Tweeted By @andrewgwils
Despite its popularity in the covariate shift setting, Bayesian model averaging can surprisingly hurt OOD generalization! https://t.co/AqbH4f29FG 1/5 https://t.co/OawZs6AV7n pic.twitter.com/SiPHp8mQrm
— Andrew Gordon Wilson (@andrewgwils) June 23, 2021