No abstract

Poker has been considered a challenging problem in both artificial intelligence and game theory because poker is characterized by imperfect information and uncertainty, which are similar to many realistic problems like auctioning, pricing, cyber security, and operations. However, it is not clear that playing an equilibrium policy in multi-player games would be wise so far, and it is infeasible to theoretically validate whether a policy is optimal. Therefore, designing an effective optimal policy learning method has more realistic significance. This paper proposes an optimal policy learning method for multi-player poker games based on Actor-Critic reinforcement learning. Firstly, this paper builds the Actor network to make decisions with imperfect information and the Critic network to evaluate policies with perfect information. Secondly, this paper proposes a novel multi-player poker policy update method: asynchronous policy update algorithm (APU) and dual-network asynchronous policy update algorithm (Dual-APU) for multi-player multi-policy scenarios and multi-player sharing-policy scenarios, respectively. Finally, this paper takes the most popular six-player Texas hold ’em poker to validate the performance of the proposed optimal policy learning method. The experiments demonstrate the policies learned by the proposed methods perform well and gain steadily compared with the existing approaches. In sum, the policy learning methods of imperfect information games based on Actor-Critic reinforcement learning perform well on poker and can be transferred to other imperfect information games. Such training with perfect information and testing with imperfect information models show an effective and explainable approach to learning an approximately optimal policy.

With the development and appliance of multi-agent systems, multi-agent cooperation is becoming an important problem in artificial intelligence. Multi-agent reinforcement learning (MARL) is one of the most effective methods for solving multi-agent cooperative tasks. However, the huge sample complexity of traditional reinforcement learning methods results in two kinds of training waste in MARL for cooperative tasks: all homogeneous agents are trained independently and repetitively, and multi-agent systems need training from scratch when adding a new teammate. To tackle these two problems, we propose the knowledge reuse methods of MARL. On the one hand, this paper proposes sharing experience and policy within agents to mitigate training waste. On the other hand, this paper proposes reusing the policies learned by original teams to avoid knowledge waste when adding a new agent. Experimentally, the Pursuit task demonstrates how sharing experience and policy can accelerate the training speed and enhance the performance simultaneously. Additionally, transferring the learned policies from the N-agent enables the (N+1)–agent team to immediately perform cooperative tasks successfully, and only a minor training resource can allow the multi-agents to reach optimal performance identical to that from scratch.

In the implementation of deep reinforcement learning (DRL), action persistence strategies are often adopted so that agents maintain their actions for a fixed or variable number of steps. The choice of the persistent duration for agent actions usually has notable effects on the performance of reinforcement learning algorithms. Aiming at the research gap of global dynamic optimal action persistence and its application in multi-agent systems, we propose a novel framework: global dynamic action persistence (GLDAP), which achieves global action persistence adaptation for deep reinforcement learning. We introduce a closed-loop method that is used to learn the estimated value and the corresponding policy of each candidate action persistence. Our experiment shows that GLDAP achieves an average of 2.5% ∼ 90.7% performance improvement and 3 ∼ 20 times higher sampling efficiency over several baselines across various single-agent and multi-agent domains. We also validate the ability of GLDAP to determine the optimal action persistence through multiple experiments.