A Hessian-Aware Stochastic Differential Equation for Modelling SGD

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Li, Xiang, Shen, Zebang, Zhang, Liang, He, Niao
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910981073928192
author Li, Xiang
Shen, Zebang
Zhang, Liang
He, Niao
author_facet Li, Xiang
Shen, Zebang
Zhang, Liang
He, Niao
contents Continuous-time approximation of Stochastic Gradient Descent (SGD) is a crucial tool to study its escaping behaviors from stationary points. However, existing stochastic differential equation (SDE) models fail to fully capture these behaviors, even for simple quadratic objectives. Built on a novel stochastic backward error analysis framework, we derive the Hessian-Aware Stochastic Modified Equation (HA-SME), an SDE that incorporates Hessian information of the objective function into both its drift and diffusion terms. Our analysis shows that HA-SME achieves the order-best approximation error guarantee among existing SDE models in the literature, while significantly reducing the dependence on the smoothness parameter of the objective. Empirical experiments on neural network-based loss functions further validate this improvement. Further, for quadratic objectives, under mild conditions, HA-SME is proved to be the first SDE model that recovers exactly the SGD dynamics in the distributional sense. Consequently, when the local landscape near a stationary point can be approximated by quadratics, HA-SME provides a more precise characterization of the local escaping behaviors of SGD. With the enhanced approximation guarantee, we further conduct an escape time analysis using HA-SME, showcasing how it can be employed to analytically study the escaping behavior of SGD for general function classes.
format Preprint
id arxiv_https___arxiv_org_abs_2405_18373
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle A Hessian-Aware Stochastic Differential Equation for Modelling SGD
Li, Xiang
Shen, Zebang
Zhang, Liang
He, Niao
Machine Learning
Optimization and Control
Continuous-time approximation of Stochastic Gradient Descent (SGD) is a crucial tool to study its escaping behaviors from stationary points. However, existing stochastic differential equation (SDE) models fail to fully capture these behaviors, even for simple quadratic objectives. Built on a novel stochastic backward error analysis framework, we derive the Hessian-Aware Stochastic Modified Equation (HA-SME), an SDE that incorporates Hessian information of the objective function into both its drift and diffusion terms. Our analysis shows that HA-SME achieves the order-best approximation error guarantee among existing SDE models in the literature, while significantly reducing the dependence on the smoothness parameter of the objective. Empirical experiments on neural network-based loss functions further validate this improvement. Further, for quadratic objectives, under mild conditions, HA-SME is proved to be the first SDE model that recovers exactly the SGD dynamics in the distributional sense. Consequently, when the local landscape near a stationary point can be approximated by quadratics, HA-SME provides a more precise characterization of the local escaping behaviors of SGD. With the enhanced approximation guarantee, we further conduct an escape time analysis using HA-SME, showcasing how it can be employed to analytically study the escaping behavior of SGD for general function classes.
title A Hessian-Aware Stochastic Differential Equation for Modelling SGD
topic Machine Learning
Optimization and Control
url https://arxiv.org/abs/2405.18373