SCS Undergraduate Thesis Topics
|Yucheng Low||Daniel Sleator||Investigating the Use of Machine Learning in Go|
The game of Go is practically the Holy Grail for computer game playing due to its massive branching factor and difficult evaluation. The current state of the art computer program is only able to play at an amateur level. Additionally, the computer programs tend to have specific weaknesses which can be targeted by professional human players. For this project, we constrain the problem to 9x9 Go and investigate the use of Machine Learning methods to train an evaluation function. Both supervised learning and reinforced learning schemes are considered. The resultant evaluation function can be directly applied to the game through an Alpha-Beta search, but other methods of applying it, such as in a Monte Carlo Go implementation, are also investigated.