A study on some games with reinforcement learning
- 발행기관 서강대학교 일반대학원
- 지도교수 김종락
- 발행년도 2017
- 학위수여년월 2017. 2
- 학위명 석사
- 학과 및 전공 일반대학원 수학과
- 실제URI http://www.dcollection.net/handler/sogang/000000061219
- 본문언어 영어
- 저작권 서강대학교 논문은 저작권보호를 받습니다.
초록/요약
Artificial intelligence has increased in popularity as a result of the victory in the match between Lee Sedol and AlphaGo, an artificial intelligence Go program developed by Google DeepMind. In this paper, we apply Q-learning, one of the reinforcement learning algorithms, to tic-tac-toe, which is a simpler game than Go, and renju, which is a type of Gomoku. In addition, the newly proposed mathematical game, the factorization game, is implemented in C-language, and Q-learning is applied. As a result of the analysis, Renju confirmed that the number of cases is too diverse, which makes learning difficult with Q-learning. The factorization game is a fair simulation when playing black and white 10 * 10 Go-board. For instance, when p = 4, place two or three Go-stones on the board. And we analyze the results of Q-learning applied to the factorization game. In the future, we plan to analyze the effect of the deep neural network technology, which shows good performance for a larger number of cases when applied to Renju, and also analyze the factorization game when the board size is increased.
more