Variance-Reduced Gradient Estimator for Nonconvex Zeroth-Order Distributed Optimization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Mu, Huaiyi, Tang, Yujie, Song, Jie, Li, Zhongkui
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917392858218496
author Mu, Huaiyi
Tang, Yujie
Song, Jie
Li, Zhongkui
author_facet Mu, Huaiyi
Tang, Yujie
Song, Jie
Li, Zhongkui
contents This paper investigates distributed zeroth-order optimization for smooth nonconvex problems, targeting the trade-off between convergence rate and sampling cost per zeroth-order gradient estimation in current algorithms that use either the $2$-point or $2d$-point gradient estimators. We propose a novel variance-reduced gradient estimator that either randomly renovates a single orthogonal direction of the true gradient or calculates the gradient estimation across all dimensions for variance correction, based on a Bernoulli distribution. Integrating this estimator with gradient tracking mechanism allows us to address the trade-off. We show that the oracle complexity of our proposed algorithm is upper bounded by $O(d/ε)$ for smooth nonconvex functions and by $O(dκ\ln (1/ε))$ for smooth and gradient dominated nonconvex functions, where $d$ denotes the problem dimension and $κ$ is the condition number. Numerical simulations comparing our algorithm with existing methods confirm the effectiveness and efficiency of the proposed gradient estimator.
format Preprint
id arxiv_https___arxiv_org_abs_2409_19567
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Variance-Reduced Gradient Estimator for Nonconvex Zeroth-Order Distributed Optimization
Mu, Huaiyi
Tang, Yujie
Song, Jie
Li, Zhongkui
Optimization and Control
Multiagent Systems
Systems and Control
This paper investigates distributed zeroth-order optimization for smooth nonconvex problems, targeting the trade-off between convergence rate and sampling cost per zeroth-order gradient estimation in current algorithms that use either the $2$-point or $2d$-point gradient estimators. We propose a novel variance-reduced gradient estimator that either randomly renovates a single orthogonal direction of the true gradient or calculates the gradient estimation across all dimensions for variance correction, based on a Bernoulli distribution. Integrating this estimator with gradient tracking mechanism allows us to address the trade-off. We show that the oracle complexity of our proposed algorithm is upper bounded by $O(d/ε)$ for smooth nonconvex functions and by $O(dκ\ln (1/ε))$ for smooth and gradient dominated nonconvex functions, where $d$ denotes the problem dimension and $κ$ is the condition number. Numerical simulations comparing our algorithm with existing methods confirm the effectiveness and efficiency of the proposed gradient estimator.
title Variance-Reduced Gradient Estimator for Nonconvex Zeroth-Order Distributed Optimization
topic Optimization and Control
Multiagent Systems
Systems and Control
url https://arxiv.org/abs/2409.19567