Deep Reinforcement Learning Policy in Hex Game System

Hex game is a zero-sum chess game. It has a large solution space when using 11 x 11 size of chess board. In recent years, deep reinforcement learning -based Go game systems, i.e. AlphaGo and AlphaGo Zero, have gotten huge achievement. In this paper, we design the self-learning method and system structure of Hex game. design policy network and value network referred to residual network, and use asynchronous advantage actor-critic algorithm to train policy network and value network. The comparison of deep reinforcement learning-based policy network and fixed strategy proves better effect of self-learning.