Date: 26 May 89 14:43:10-PST
From: Vision-List moderator Phil Kahn <Vision-List-Request@ADS.COM>
Errors-to: Vision-List-Request@ADS.COM
Reply-to: Vision-List@ADS.COM
Subject: Vision-List delayed redistribution
To: Vision-List@ADS.COM

Vision-List Digest	Fri May 26 14:43:10 PDT 89

 - Send submissions to Vision-List@ADS.COM
 - Send requests for list membership to Vision-List-Request@ADS.COM

Today's Topics:

 learning to play Go and Neural Networks: info request

----------------------------------------------------------------------

Date: 23 May 89 08:32:41 GMT
From: munnari!latcs1.oz.au!mahler@uunet.UU.NET (Daniel Mahler)
Subject: learning to play Go and Neural Networks: info request
Organization: Comp Sci, La Trobe Uni, Australia

[ Not a mainline vision question, but perhaps of interest to some.
		phil...		]


i am an honours student working on implementing
a neural network that learns to play Go.
this seems appropriate as Go has a much
higher (~50x) branching factor, forcing
a pattern oriented rather than a lookahead
orinted approach, and most instruction
(heuristics advocated by books & players)
are of an intuitive nature hard to 
formalise into the classical ai symbolic/logical
paradigm. My current idea is to preprocess
the board position using image/signal processing techniques (eg transforms,
filters, convolution, multi dimensional grammars)
to enhance the strategic structure of
the position over superficial similarities;
in other words i will treat the board
as a 2d signal/image. I will concentrate mainly on
the opening to early middle game phase, as these are
quiescent; later stages become more dependent on tactical
lookahead. All responses, be they bibliographic,
theoretical, practical, or philosophical, will be apprecited.
i leave to your discretion to judge
the appropriatness of replying by news or email.



------------------------------

End of VISION-LIST
********************
