Sumio Watanabe, Tokyo Institute of Technology, Japan


Difference between Bayes Cross Validation and WAIC for Conditional Independent Samples


It is well known that Bayes cross validation and WAIC are asymptotically equivalent for an identically and independently distributed sample. However, they are not in general if otherwise. In this talk, We show that WAIC is the better estimator of the generalization loss for conditional independent samples.