A Single-Loop First-Order Algorithm for Linearly Constrained Bilevel Optimization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Shen, Wei, Zhang, Jiawei, Huang, Minhui, Shen, Cong
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915774939004928
author Shen, Wei
Zhang, Jiawei
Huang, Minhui
Shen, Cong
author_facet Shen, Wei
Zhang, Jiawei
Huang, Minhui
Shen, Cong
contents We study bilevel optimization problems where the lower-level problems are strongly convex and have coupled linear constraints. To overcome the potential non-smoothness of the hyper-objective and the computational challenges associated with the Hessian matrix, we utilize penalty and augmented Lagrangian methods to reformulate the original problem as a single-level one. Especially, we establish a strong theoretical connection between the reformulated function and the original hyper-objective by characterizing the closeness of their values and derivatives. Based on this reformulation, we propose a single-loop, first-order algorithm for linearly constrained bilevel optimization (SFLCB). We provide rigorous analyses of its non-asymptotic convergence rates, showing an improvement over prior double-loop algorithms -- form $O(ε^{-3}\log(ε^{-1}))$ to $O(ε^{-3})$. The experiments corroborate our theoretical findings and demonstrate the practical efficiency of the proposed SFLCB algorithm. Simulation code is provided at https://github.com/ShenGroup/SFLCB.
format Preprint
id arxiv_https___arxiv_org_abs_2510_24710
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A Single-Loop First-Order Algorithm for Linearly Constrained Bilevel Optimization
Shen, Wei
Zhang, Jiawei
Huang, Minhui
Shen, Cong
Optimization and Control
Information Theory
Machine Learning
We study bilevel optimization problems where the lower-level problems are strongly convex and have coupled linear constraints. To overcome the potential non-smoothness of the hyper-objective and the computational challenges associated with the Hessian matrix, we utilize penalty and augmented Lagrangian methods to reformulate the original problem as a single-level one. Especially, we establish a strong theoretical connection between the reformulated function and the original hyper-objective by characterizing the closeness of their values and derivatives. Based on this reformulation, we propose a single-loop, first-order algorithm for linearly constrained bilevel optimization (SFLCB). We provide rigorous analyses of its non-asymptotic convergence rates, showing an improvement over prior double-loop algorithms -- form $O(ε^{-3}\log(ε^{-1}))$ to $O(ε^{-3})$. The experiments corroborate our theoretical findings and demonstrate the practical efficiency of the proposed SFLCB algorithm. Simulation code is provided at https://github.com/ShenGroup/SFLCB.
title A Single-Loop First-Order Algorithm for Linearly Constrained Bilevel Optimization
topic Optimization and Control
Information Theory
Machine Learning
url https://arxiv.org/abs/2510.24710