返回
2.2 Understanding and Visualizing Convolutional Neural Networks

2.2 Understanding and Visualizing Convolutional Neural Networks

ID：40703532

大小：990.14 KB

页数：9页

时间：2019-08-06

2.2 Understanding and Visualizing Convolutional Neural Networks_第1页
2.2 Understanding and Visualizing Convolutional Neural Networks_第2页
2.2 Understanding and Visualizing Convolutional Neural Networks_第3页
2.2 Understanding and Visualizing Convolutional Neural Networks_第4页
2.2 Understanding and Visualizing Convolutional Neural Networks_第5页
资源描述：

《2.2 Understanding and Visualizing Convolutional Neural Networks》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、CS231nConvolutionalNeuralNetworksforVisualRecognition(thispageiscurrentlyindraftform)VisualizingwhatConvNetslearnSeveralapproachesforunderstandingandvisualizingConvolutionalNetworkshavebeendevelopedintheliterature,partlyasaresponsethecommoncriticismthatthelearn

2、edfeaturesinaNeuralNetworkarenotinterpretable.Inthissectionwebrieflysurveysomeoftheseapproachesandrelatedwork.Visualizingtheactivationsandfirst-layerweightsLayerActivations.Themoststraight-forwardvisualizationtechniqueistoshowtheactivationsofthenetworkduringthe

3、forwardpass.ForReLUnetworks,theactivationsusuallystartoutlookingrelativelyblobbyanddense,butasthetrainingprogressestheactivationsusuallybecomemoresparseandlocalized.Onedangerouspitfallthatcanbeeasilynoticedwiththisvisualizationisthatsomeactivationmapsmaybeallze

4、roformanydifferentinputs,whichcanindicatedeadfilters,andcanbeasymptomofhighlearningrates.Typical-lookingactivationsonthefirstCONVlayer(left),andthe5thCONVlayer(right)ofatrainedAlexNetlookingatapictureofacat.Everyboxshowsanactivationmapcorrespondingtosomefilter.

5、Noticethattheactivationsaresparse(mostvaluesarezero,inthisvisualizationshowninblack)andmostlylocal.Conv/FCFilters.Thesecondcommonstrategyistovisualizetheweights.TheseareusuallymostinterpretableonthefirstCONVlayerwhichislookingdirectlyattherawpixeldata,butitispo

6、ssibletoalsoshowthefilterweightsdeeperinthenetwork.Theweightsareusefultovisualizebecausewell-trainednetworksusuallydisplayniceandsmoothfilterswithoutanynoisypatterns.Noisypatternscanbeanindicatorofanetworkthathasn'tbeentrainedforlongenough,orpossiblyaverylowreg

7、ularizationstrengththatmayhaveledtooverfitting.Typical-lookingfiltersonthefirstCONVlayer(left),andthe2ndCONVlayer(right)ofatrainedAlexNet.Noticethatthefirst-layerweightsareveryniceandsmooth,indicatingnicelyconvergednetwork.Thecolor/grayscalefeaturesareclustered

8、becausetheAlexNetcontainstwoseparatestreamsofprocessing,andanapparentconsequenceofthisarchitectureisthatonestreamdevelopshigh-frequencygrayscalefeaturesandtheotherlow-freque

当前文档最多预览五页，下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页，下载文档查看全文
温馨提示：
1. 部分包含数学公式或PPT动画的文件，查看预览时可能会显示错乱或异常，文件下载后无此问题，请放心下载。
2. 本文档由用户上传，版权归属用户，天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容，确认文档内容符合您的需求后进行下载，若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误，付费完成后未能成功下载的用户请联系客服处理。