Stochastic Gradient Markov Chain Monte Carlo

Markov chain Monte Carlo (MCMC) algorithms are generally regarded as the gold standard technique for Bayesian inference. They theoretically well-understood and conceptually simple to apply in practice. The drawback of MCMC is that performing exact inference requires all data be processed at each iteration algorithm. For large datasets, computational cost can prohibitive, which has led recent developments scalable have a significantly lower than MCMC. In this article, we focus on particular class algorithms, stochastic gradient (SGMCMC) utilizes subsampling techniques reduce per-iteration We provide an introduction some popular SGMCMC review supporting theoretical results, well comparing efficiency against benchmark examples. R code available online https://github.com/chris-nemeth/sgmcmc-review-paper.