Scholarlink

<p>Adversarial team games (ATGs) with <i>en ante</i> coordination involve a team competing against an adversary under conditions of incomplete information, where team members can coordinate their strategies before the game starts. <i>Team-maxmin equilibrium with correlation</i> (TMECor) is a core solution concept in this setting. While some existing algorithms leverage game transformation techniques to compute TMECor by converting ATGs into two-player zero-sum games, these methods often encounter scalability issues. The exponential growth in the transformed game tree size impedes computational speed and consumes significant storage resources. In this paper, we propose a novel method based on perfect-recall refinement that achieves game tree transformation while maintaining the original size. Additionally, we introduce an innovative teammate modeling approach, allowing team members to infer private information based on observations of their teammates’ actions. Experimental results on benchmark testbeds show a significant improvement in equilibrium computation efficiency without expanding the game size, indicating the effectiveness of our method. Our work not only addresses the computational complexities associated with ATGs but also provides valuable insights for further research on enhancing strategic decision-making in collaborative adversarial settings.</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

PRR-TM: finding equilibria in adversarial team games via perfect-recall refinement and teammate modeling

  • 期刊论文

  • 发表日期：

    2025年3月28日

查看全文
  • Chen Qiu,
  • Weixin Huang,
  • Hongji Xiong,
  • Jiajia Zhang,
  • Xuan Wang

摘要

Adversarial team games (ATGs) with en ante coordination involve a team competing against an adversary under conditions of incomplete information, where team members can coordinate their strategies before the game starts. Team-maxmin equilibrium with correlation (TMECor) is a core solution concept in this setting. While some existing algorithms leverage game transformation techniques to compute TMECor by converting ATGs into two-player zero-sum games, these methods often encounter scalability issues. The exponential growth in the transformed game tree size impedes computational speed and consumes significant storage resources. In this paper, we propose a novel method based on perfect-recall refinement that achieves game tree transformation while maintaining the original size. Additionally, we introduce an innovative teammate modeling approach, allowing team members to infer private information based on observations of their teammates’ actions. Experimental results on benchmark testbeds show a significant improvement in equilibrium computation efficiency without expanding the game size, indicating the effectiveness of our method. Our work not only addresses the computational complexities associated with ATGs but also provides valuable insights for further research on enhancing strategic decision-making in collaborative adversarial settings.