Differentially Private Bayesian Learning on Distributed Data