Stochastic Gradient Geodesic MCMC Methods

Neural Information Processing Systems 

We propose two stochastic gradient MCMC methods for sampling from Bayesian posterior distributions defined on Riemann manifolds with a known geodesic flow, e.g.