欢迎来到天天文库
浏览记录
ID:40352154
大小:1.05 MB
页数:40页
时间:2019-07-31
《Artificial Neural Network》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、ArtificialNeuralNetworksKwok-LeungTsuiSystemsEngineering&EngineeringManagementCityUniversityofHongKong3/6/20121HumanBrain3/6/20122RealNeuralNetworkNeuronNeuralNetwork3/6/20123HumanLearning•Numberofneurons:~1010•Connectionsperneuron:~104to105•Neuronswitchingtime:~0.001second•Scenerecog
2、nitiontime:~0.1second3/6/20124ArtificialNeuralNetworks(ANN)xy11x2MysteryHappensx3HereykxpMcCullochandPitts(1943)–firstmodelingwithANN3/6/20125ANNModels3/6/20126ANNModels3/6/20127TransferFunctions3/6/20128FeedForward(4-3-2ANN)3/6/20129RemarksofANN3/6/201210ModelingofANN3/6/201211Modeli
3、ngofANNLinearOutputExample(4-2-1ANN):3/6/201212ModelingofANN3/6/201213ModelingofANN3/6/201214ModelingofANNComputetrainingerror:3/6/201215ModelingofANNLogisticOutputExample(4-2-1ANN):3/6/201216ModelingofANNComputetrainingerror:3/6/201217ModelingofANNComputetestingerror:3/6/201218Modeli
4、ngofANNComputetestingerror(linearoutput):3/6/201219ModelingofANNComputetestingerror(linearoutput):3/6/201220ModelingofANNComputetestingerror(logisticoutput):3/6/201221ModelingofANN3/6/201222BackpropagationofANNAn(2-2-1ANNExample):3/6/201223BackpropagationofANNAn(2-2-1ANNExample):3/6/2
5、01224BackpropagationofANNAn(2-2-1ANNExample):3/6/201225ScalingMethods3/6/201226ANNModels3/6/201227MathematicalRepresentationY…YYlayer,Knodes1KZZ…Z12MZlayer,MnodesXXXX…1234XXlayer,PnodesPZm(0mmX),m1,...,MForregressiongT()TkkTZ,k1,...,Kkk0ForK-classTek(1)classificationgT()
6、fX()gT(),k1,...,KkKTkkonecanchooseell1v()1(1ve)3/6/201228OverfittinginANNs•Stoptrainingwhenerrorgoesuponvalidationset3/6/201229Applications3/6/201230ANNsinSpeechRecognition[Haung/Lippman1988]3/6/201231ANNsforFaceRecognitionleftstrtrightupTypicalinputimagesHeadpose(1-of-4):90%
7、accuracyFacerecognition(1-of-20):90%accuracy3/6/201232ANNsforFaceRecognitionleftstrtrightup3/6/201233UnmannedVehicleDeanPomerleauCMU3/6/201234UnmannedVehicle3/6/201235SummaryANNpresent…”apowerfulandverygeneralapproachforregressionandclassification,andhasbeenshowntocompetewellwiththebe
8、stlea
此文档下载收益归作者所有