comp.ai.neural-nets #5729 (1 more) [1] From: alph@wsinfo01.win.tue.nl (G.J.W. Hagenaars) [1] SUMMARY Neural Networks and Games (was: Request for stuff) Summary: yes Keywords: neural networks, games Date: Fri Mar 06 05:30:35 CST 1992 Lines: 202 On the 5th of february we posted the following message on several newgroups (comp.ai.neural-nets, rec.games.backgammon, rec.games.go, rec.games.chess, rec.games.misc) >This is a request for articles, ideas and anything else you might >think useful after reading this. >We have decided that we are going to do a master's thesis on the >subject of Neural Networks and games (boardgames, Go in >particular). >So we are looking for anything we can get on this subject. >Some ideas: games, complexity of games, strategy with AI >We would also like to get ftp-sites for articles, programs and >the like. >Feel free to respond (via e-mail preferably). >If we get enough responds (>2 :-) we will make a summary of the data >we got, and post it to the net (deal? deal!). > >Jack van Rijswijck javhar@win.tue.nl >Gert-Jan Hagenaars alph@win.tue.nl > > Department of Mathematics > Eindhoven University of Technology > The Netherlands As the response to our message exceeded the threshold we set (2) we'll summarize. But before all that, all this: 1) Before we start working on our thesis we will make a taxonomy of (fairly) recent developments on Neural Networks and `Strategy', so information is still welcome; 2) A good place to start looking for stuff is the FAQ-posting; 3) We don't actually know what is costumary about publishing email-addresses. We think it would be a bad idea to post email-addresses on netnews, as people would probably try and send their requests to the writers of the articles themselves, thus clogging mailboxes all over the place. So, we decided that we will only give email-addresses from people who replied to our posting; 4) When using anonymous ftp, use your email-address as password: only then the ftp-site will be able to contact you when necessary; 5) This space unintentionally left blank ===================================================================== * * * S U M M A R Y * * * As our prime interest is the application of neural networks in game playing, we'll start this summary with: ON NEURAL NETS AND GO David Stoutamire wrote a report on Machine Learning applied to Go, available on milton.u.washington.edu (128.95.136.1), directory public/go, report.ps.Z. Mind you, this takes a LONG time to print! There is also a lot of go-related software available on milton. Herbert Enderton has been working on a neural net go program. His tech report entitled 'the Golem go program' can be obtained ($2, overseas $5) at: Computer Science Documentation; School of Computer Science; Carnegie Mellon University; Pittsburgh PA, 15213-3890; reports@cs.cmu.edu; 412/268-2596. It appears that the network plays professional looking moves, but is not very good at life and death yet. ON NEURAL NETS AND GAMES Gerald Tesauro has done quite some work with neural nets that play backgammon. First, he trained a net with expert data sets. The following articles report on this: A Parallel Network that learns to play Backgammon G. Tesauro & T.J. Sejnowski Artificial Intelligence, vol 39, pp. 357-390, 1989. Neurogammon: A Neural Network Backgammon Program G. Tesauro & T.J. Sejnowski IJCNN Proceedings, vol 3, pp. 33-39, 1990. To appear in Machine Learning is this article, in which he comments on temporal difference learning (i.e. training a net from scratch by playing a copy of itself): Practical Issues in Temporal Difference Learning G. Tesauro Computer Science, #76307 submitted Also to be found in Artificial Intelligence: A Pattern Classification Approach to Evaluation Function Learning Kai-Fu Lee and Sanjoy Mahajan Artificial Intelligence, vol 36, pp. 1-25, 1988. We haven't checked this one out yet, so we're not really sure that it deals with games: RAMBOT: A Connectionist Expert System That Learns by Example Michael C. Mozer Proceedings of the IEEE First International Conference on Neural Networks vol 2 (Maureen Caudill editor), pp. 693-700, 1987. Not very recent, but very legendary is A.L. Samuel's work in the sixties: Some studies in machine learning using the game of checkers A.L. Samuel IBM journal of Research and Development, vol 3, nr. 3, pp. 210-229, 1959. Some studies in machine learning using the game of checkers 2 - recent progress A.L. Samuel IBM journal of Research and Development, vol 11, nr. 6, pp. 601-616, 1967. Concerning the application of neural networks in bridge: Knowledge acquisition and reasoning based on neural networks - the research of a bridge bidding system He Yo, Zhen Xianjun, Ye Yizheng, Li Zhongrong INNC '90, Paris, vol 1, pp. 416-423. Experience gained with a neural network that learns to play bridge M. Kohle & F. Schonbauer Proc. of the 5th Austrian Artificial Intelligence meeting, pp. 224-229. ON COMPUTERS AND GO Bruce Wilcox, author of Nemesis, published a series of articles entitled "Instant Go" in the American Go Journal. Also, he reported in SIGART newsletter, october 1985, no 94, on his go programs. Nemesis' rating is about 15 kyu. Not really using neural nets, but still interesting: FIELD: a data type for automated perception in GO Z. Kok and V. Akman proc. sixth int. symposium on computer and information sciences vol 2, pp. 641-650 Elsevier Science Publishers B.V., Amsterdam 1991. ON COMPUTERS AND GAMES Barney Pell (Barney.Pell@cl.cam.ac.uk) did some very interesting work on games and learning in a more general context: metagame. His aim is to program a computer to analyse and play any game, when supplied the rules. The annual computer olympiad involves tournaments in a variety of games. These publications contain a wealth of interesting articles: Heuristic Programming in Artificial Intelligence - the first computer olympiad D.N.L. Levy & D.F. Beal eds. Ellis Horwood ltd, 1989. Heuristic Programming in Artificial Intelligence 2 - the second computer olympiad D.N.L. Levy & D.F. Beal eds. Ellis Horwood, 1991. Heuristic Programming in Artificial Intelligence 3 - the third computer olympiad H.J. van den Herik & L.V. Allis eds. Ellis Horwood, 1992. To conclude: there are a lot of papers on neural networks available at the ftp-site archive.cis.ohio-state.edu (128.146.8.52), directory pub/neuroprose. Thanks very much to the following people for responding: Justin Wells (bfield!jwells@relay.eu.net) Tom Verhoeff (wstomv@win.tue.nl) Erik Tjong Kim Sang (erikt@let.rug.nl) Gerald Tesauro at ibm David Stoutamire (davids@icsi.berkeley.edu) Randall Smith (rsmith@astroa.physics.wisc.edu) Ed Rybak (rybak@sequent.com) Barney Pell (Barney.Pell@cl.cam.ac.uk) Dan Olson (dano@ssc-vax.boeing.com) Adrian Mariano (adrian@u.washington.edu) Chris Love (cdlove@dretor.dciem.dnd.ca) George Loegel (ettare@zip.eecs.umich.edu) David Fotland (fotland@hpihoc.cup.hp.com) Herbert Enderton (Herbert_Enderton@moriarty.theory.cs.cmu.edu) David Dowe (dld@dibbler.cs.monash.edu.au) David Doshay (doshay@ursa.arc.nasa.gov) Robert Baruch (baruch@nynexst.com) Varol Akman (akman@trbilun.bitnet) Keep us posted! Gert-Jan Hagenaars alph@win.tue.nl Jack van Rijswijck javhar@win.tue.nl Department of Mathematics Eindhoven University of Technology the Netherlands