Adaptive strategy optimization in game-theoretic paradigm using reinforcement learning

• Main Authors: Cheong, Kang Hao, Zhao, Jie
• Other Authors: School of Physical and Mathematical Sciences
• Format: Article
• Language: English
• Published: 2024
• Subjects: Mathematical Sciences; Adaptive strategy; Parrondo paradox
• Online Access: https://hdl.handle.net/10356/181728
• Institution: Nanyang Technological University

Parrondo's paradox refers to the counterintuitive phenomenon whereby two losing strategies, when alternated in a certain manner, can result in a winning outcome. Understanding the optimal sequence in Parrondo's games is of significant importance for maximizing profits in various contexts. However, the current predefined sequences may not adapt well to changing environments, limiting their potential for achieving the best performance. We posit that the optimal strategy that determines which game to play should be learnable through experience. In this Letter, we propose an efficient and robust approach that leverages Q learning to adaptively learn the optimal sequence in Parrondo's games. Through extensive simulations of coin-tossing games, we demonstrate that the learned switching strategy in Parrondo's games outperforms other predefined sequences in terms of profit. Furthermore, the experimental results show that our proposed method can be easily adjusted to adapt to different cases of capital-dependent games and history-dependent games.