BetterDepth-WidthTrade-offsforNeuralNetworksthroughthelensofDynamicalSystemsVaggosChatziafratis1SaiGaneshNagarajan2IoannisPanageas2Abstractunderstandthatthenatureofcomputationdonebydeepandshallowne...
Depth-WidthTradeoffsinApproximatingNaturalFunctionswithNeuralNetworksItaySafran1OhadShamir1Abstractcentempiricalevidencesuggeststhatstandardfeedforwarddeepnetworksarehardertooptimizethanshallowerne...