Differentially private Bayesian learning on distributed data