Saved in:
Bibliographic Details
Main Authors: Niu, Youcheng, Xu, Jinming, Sun, Ying, Chai, Li, Chen, Jiming
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2412.11218
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910033946607616
author Niu, Youcheng
Xu, Jinming
Sun, Ying
Chai, Li
Chen, Jiming
author_facet Niu, Youcheng
Xu, Jinming
Sun, Ying
Chai, Li
Chen, Jiming
contents This paper considers a class of distributed bilevel optimization (DBO) problems with a coupled inner-level subproblem. Existing approaches typically rely on hypergradient estimations involving computationally expensive Hessian evaluation. To address this, we approximate the DBO problem as a minimax problem by properly designing a penalty term that enforces both the constraint imposed by the inner-level subproblem and the consensus among the decision variables of agents. Moreover, we propose a loopless distributed algorithm, AHEAD, that employs multiple-timescale updates to solve the approximate problem asymptotically without requiring Hessian computation. Theoretically, we establish sharp convergence rates for nonconvex-strongly-convex settings and for distributed minimax problems as special cases. Our analysis reveals a clear dependence of convergence performance on node heterogeneity, penalty parameters, and network connectivity, with a weaker assumption on heterogeneity that only requires bounded gradients at the optimum. Numerical experiments corroborate our theoretical results.
format Preprint
id arxiv_https___arxiv_org_abs_2412_11218
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Hessian-Free Distributed Bilevel Optimization via Penalization with Time-Scale Separation
Niu, Youcheng
Xu, Jinming
Sun, Ying
Chai, Li
Chen, Jiming
Optimization and Control
This paper considers a class of distributed bilevel optimization (DBO) problems with a coupled inner-level subproblem. Existing approaches typically rely on hypergradient estimations involving computationally expensive Hessian evaluation. To address this, we approximate the DBO problem as a minimax problem by properly designing a penalty term that enforces both the constraint imposed by the inner-level subproblem and the consensus among the decision variables of agents. Moreover, we propose a loopless distributed algorithm, AHEAD, that employs multiple-timescale updates to solve the approximate problem asymptotically without requiring Hessian computation. Theoretically, we establish sharp convergence rates for nonconvex-strongly-convex settings and for distributed minimax problems as special cases. Our analysis reveals a clear dependence of convergence performance on node heterogeneity, penalty parameters, and network connectivity, with a weaker assumption on heterogeneity that only requires bounded gradients at the optimum. Numerical experiments corroborate our theoretical results.
title Hessian-Free Distributed Bilevel Optimization via Penalization with Time-Scale Separation
topic Optimization and Control
url https://arxiv.org/abs/2412.11218