EnsembleCard : A Strategy Ensemble Bot For Two-Player No-Limit Texas Hold’em Poker
2022 34th Chinese Control and Decision Conference (CCDC)(2022)
摘要
Nash Equilibrium is a powerful paradigm to create advanced AI in two-player zero-sum setting, which has a great advantage in getting a statistically significant result. However, playing millions of matches is time-consuming and even more problematic when human player participate in. Single paradigm is hard to hold absolute advantages in this real-world setting because of uncertainty in games. In this paper, we proposed a robust strategy ensemble framework in decision-making tasks, which is useful for real-world settings and has dominant advantage in comparison to single paradigm-based algorithm. We create a strong poker AI with two base decision-makers(including NE searching-base and rule-based), where we improve DeepStack and unlock the potential of CFVnet for creating a high-performance AI. Finally, we applied these approaches to build a poker AI for two-player no-limit Texas Hold’em and won the silver medal of the 2021 Chinese computer poker tournament. And additional experimental results also proved the effectiveness of strategy ensemble framework.
更多查看译文
关键词
Strategy Ensemble,Imperfect Information Game,Nash Equilibrium,Texas Hold’em Poker
AI 理解论文
溯源树
样例
![](https://originalfileserver.aminer.cn/sys/aminer/pubs/mrt_preview.jpeg)
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要