English

On the Stability and Generalization of First-order Bilevel Minimax Optimization

Machine Learning 2026-04-23 v1 Artificial Intelligence Machine Learning

Abstract

Bilevel optimization and bilevel minimax optimization have recently emerged as unifying frameworks for a range of machine-learning tasks, including hyperparameter optimization and reinforcement learning. The existing literature focuses on empirical efficiency and convergence guarantees, leaving a critical theoretical gap in understanding how well these algorithms generalize. To bridge this gap, we provide the first systematic generalization analysis for first-order gradient-based bilevel minimax solvers with lower-level minimax problems. Specifically, by leveraging algorithmic stability arguments, we derive fine-grained generalization bounds for three representative algorithms, including single-timescale stochastic gradient descent-ascent, and two variants of two-timescale stochastic gradient descent-ascent. Our results reveal a precise trade-off among algorithmic stability, generalization gaps, and practical settings. Furthermore, extensive empirical evaluations corroborate our theoretical insights on realistic optimization tasks with bilevel minimax structures.

Keywords

Cite

@article{arxiv.2604.20115,
  title  = {On the Stability and Generalization of First-order Bilevel Minimax Optimization},
  author = {Xuelin Zhang and Peipei Yuan},
  journal= {arXiv preprint arXiv:2604.20115},
  year   = {2026}
}