DynamicalIsometryandaMeanFieldTheoryofRNNs:GatingEnablesSignalPropagationinRecurrentNeuralNetworksMinminChen1JeffreyPennington2SamuelS.Schoenholz2Abstracttion(Gravesetal.,2013)andrecommendationsyst...
DynamicalIsometryandaMeanFieldTheoryofCNNs:HowtoTrain10,000-LayerVanillaConvolutionalNeuralNetworksLechaoXiao12YasamanBahri12JaschaSohl-Dickstein1SamuelS.Schoenholz1JeffreyPennington1AbstractFigure...
Graph-basedIsometryInvariantRepresentationLearningRenataKhasanova1PascalFrossard1AbstractTIGraNetSTNConvLearningtransformationinvariantrepresenta-Figure1.Illustrativetransformation-invarianthandwri...