Incomplete Information Competition Strategy Based on Improved Asynchronous Advantage Actor Critical Model

Abstract

In recent years, game theory has been widely used in the field of deep learning, mainly including intelligent competition strategies of complete information games and incomplete information games. This paper focuses on incomplete information games, and proposes a low-dimensional semantic feature based on category coding and an incomplete information competition strategy based on the improved Asynchronous Advantage Actor-Critic (A3C) network model. First, the A3C network model in deep reinforcement learning is adopted in the competition strategy, and its network structure is improved according to the semantic features based on category coding. The improved A3C model is implemented in parallel by a series of "workers". The "workers" is a new deep learning model structure proposed in this paper. Secondly, this article combines supervised learning and Deep Reinforcement Learning (DRL) to propose a new competitive strategy. Through conducting a large number of real-time experiments with human players on online competitive websites, the comparison with the existing methods in terms of the ratio of winning and losing and the ranking rate, the experimental results indicate the superiority of the new competitive strategy.