9_10_convolutional_rbm

9_10_convolutional_rbm

ID:40704085

大小:2.50 MB

页数:5页

时间:2019-08-06

9_10_convolutional_rbm_第1页
9_10_convolutional_rbm_第2页
9_10_convolutional_rbm_第3页
9_10_convolutional_rbm_第4页
9_10_convolutional_rbm_第5页
资源描述:

《9_10_convolutional_rbm》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、NeuralnetworksComputervision-convolutionalRBM2CONVOLUTIONALRBMConvolutionalDeepBeliefNetworksforScalableUnsupervisedLearningofHierarchicalRepresentationsTopics:ConvolutionalDeepBeliefNetworksforScalableUnsupervisedLearningofHierarchicalRepresentationsConvolutionalDeepBeliefNetworksf

2、orScalableUnsupervisedLearningofHierarchicalRepresentations(NWconvolutionalRBM,NVNH+1);thefilterweightsaresharedLeeetal.2009kacrossallthehiddenunitswithinthegroup.Inaddi-NpkP(poolinglayer)Pα•((HowaboutdesigningconvolutionalunsupervisednetworksNNWWtion,eachhiddengrouphasabias,,NNVV

3、NNHH+1);thefilterweightsareshared+1);thefilterweightsaresharedbkandallvisiblekkacrossallthehiddenunitswithinthegroup.Inaddi-acrossallthehiddenunitswithinthegroup.Inaddi-NNppkkPP(poolinglayer)(poolinglayer)unitsshareasinglebiasc.PPααNkktion,eachhiddengrouphasabiastion,eachhiddengroupha

4、sabiasbbandallvisibleandallvisibleHChi,jH(detectionlayer)‣let’sconsiderthecaseoftheRBMkkunitsshareasinglebiasunitsshareasinglebiasWedefinetheenergyfunctioncc..E(v,h)as:NNHkkkkHCChhi,ji,jHH(detectionlayer)(detectionlayer)‣couldusesameconvolutionalconnectivitybetweeninput(1v)andhiddenl

5、ayer(h)WedefinetheenergyfunctionWedefinetheenergyfunctionP(v,h)=exp(EEE(((vvv,,,hhh)))as:)as:WkZ11PP((vv,,hh)=)=exp(exp(XKEE((vXNv,,Hhh))))XNWkWWkkZZkk‣hijarethehiddenunitsofNE(v,h)=hijWrsvi+r1,j+s1NVWvV(visiblelayer)XXKKXNXNHHXNXNWWthekthfeaturemapk=1i,j=1r,s=1EE((vv,,hh)=)=h

6、hkkWWkkvv‣WkNNNNWWvvXKXNHijijrsrsii++rrXNV11,j,j++ss11rsaretheweightstotheVVVV(visiblelayer)(visiblelayer)Figure1.ConvolutionalRBMwithprobabilisticmax-kk=1=1i,ji,j=1=1r,sr,s=1=1kkthfeaturemapbkhijcvij.(1)pooling.Forsimplicity,onlygroupkofthedetectionlayerXXKKk=1XNXNHHi,j=1XNXN

7、VVi,j=1‣WFigure1.~Figure1.karetheweightswithConvolutionalRBMwithprobabilisticmax-ConvolutionalRBMwithprobabilisticmax-kkandthepooinglayerareshown.ThebasicCRBMcorre-bbkkhhijijccvvijij..(1)(1)flippedrowsandcolumnspooling.Forsimplicity,onlygrouppooling.Forsimplicity,onlygroupspondst

8、oasimplifiedstructur

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。
相关文章
更多
相关标签