Application of Deep Reinforcement Learning in the Board Game

Abstract

Deep reinforcement learning is one of the core technologies, which leads to artificial intelligence. It has been used widely not only in robot, autonomous driving, power station control and human-computer interaction in mobile e-commerce platforms, but also in gambling games. The game software AlphaGo of artificial intelligence defeated Ke Jie in 2017. Since then, the study of artificial intelligence and deep reinforcement is widely concerned. We build a Neural Network model using TensorFlow and Monte Carlo tree search algorithm, so that agents can reach the level of playing chess with human beings in board games. The gobang interface is realized by Python, and the PhoenixGo recurrence is completed. The computation is reduced and the training effect of the agent is improved by optimizing the establishment of gobang strategic value network. Black hand first can achieve a 73% winning rate, and white hand first is 64%, when playing against human players.