面向智能博弈游戏的卷积神经网络估值方法

CONVOLUTIONAL NEURAL NETWORK VALUATION METHOD FOR INTELLIGENT GAME

非完备信息博弈中存在的许多问题在日常生活也同样存在,研究它对解决人们日常中的问题以及提高生活质量有重要意义。德州扑克是典型的非完备信息博弈牌类游戏,针对德州扑克博弈提出一种基于卷积神经网络的估值算法模型。选择用大师之间的博弈历史记录来训练该模型,从而达到学习大师的目的。将该估值模型的博弈程序与前人设计的博弈程序进行博弈,实验结果表明:学习人类大师经验的卷积神经网络估值方法可以提供更好的决策,增强了德州扑克博弈程序的牌力。

Many problems in the imperfect information game also exist in daily life.Studying about it is of great significance to solve our daily problems and improve people s quality of life.Texas Hold em is a typical imperfect information game.This paper proposes a valuation algorithm based on convolutional neural network for Texas Hold em.We chose to use the game history between masters to train the model so as to learn the experience of masters.The game agent based on the valuation model gamed with the agent designed by the predecessors.The experiment proves that the convolutional neural network valuation method based on human master experience can provide better decision-making and enhance the power of the agent.