基于焊缝点云数据引导的机器人自动焊接技术研究
- 阳江2025年广东海洋大学阳江校区合同制工作人员招聘(第一批)笔试历年参考题库附带答案详解
- 船舶买卖合同范本及法律风险提示
- 小学信息技术课堂教学课件
- 学生作文写作提升技巧汇编
- 工厂节能降耗技术改造方案分析
- 一年级语文经典课文背诵指导
- 幼儿园自然拼读教学设计与测评
- 压力性损伤护理报告标准格式
- 【《离心泵叶轮的水力设计过程案例综述》2200字】
- 胃手术并发症及处理
- 2025至2030 中国热成型钢(PHS)行业现状调查与前景策略研究报告
- 执法监督培训课件
- 股权投资基金培训课件
- 千川投手培训课件
- 2025年中国注塑机熔胶筒螺杆市场调查研究报告
- 职业培训班级管理制度
- 第一章第二节《孟德尔自由组合定律应用9331变形及致死现象》课件-人教版必修二
- DB31/T 1093-2018混凝土砌块(砖)用再生骨料技术要求
- 培训机构教务老师工作计划
评论
0/150
提交评论