Special Properties of Gradient Descent with Large Learning Rates

(2023) Special Properties of Gradient Descent with Large Learning Rates.

Text 2637_special_properties_of_gradient.pdf Download (4MB)

Abstract

When training neural networks, it has been widely observed that a large step size is essential in stochastic gradient descent (SGD) for obtaining superior models. However, the effect of large step sizes on the success of SGD is not well understood theoretically. Several previous works have attributed this success to the stochastic noise present in SGD. However, we show through a novel set of experiments that the stochastic noise is not sufficient to explain good non-convex training, and that instead the effect of a large learning rate itself is essential for obtaining best performance.We demonstrate the same effects also in the noise-less case, i.e. for full-batch GD. We formally prove that GD with large step size ---on certain non-convex function classes --- follows a different trajectory than GD with a small step size, which can lead to convergence to a global minimum instead of a local one. Our settings provide a framework for future analysis which allows comparing algorithms based on behaviors that can not be observed in the traditional settings.

Item Type:	Conference or Workshop Item (A Paper) (Paper)
Divisions:	Sebastian Stich (SS)
Conference:	ICML International Conference on Machine Learning
Depositing User:	Sebastian Stich
Date Deposited:	30 Jun 2023 11:09
Last Modified:	30 Jun 2023 11:09
Primary Research Area:	NRA1: Trustworthy Information Processing
URI:	https://publications.cispa.saarland/id/eprint/3977

Actions

Actions (login required)

View Item