How do you tackle the problem of learning to play in Tic-Tac-Toe like games? From the point of view of machine learning this is tipically seen as a "reinforcement learning" task. I have never studied that area in detail, since during the development of my thesis I was centered in supervised learning. During ICML 2003 I noticed a strong interest in the community towards reinforcement learning problems (5 sessions were dedicated to the topic).
Today I have been trying to order my ideas and tackle some small toy problems about learning in games. I have noticed that in order to perform research in the area I do not only need to read about Game Theory but also about reinforcement learning. I have found two valuable resources that are now addequately added to my reading list (and both are online).
- Reinforcement Learning: An Introduction by Richard S. Sutton and Andrew G. Barto
- Michael Duff's PhD Thesis that explains reinforcement learning in Bayesian terms.
An also interesting page is David J.C. MacKay's homepage that contains a full version of his recently published book on Information Theory, Inference, and Learning Algorithms
This is one of this days in life when you feel absolutely unproductive. I think that I have been reading too much and now my brain is busy processing. I have been in front of the computer for more than 10 hours today with an astonishing result of 5 slides.