Scholarlink

<p>Imperfect-information games (IIGs) have attracted significant attention due to the complexity arising from hidden information. In recent years, efficiently solving optimal strategies in IIGs has become a major challenge. Equilibrium strategies, with their solid theoretical foundation and demonstrated performance, have achieved great success. The agents like Libratus and DeepNash demonstrate superior performance in public evaluations. This survey comprehensively reviews two key methods for solving Nash equilibrium strategies in two-player zero-sum IIGs: counterfactual regret minimization (CFR) and fictitious self-play (FSP), including their improved variants. We provide a detailed overview from multiple dimensions, including problem modeling, solution paradigms, and classical solving methods. To further evaluate these methods, we conducted a series of evaluation experiments to deeply analyze and compare their performance. Finally, this survey summarizes current challenges and outlines future research directions, aiming to provide useful insights for advancing the field.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Nash equilibrium strategy solving in two-player imperfect-information games: a survey

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月25日

查看全文
  • Huale Li,
  • Zhaobin Wang,
  • Zhen Yang,
  • Shuhan Qi,
  • Jiajia Zhang

摘要

Imperfect-information games (IIGs) have attracted significant attention due to the complexity arising from hidden information. In recent years, efficiently solving optimal strategies in IIGs has become a major challenge. Equilibrium strategies, with their solid theoretical foundation and demonstrated performance, have achieved great success. The agents like Libratus and DeepNash demonstrate superior performance in public evaluations. This survey comprehensively reviews two key methods for solving Nash equilibrium strategies in two-player zero-sum IIGs: counterfactual regret minimization (CFR) and fictitious self-play (FSP), including their improved variants. We provide a detailed overview from multiple dimensions, including problem modeling, solution paradigms, and classical solving methods. To further evaluate these methods, we conducted a series of evaluation experiments to deeply analyze and compare their performance. Finally, this survey summarizes current challenges and outlines future research directions, aiming to provide useful insights for advancing the field.