-
TDgammon?
Hi, me again, still trying to come up with a solution for Backgammon AI.
When I read about Gerald Tesauro's TDgammon - a backgammon AI implementation that can learn for itself yet can also play pretty well, I thought it'd be nice to dive into the subject and do something similar myself.
Wikipedia, just like other results in Google, had only shown me scattered information on puzzling subjects: Reinforcement Learning, Temporal Difference, TD(Lambda), Artificial Neural Networks, Feed-forward Backpropagation, Hidden Layers, etc. Unfortunately, being as thick as I am, I got quite lost in the land of terminology and couldn't put these little pieces together.
Apparently the water is deeper than I had hoped.
So any helpful tips on where do I begin, or how do I put it all together, would be greatly appreciated.
-
If I remember correctly, TDGammon use mostly temporal difference learning, so you'd really just want to look up on that.
-
Thanks, Happy_Reaper, I did manage to find an online book that mentions TDGammon has also to do with neural networks (or back propagating multi layered perceptron network), so will try to see how the two work together.