In this paper, we consider a decision-maker who tries to learn the distribution of outcomes from previously observed cases. For each observed database of cases the decision-maker predicts a set of priors expressing his beliefs about the underlying probability distribution. We impose a version of...