Being Bayesian, Even Just a Bit, Fixes Overconfidence in ReLU Networks

BeingBayesian,EvenJustaBit,FixesOverconfidenceinReLUNetworksAgustinusKristiadi1MatthiasHein1PhilippHennig12Abstract2016).Themostobviouscasesaretestpointsthatlie“faraway”fromthetrainingdata.Manyme...