Summary of “A Bayesian Perspective on generalization and stochastic gradient descent, ICLR 2018”

This paper is a follow-up paper of “UNDERSTANDING DEEP LEARNING REQUIRES RETHINKING GENERALIZATION, ICLR 2017”,

Important/Interesting observations from the paper

“We also demonstrate that, when one holds the learning rate fixed, there is an optimum batch size which maximizes the test set accuracy.”