Solving Life and Death Problem of Go with Deep Learning

• Main Authors: Guan-Lun Cheng, 程冠倫
• Other Authors: Shi-Jim Yen
• Format: Others
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/2225z3

碩士 === 國立東華大學 === 資訊工程學系 === 105 === The Life and Death problem is a basic concept of Go. This problem is an importantissue to improve Computer Go, because it needs a perfect solution, require no mistake, and it is approved in game usually. This problem was solved by the search tree methodor simulated with patterns. AlphaGo and DarkforestGo applies Deep Convolutional Neural Network on their policy model learning, which can increase the accuracy in prediction of the next move of Go than normal neural network. This paper proposed a DCNN model to solve the Life-and-Death Problem, predicts the correct next move in the problem. Our experimental results show the prediction of DCNN model can solve the problem perfect usually. This paper also analyzes some special cases of prediction error. Finally, this paper integrates the prediction with DarkforestGo DCNN model for increasing the prediction rate.