The promises and pitfalls of Stochastic Gradient Langevin Dynamics

Abstract : Stochastic Gradient Langevin Dynamics (SGLD) has emerged as a key MCMC algorithm for Bayesian learning from large scale datasets. While SGLD with decreasing step sizes converges weakly to the posterior distribution, the algorithm is often used with a constant step size in practice and has demonstrated successes in machine learning tasks. The current practice is to set the step size inversely proportional to N where N is the number of training samples. As N becomes large, we show that the SGLD algorithm has an invariant probability measure which significantly departs from the target posterior and behaves like Stochastic Gradient Descent (SGD). This difference is inherently due to the high variance of the stochastic gradients. Several strategies have been suggested to reduce this effect; among them, SGLD Fixed Point (SGLDFP) uses carefully designed control variates to reduce the variance of the stochastic gradients. We show that SGLDFP gives approximate samples from the posterior distribution, with an accuracy comparable to the Langevin Monte Carlo (LMC) algorithm for a computational cost sublinear in the number of data points. We provide a detailed analysis of the Wasserstein distances between LMC, SGLD, SGLDFP and SGD and explicit expressions of the means and covariance matrices of their invariant distributions. Our findings are supported by limited numerical experiments.
Complete list of metadatas

Cited literature [38 references]  Display  Hide  Download

https://hal.inria.fr/hal-01934291
Contributor : Nicolas Brosse <>
Submitted on : Sunday, November 25, 2018 - 8:04:18 PM
Last modification on : Wednesday, December 4, 2019 - 1:34:07 PM
Long-term archiving on: Tuesday, February 26, 2019 - 12:38:53 PM

File

main.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-01934291, version 1

Citation

Nicolas Brosse, Alain Durmus, Éric Moulines. The promises and pitfalls of Stochastic Gradient Langevin Dynamics. NeurIPS 2018 (Advances in Neural Information Processing Systems 2018), Dec 2018, Montreal, France. ⟨hal-01934291⟩

Share

Metrics

Record views

136

Files downloads

29