[IJCAI 2011 Ciresan] Flexible, high performance convolutional neural networks for image classification

《[IJCAI 2011 Ciresan] Flexible, high performance convolutional neural networks for image classification》由會(huì)員上傳分享，免費(fèi)在線(xiàn)閱讀，更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫(kù)。

1、ProceedingsoftheTwenty-SecondInternationalJointConferenceonArtificialIntelligenceFlexible,HighPerformanceConvolutionalNeuralNetworksforImageClassi?cationDanC.Cires?an,UeliMeier,JonathanMasci,LucaM.Gambardella,JurgenSchmidhuber¨IDSIA,USIandSUPSIGalleria2,6928Manno-Lu

2、gano,Switzerland{dan,ueli,jonathan,luca,juergen}@idsia.chAbstract(CNNs)[LeCunetal.,1998;Behnke,2003;Simardetal.,2003],whoseweights(?lters)arerandomlyinitializedandWepresentafast,fullyparameterizableGPUim-changedinasupervisedwayusingback-propagation(BP).plementationo

3、fConvolutionalNeuralNetworkDespitethehardwareprogressofthepastdecades,compu-variants.Ourfeatureextractorsareneithercare-tationalspeedisstillalimitingfactorforCNNarchitecturesfullydesignednorpre-wired,butratherlearnedincharacterizedbymanybuildingblockstypicallysetbyt

4、rialasupervisedway.Ourdeephierarchicalarchitec-anderror.Tosystematicallytesttheimpactofvariousarchi-turesachievethebestpublishedresultsonbench-tecturesonclassi?cationperformance,wepresentafastCNNmarksforobjectclassi?cation(NORB,CIFAR10)implementationonGraphicsProces

5、singUnits(GPUs).Previ-andhandwrittendigitrecognition(MNIST),withousGPUimplementationsofCNNs[Chellapillaetal.,2006;errorratesof2.53%,19.51%,0.35%,respectively.UetzandBehnke,2009;Strigletal.,2010]werehard-codedDeepnetstrainedbysimpleback-propagationper-tosatisfyGPUhar

6、dwareconstraintsorusegeneralpurposeformbetterthanmoreshallowones.Learningislibraries,whereasourimplementationis?exibleandfullyon-surprisinglyrapid.NORBiscompletelytrainedline(i.e.,weightupdatesaftereachimage).Anotableexcep-within?veepochs.TesterrorratesonMNISTtionis

7、[Jarrettetal.,2009]whoperformedathoroughanaly-dropto2.42%,0.97%and0.48%after1,3and17sisofthein?uenceofallbuildingblocksofamultistagear-epochs,respectively.chitectureonrecognitionperformance.OurimplementationallowsfortraininglargeCNNswithindaysinsteadofmonths,1Introd

8、uctionsuchthatwecaninvestigatethein?uenceofvariousstructuralThehumanvisualsystemef?cientlyrecognizesandlocal-parametersbyexploringlargepar