2012-ImageNet Classification with Deep Convolutional Neural Networks

2012-ImageNet Classification with Deep Convolutional Neural Networks

ID：40254455

大小：1.35 MB

頁數：9頁

時間：2019-07-29

2012-ImageNet Classification with Deep Convolutional Neural Networks_第1頁
2012-ImageNet Classification with Deep Convolutional Neural Networks_第2頁
2012-ImageNet Classification with Deep Convolutional Neural Networks_第3頁
2012-ImageNet Classification with Deep Convolutional Neural Networks_第4頁
2012-ImageNet Classification with Deep Convolutional Neural Networks_第5頁
2012-ImageNet Classification with Deep Convolutional Neural Networks_第6頁
2012-ImageNet Classification with Deep Convolutional Neural Networks_第7頁
2012-ImageNet Classification with Deep Convolutional Neural Networks_第8頁
2012-ImageNet Classification with Deep Convolutional Neural Networks_第9頁
資源描述：

《2012-ImageNet Classification with Deep Convolutional Neural Networks》由會員上傳分享，免費在線閱讀，更多相關內容在學術論文-天天文庫。

1、ImageNetClassi?cationwithDeepConvolutionalNeuralNetworksAlexKrizhevskyIlyaSutskeverGeoffreyE.HintonUniversityofTorontoUniversityofTorontoUniversityofTorontokriz@cs.utoronto.cailya@cs.utoronto.cahinton@cs.utoronto.caAbstractWetrainedalarge,deepconvolutionalneuralnetworktocla

2、ssifythe1.2millionhigh-resolutionimagesintheImageNetLSVRC-2010contestintothe1000dif-ferentclasses.Onthetestdata,weachievedtop-1andtop-5errorratesof37.5%and17.0%whichisconsiderablybetterthanthepreviousstate-of-the-art.Theneuralnetwork,whichhas60millionparametersand650,000neu

3、rons,consistsof?veconvolutionallayers,someofwhicharefollowedbymax-poolinglayers,andthreefully-connectedlayerswitha?nal1000-waysoftmax.Tomaketrain-ingfaster,weusednon-saturatingneuronsandaveryef?cientGPUimplemen-tationoftheconvolutionoperation.Toreduceover?ttinginthefully-co

4、nnectedlayersweemployedarecently-developedregularizationmethodcalled“dropout”thatprovedtobeveryeffective.WealsoenteredavariantofthismodelintheILSVRC-2012competitionandachievedawinningtop-5testerrorrateof15.3%,comparedto26.2%achievedbythesecond-bestentry.1IntroductionCurrent

5、approachestoobjectrecognitionmakeessentialuseofmachinelearningmethods.Toim-provetheirperformance,wecancollectlargerdatasets,learnmorepowerfulmodels,andusebet-tertechniquesforpreventingover?tting.Untilrecently,datasetsoflabeledimageswererelativelysmall—ontheorderoftensofthou

6、sandsofimages(e.g.,NORB[16],Caltech-101/256[8,9],andCIFAR-10/100[12]).Simplerecognitiontaskscanbesolvedquitewellwithdatasetsofthissize,especiallyiftheyareaugmentedwithlabel-preservingtransformations.Forexample,thecurrent-besterrorrateontheMNISTdigit-recognitiontask(<0.3%)ap

7、proacheshumanperformance[4].Butobjectsinrealisticsettingsexhibitconsiderablevariability,sotolearntorecognizethemitisnecessarytousemuchlargertrainingsets.Andindeed,theshortcomingsofsmallimagedatasetshavebeenwidelyrecognized(e.g.,Pintoetal.[21]),butithasonlyrecentlybecomeposs

8、ibletocol-lectlabeleddatasetswithmillionsofimages.ThenewlargerdatasetsincludeLabel

當前文檔最多預覽五頁，下載文檔查看全文

此文檔下載收益歸作者所有

當前文檔最多預覽五頁，下載文檔查看全文
溫馨提示：
1. 部分包含數學公式或PPT動畫的文件，查看預覽時可能會顯示錯亂或異常，文件下載后無此問題，請放心下載。
2. 本文檔由用戶上傳，版權歸屬用戶，天天文庫負責整理代發(fā)布。如果您對本文檔版權有爭議請及時聯系客服。
3. 下載前請仔細閱讀文檔內容，確認文檔內容符合您的需求后進行下載，若出現內容與標題不符可向本站投訴處理。
4. 下載文檔時可能由于網絡波動等原因無法下載或下載錯誤，付費完成后未能成功下載的用戶請聯系客服處理。