Fast Last-Iterate Convergence of Learning in Games Requires Forgetful Algorithms

• URL: http://arxiv.org/abs/2406.10631v1
• Date: Sat, 15 Jun 2024 13:26:17 GMT
• Title: Fast Last-Iterate Convergence of Learning in Games Requires Forgetful Algorithms
• Authors: Yang Cai, Gabriele Farina, Julien Grand-Clément, Christian Kroer, Chung-Wei Lee, Haipeng Luo, Weiqiang Zheng,
• Abstract summary: Self-play via online learning is one of the premier ways to solve large-scale two-player zero-sum games. We show that OMWU offers several advantages including logarithmic dependence on the size of the payoff matrix. We prove that a broad class of algorithms that do not forget the past quickly all suffer the same issue.
• Score: 71.73971094342349
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Self-play via online learning is one of the premier ways to solve large-scale two-player zero-sum games, both in theory and practice. Particularly popular algorithms include optimistic multiplicative weights update (OMWU) and optimistic gradient-descent-ascent (OGDA). While both algorithms enjoy $O(1/T)$ ergodic convergence to Nash equilibrium in two-player zero-sum games, OMWU offers several advantages including logarithmic dependence on the size of the payoff matrix and $\widetilde{O}(1/T)$ convergence to coarse correlated equilibria even in general-sum games. However, in terms of last-iterate convergence in two-player zero-sum games, an increasingly popular topic in this area, OGDA guarantees that the duality gap shrinks at a rate of $O(1/\sqrt{T})$, while the best existing last-iterate convergence for OMWU depends on some game-dependent constant that could be arbitrarily large. This begs the question: is this potentially slow last-iterate convergence an inherent disadvantage of OMWU, or is the current analysis too loose? Somewhat surprisingly, we show that the former is true. More generally, we prove that a broad class of algorithms that do not forget the past quickly all suffer the same issue: for any arbitrarily small $\delta>0$, there exists a $2\times 2$ matrix game such that the algorithm admits a constant duality gap even after $1/\delta$ rounds. This class of algorithms includes OMWU and other standard optimistic follow-the-regularized-leader algorithms.

Related papers

• Last-Iterate Convergence Properties of Regret-Matching Algorithms in Games [72.43065138581589]
  We study the last-iterate convergence properties of various popular variants of RM$+$. We show numerically that several practical variants such as simultaneous RM$+$, alternating RM$+$, and predictive RM$+$, all lack last-iterate convergence guarantees. We then prove that recent variants of these algorithms based on a smoothing technique do enjoy last-iterate convergence.
  arXiv  Detail & Related papers  (2023-11-01T17:34:58Z)
• Doubly Optimal No-Regret Learning in Monotone Games [10.760195409078591]
  We propose the first doubly optimal no-regret learning algorithm for smooth monotone games. Our algorithm achieves both (i) the optimal $O(sqrtT)$ regret in the adversarial setting under smooth and convex loss functions and (ii) the optimal $O(frac1T)$ last-iterate convergence rate to a Nash equilibrium.
  arXiv  Detail & Related papers  (2023-01-30T17:55:53Z)
• Global Nash Equilibrium in Non-convex Multi-player Game: Theory and Algorithms [66.8634598612777]
  We show that Nash equilibrium (NE) is acceptable to all players in a multi-player game. We also show that no one can benefit unilaterally from the general theory step by step.
  arXiv  Detail & Related papers  (2023-01-19T11:36:50Z)
• Policy Optimization for Markov Games: Unified Framework and Faster Convergence [81.3266426402464]
  We show that the state-wise average policy of this algorithm converges to an approximate Nash equilibrium (NE) of the game. We extend this algorithm to multi-player general Markov Games and show a $mathcalwidetildeO(T-1/2)$ convergence rate to Correlated Equilibria (CCE)
  arXiv  Detail & Related papers  (2022-06-06T14:23:13Z)
• Almost Optimal Algorithms for Two-player Markov Games with Linear Function Approximation [92.99933928528797]
  We study reinforcement learning for two-player zero-sum Markov games with simultaneous moves. We propose an algorithm Nash-UCRL-VTR based on the principle "Optimism-in-Face-of-Uncertainty" We show that Nash-UCRL-VTR can provably achieve an $tildeO(dHsqrtT)$ regret, where $d$ is the linear function dimension.
  arXiv  Detail & Related papers  (2021-02-15T09:09:16Z)
• Near-Optimal Reinforcement Learning with Self-Play [50.29853537456737]
  We focus on self-play algorithms which learn the optimal policy by playing against itself without any direct supervision. We propose an optimistic variant of the emphNash Q-learning algorithm with sample complexity $tildemathcalO(SAB)$, and a new emphNash V-learning algorithm with sample complexity $tildemathcalO(S(A+B))$.
  arXiv  Detail & Related papers  (2020-06-22T05:00:13Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.