Why Cold Posteriors? On the Suboptimal Generalization of Optimal Bayes Estimates
2020-11-23pproximateinference AABI Symposium 2021Unverified0· sign in to hype
Chen Zeno, Itay Golan, Ari Pakman, Daniel Soudry
Unverified — Be the first to reproduce this paper.
ReproduceAbstract
Recent works have shown that the predictive accuracy of Bayesian deep learning models exhibit substantial improvements when the posterior is raised to a 1/T power with T<1. In this work, we explore several possible reasons for this surprising behavior.