透過您的圖書館登入
IP:3.17.154.171
  • 學位論文

在HappyGO上應用模擬棋局平衡

Simulation Balancing for HappyGO

指導教授 : 吳毅成

摘要


這篇論文的目的是應用模擬棋局平衡(Simulation Balancing)於HappyGO上,並且藉由調整參數以及不同的訓練資料等等期望能找到更好的結果。HappyGO曾獲得2010年中華民國人工智慧學會9x9電腦圍棋組的銀牌。 根據我們的實驗分析,模擬棋局平衡能讓程式的勝率從原本訓練前(特徵權重皆為0)的9.3%,提升到48.7%,我們的勝率算法是讓HappyGO以一步500場模擬棋局對Gnu GO 3.8等級十。

並列摘要


The main purpose of this thesis is to apply simulation balancing to HappyGO and expect to find better results by adjusting the parameters, different training data and so on. HappyGO is one the Conputer GO program, and won silver of TAAI 9x9 Computer Go group. According to our experiments on simulation balancing, HappyGO with 500 playouts per move has a 9.3% win rate against Gnu GO 3.8 level 10 before training and raises to 48.7% after training.

並列關鍵字

Simulation Simulation Balancing GO HappyGO AI 9x9

參考文獻


[2] 王永樂,基於UCT之九路電腦圍棋程式HappyGO的設計與實作,交通大學資訊學院碩士在職專班資訊組學位論文,2009。
[5] B. Bouzy, “History and territory heuristics for Monte-Carlo Go,” New Mathematics and Natural Computation 2, pp. 1–8, 2006.
[7] K. Chen, P. Zhang, “A New Heuristic Search Algorithm for Capturing Problems in Go,” ICGA Journal 29(4), pp. 183–190, 2006.
[12] S. Gelly, D. Silver, “Combining online and offine knowledge in UCT,” ICML, vol. 227, pp. 273-280, 2007.
[14] S. Gelly and D. Silver, “Monte-Carlo Tree Search and Rapid Action Value Estimation in Computer Go,” Artificial Intelligence, pp. 1–33, 2011.

延伸閱讀