Accelerating Min-Max Optimization via Power-Law Stepsizes

arXiv:2606.01764 · math.OC, cs.GT, cs.LG · Submitted 2026-06-01 · Read on arXiv

math.OC, cs.GT, cs.LG

Submitted: 2026-06-01

Updated: 2026-10-30

Terminology

• Convergence Guarantees — 1× in this paper · explained in 3 episodes

Related papers

• Lions and Muons: Optimization via Stochastic Frank-Wolfe under Heavy-Tailed Noise
• Breaking the Stochasticity Barrier: An Adaptive Variance-Reduced Method for Variational Inequalities
• Service-Induced Congestion in Memory-Constrained LLM Serving
• Adam-HNAG: A Convergent Reformulation of Adam with Accelerated Rate
• Incremental Learning in Mirror Flows
• Beyond Aggregate Drift: Optimal Client-Specific Windows in Federated Learning