The Catch-Up Phenomenon in Bayesian Inference

Standard Bayesian model selection/averaging sometimes learn too slowly: there exist other learning methods that lead to better predictions based on less data. We give a novel analysis of this "catch-u