Entropy bounds on Bayesian learning