Acceleration via Symplectic Discretization of High-Resolution Differential Equations
Optimization and Control
2019-11-05 v2 Machine Learning
Numerical Analysis
Numerical Analysis
Machine Learning
Abstract
We study first-order optimization methods obtained by discretizing ordinary differential equations (ODEs) corresponding to Nesterov's accelerated gradient methods (NAGs) and Polyak's heavy-ball method. We consider three discretization schemes: an explicit Euler scheme, an implicit Euler scheme, and a symplectic scheme. We show that the optimization algorithm generated by applying the symplectic scheme to a high-resolution ODE proposed by Shi et al. [2018] achieves an accelerated rate for minimizing smooth strongly convex functions. On the other hand, the resulting algorithm either fails to achieve acceleration or is impractical when the scheme is implicit, the ODE is low-resolution, or the scheme is explicit.
Keywords
Cite
@article{arxiv.1902.03694,
title = {Acceleration via Symplectic Discretization of High-Resolution Differential Equations},
author = {Bin Shi and Simon S. Du and Weijie J. Su and Michael I. Jordan},
journal= {arXiv preprint arXiv:1902.03694},
year = {2019}
}
Comments
Published in Neurips 2019