Learning To Predict Life And Death With Ann In Computer Go
International Conference on Computing, Communications and Control Technologies, Vol 1, Proceedings(2004)
摘要
Tsumego ( FA- Life and Death) is a computer Go game sub-problem that determines whether a group of stones is safe and free from ever being captured - the "life" state, or may potentially be captured after a sequence of moves, no matter how he/she defenses it - the "death" state. The ability to quickly and accurately determine "life and death" is crucial to becoming an expert at Go. Humans usually learn this skill by studying examples in classic Go books written by professional players. Most examples usually use only a corner of the Go board or a side of the board to illustrate how to archive "alive" or "dead" state. This paper shows how we designed and used an Artificial Neural Network (ANN) to simulate this learning process. We performed two different experiments - (1) stones in "final position" and (2) stones not in "final position. Using back propagation learning algorithm with a training set of over a 1000 examples of 8x8 corner positions, we were able to achieve a very high accuracy rate of 97% for final position tests and 94.5% for nonfinal position tests. This paper describes the design of our ANN and algorithm as well as provides an analysis and comparison of our experiment results.
更多查看译文
关键词
computer Go, Artificial Neural Network, learning
AI 理解论文
溯源树
样例
![](https://originalfileserver.aminer.cn/sys/aminer/pubs/mrt_preview.jpeg)
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要