แนวคิดหลัก
Policy Space Response Oracles (PSRO) is a game-reasoning framework that combines traditional equilibrium computation with learning, providing a versatile approach for large-scale games.
บทคัดย่อ
PSRO is a fast-developing framework for large games, combining equilibrium computation with learning. It addresses challenges in game theory and has diverse applications. PSRO variants enhance strategy exploration efficiency and performance through MSS-RO combinations. The framework has been successfully applied to various domains, including mechanism design and robust reinforcement learning.
สถิติ
PSRO has been applied to security games [Wang et al., 2019; Wright et al., 2019], bargaining games [Li et al., 2023b; Wang and Wellman, 2024], Colonel Blotto games [An and Zhou, 2023], Pursuit-Evasion games [Li et al., 2023a], auctions [Li and Wellman, 2021], and mechanism design [Zhang et al., 2023].
Algorithms inspired by PSRO have reached state-of-the-art performance in large-scale games such as Barrage Stratego [McAleer et al., 2020] and in StarCraft [Vinyals et al., 2019].
คำพูด
"PSRO alternates between the analysis of the current game model, defining a new learning target, and game model refinement by including the new strategies generated via learning."
"Algorithms inspired by PSRO have reached state-of-the-art performance in large-scale games such as Barrage Stratego and in StarCraft."