Logo image
Penalty-Based First-Order Methods for Bilevel Optimization with Minimax and Constrained Lower-Level Problems
Preprint   Open access

Penalty-Based First-Order Methods for Bilevel Optimization with Minimax and Constrained Lower-Level Problems

Yiyang Shen, Yutian He, Weiran Wang and Qihang Lin
ArXiv.org
Cornell University
05/08/2026
DOI: 10.48550/arxiv.2605.08006
url
https://doi.org/10.48550/arxiv.2605.08006View
Preprint (Author's original) This preprint has not been evaluated by subject experts through peer review. Preprints may undergo extensive changes and/or become peer-reviewed journal articles. Open Access

Abstract

We study a class of bilevel optimization problems in which both the upper- and lower-level problems have minimax structures. This setting captures a broad range of emerging applications. Despite the extensive literature on bilevel optimization and minimax optimization separately, existing methods mainly focus on bilevel optimization with lower-level minimization problems, often under strong convexity assumptions, and are not directly applicable to the minimax lower-level setting considered here. To address this gap, we develop penalty-based first-order methods for bilevel minimax optimization without requiring strong convexity of the lower-level problem. In the deterministic setting, we establish that the proposed method finds anε -KKT point withÕ(ε⁻⁴)oracle complexity. We further show that bilevel problems with convex constrained lower-level minimization can be reformulated as special cases of our framework via Lagrangian duality, leading to anÕ(ε⁻⁴)complexity bound that improves upon the existingÕ(ε⁻⁷)result. Finally, we extend our approach to the stochastic setting, where only stochastic gradient oracles are available, and prove that the proposed stochastic method finds a nearlyε -KKT point withÕ(ε⁻⁹)oracle complexity.
Computer Science - Learning Mathematics - Optimization and Control Statistics - Machine Learning

Details

Metrics

1 Record Views
Logo image