{"id":"arxiv/2605.08006","url":"https://arxiv.org/abs/2605.08006","tags":["math.OC","cs.LG","stat.ML"],"score":0,"title":"Penalty-Based First-Order Methods for Bilevel Optimization with Minimax and Constrained Lower-Level Problems","domain":"math.OC","source":"arxiv","authors":["Yiyang Shen","Yutian He","Weiran Wang","Qihang Lin"],"reviews":0,"subject":"math.OC","summary":"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…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-05-08","reactions":{}}