English

Error estimates between SGD with momentum and underdamped Langevin diffusion

Machine Learning 2024-10-24 v1 Machine Learning Probability

Abstract

Stochastic gradient descent with momentum is a popular variant of stochastic gradient descent, which has recently been reported to have a close relationship with the underdamped Langevin diffusion. In this paper, we establish a quantitative error estimate between them in the 1-Wasserstein and total variation distances.

Cite

@article{arxiv.2410.17297,
  title  = {Error estimates between SGD with momentum and underdamped Langevin diffusion},
  author = {Arnaud Guillin and Yu Wang and Lihu Xu and Haoran Yang},
  journal= {arXiv preprint arXiv:2410.17297},
  year   = {2024}
}
R2 v1 2026-06-28T19:31:59.661Z