BeingBayesian,EvenJustaBit,FixesOverconfidenceinReLUNetworksAgustinusKristiadi1MatthiasHein1PhilippHennig12Abstract2016).Themostobviouscasesaretestpointsthatlie“faraway”fromthetrainingdata.Manyme...
BeingBayesianaboutCategoricalProbabilityTaejongJoo1UijungChung1Min-GwanSeo1AbstractBayesianNNs(BNNs;MacKay,1992)canaddresstheaforementionedissuesofsoftmax.BNNsprovidequan-Neuralnetworksutilizetheso...
BeingRobust(inHighDimensions)CanBePracticalIliasDiakonikolas1GautamKamath2DanielM.Kane3JerryLi2AnkurMoitra2AlistairStewart1Abstractgivencomefromanicedistribution,butthatanadversaryhasthepowertoarbi...