Guszti.Bartfai@comp.vuw.ac.nz (Guszti Bartfai) (03/18/91)
Hello NNetters, Sorry for keep you waiting for this summary so long. Here are some interesting info that you might find useful and worth investigating further. * there is a newsgroup rec.games.go, that might be useful to follow. * there are two sites where you can find go information. ftp scam.berkeley.edu in /src/local I think blake.u.washington.edu who knows where. * Gerry Tesauro, who is at IBM Watson, created the winner of the Computer Games Olympiad Backgammon competition in 1989. This is the first time a learning program won over hand-crafted programs. A brief report is in Neural Computation vol 1 no. 3 (Fall 1989). The initial work was reported at NIPS 1 in 1988 and is in the proceedings (Neural Information Processing Systems vol 1, Touretzky, Ed., Morgan Kaufmann 1989) Other papers are Tesauro & Sejnowski, "A parallel network that learns to play backgammon", Artificial Intelligence 39, 357 (1989), and Tesauro, "Neural Network defeats creator in backgammon match" T.R. CCSR-88-6, U. of Illinois at Urbana-Champagne Center for Complex Research. * From: Dave Stoutamire <daves%curie.ces.cwru.edu@RELAY.CS.NET> "My thesis is on machine learning applied to go: it is not trivial, mainly for computatinal reasons. I ended up not using NNs for go because I didn't want to wait years for them to converge. Herbert Enderton at CMU (andrew.cmu.edu) has been doing some NNs for go for his PhD work." * From: Barney.Pell@computer-lab.cambridge.ac.uk "The following people (certainly the first) have done some work on NNs and GO: % Herbert.Enderton Herbert.Enderton@moriarty.theory.cs.cmu.edu %Howard Landman landman@sun.com I do work on Machine Learning and GO, but using logic and/or probability, not NNs." * From: landman@Eng.Sun.COM (Howard A. Landman) "...papers dealing with the computational complexity of GO Lichtenstein & Sipser, "Go is PSPACE-hard" (don't have the date or journal in front of me but they were posted to the net a couple of months ago) Yedwab, Laura, "On Playing Well In A Sum Of Games", MIT Masters thesis under Ron Rivest, available by sending email to (I think) publications@lcs.mit.edu" * From: heidi@ucthpx.uct.ac.za (Heidi de Wet) "...On a related topic, I saw a reference once to a ... Go program which reportedly played to 10 kyu (the Nemesis Go programs play to 20 and 15 kyu). The heart of the program was a so-called cellular automaton - a sort of one-layer NN with feedback. Each cell was one location on the board, and 'vacant' cells could take on one of 15 or so states. At each move, the net would cycle about 5 times before reaching a stable state, and picking its move. ...It comes from a collection of articles from the "Over the Horizon" column in the British weekly newspaper "Computing", by one Tony Durham. The book is entitled "Computing Horizons", by Tony Durham, published by Addison- Wesly, 1988. ISBN 0-201-18046-4." I have put all the replies into a file, which I'm happy to send anyone who is interseted. Thanks everyone who has responded. -- --- Guszti Bartfai --- Mail: P.O. Box 600, Wellington, NZ. Phone: +64 4 721-000 ext. 8580 E-mail: guszti@comp.vuw.ac.nz FAX: +64 4 712 070