摘要
以博弈树搜索为核心的α-β剪枝算法,受限于估值函数对设计者棋力水平的依赖,难以实现更进一步的提升。论文提出的UCT(Upper Confidence Bound Apply to Tree)算法结合了UCB公式和蒙特卡洛树搜索算法,弱化了算法本身对估值函数的依赖性,最大化利用计算机的算力优势,提升算法的整体效率,并利用其并行化优势优化算法,基于点格棋进行了算法的实现。
Alpha-beta pruning algorithm based on searching tree is limited by the dependence of the evaluation function on the designer’s chess ability,and it is difficult to achieve further improvement. Paper puts forward the UCT(the Upper Confidence Bound Apply to Tree) algorithm is a combination of UCB formula and Monte Carlo Tree search,which weakens the dependence of the algorithm on the estimation function,maximizes the computer ’ s computing power,enhances the overall efficiency of the algorithm,takes advantage of the parallel optimization algorithm,and realizes the algorithm based on the dots and boxes.
作者
张宜放
孟坤
ZHANG Yifang;MENG Kun(School of Computer,Beijing Information Science and Technology University,100101,China;Sensing and Computational Intelligence Joint Lab,Beijing Information and Science and Technology University,100101,China)
出处
《智能计算机与应用》
2020年第4期27-31,共5页
Intelligent Computer and Applications
基金
北京信息科技大学2019年促进高校内涵发展-大学生科研训练项目(5101923400)
科技计划一般项目(KM201911232002)资助
关键词
UCT算法
估值函数
点格棋
UCT algorithm
evaluation function
dots and boxes
