This was part of Bayesian Statistics and Statistical Learning

Bias and Variance of Bayes Cross Validation in Singular Learning Theory

Sumio Watanabe, Tokyo Institute of Technology

Monday, December 11, 2023

Abstract: Cross validation is widely employed to check a statistical model and a prior distribution,  however, it is known that cross validation error has an inverse correlation with the generalization error even if a posterior distribution is in a singular state. In this talk, I will discuss bias and variance of Bayes cross validation and show that there is a more accurate method for estimating generalization error than leave-one-out cross validation. Several experimental results are demonstrated.