Backgammon-related Thesis

[MuffinHead]

I have recently finished my thesis, "Search in Trees with Chance Nodes".
The thesis explored a new method for performing search in games which
involve chance events. The main focus of interest was on backgammon, and
I used the set of GNU Backgammon neural networks for performing
evaluation of backgammon positions, while I implemented my own search
routines. The search routines in question are the *-Minimax functions
created by Bruce Ballard in 1983, dubbed Star1 and Star2.

While the final results were not as spectacular as we would have hoped
(mainly due to how the Gnubg NN works rather than inherent problems with
the algorithms), the great hope is that the (first) real-world results I
presented will renew interest in improving search in stochastic domains.

Being a Master's thesis in Computing Science, it is not intended for a
general audience, but if you have an interest in backgammon, or AI,
there should be plenty there to chew on (mentally, that is).

My thesis is available in PDF, PS and PS.gz formats at

http://www.cs.ualberta.ca/~hauk/

Thanks again to the GNU Backgammon team (including anyone who felt I was
wasting their time asking questions about the code!). It's not often one
runs into a program that is so good, it can barely be improved. Kudos.