Fast Linear Convergence of Randomized BFGS
Optimization and Control
2021-02-05 v4 Numerical Analysis
Numerical Analysis
Abstract
Since the late 1950's when quasi-Newton methods first appeared, they have become one of the most widely used and efficient algorithmic paradigms for unconstrained optimization. Despite their immense practical success, there is little theory that shows why these methods are so efficient. We provide a semi-local rate of convergence for the randomized BFGS method which can be significantly better than that of gradient descent, finally giving theoretical evidence supporting the superior empirical performance of the method.
Cite
@article{arxiv.2002.11337,
title = {Fast Linear Convergence of Randomized BFGS},
author = {Dmitry Kovalev and Robert M. Gower and Peter Richtárik and Alexander Rogozin},
journal= {arXiv preprint arXiv:2002.11337},
year = {2021}
}
Comments
26 pages, 1 algorithm, 3 theorems, 11 lemmas, 9 figures