Error estimates between SGD with momentum and underdamped Langevin diffusion

Guillin, Arnaud, Wang, Yu, Xu, Lihu, Yang, Haoran

arXiv.org Machine Learning 

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.