Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!uunet!news1.digital.com!pa.dec.com!depot.mro.dec.com!janix.mfr.dec.com!stkai1.enet.dec.com!t_andersson
From: t_andersson@stkai1.enet.dec.com ()
Subject: Re: GO programming question useing nueral nets.
Message-ID: <1995Jun12.100602.15230@janix.mfr.dec.com>
Lines: 23
Sender: news@janix.mfr.dec.com (SDSC USENET News System)
Reply-To: t_andersson@stkai1.enet.dec.com ()
Organization: Digital Equipment GmbH Alpha Porting Center OSSC/SDSC
X-Newsreader: dxrn 6.18-32
References: <3qro9r$pd4@hustle.rahul.net> <py-cNed.predictor@delphi.com> <3r4bmm$hgr@hustle.rahul.net> <3r6ecs$1o1m@bigblue.oit.unc.edu>
Date: Mon, 12 Jun 1995 10:06:02 GMT


Matthew Carlson writes:

|>>Why a tree search? Why does everything always come down to a stupid tree 
|>>search. A tree search with a nueral net evalutor would take a 
|>>considerably longer time and would not completely sovle the problem. I 
|>>would have to train the net the values of moves. This in chess is easy 
|>>but not in GO! I keep getting responces from people who 1. think I am 
|>>really lame for trying this or 2. think alpha beta searches are a god 
|>>send. I need info here please. I have talked to people who have tried 
|>>using a tree search with a nueral net evaluator and they haven't had much 
|>>success. Besides an evaluator may output the same value for all legal 
|>>moves. Then what? Well I am stumped any help would be appreciated.
|>>-- 
|>>Matthew Carlson <carlson@rahul.net>

Why don't you create a population of neural nets, let them play each
other, and create some evolutionary pressure with a genetic algorithm.

There's no law against killing a neural network, so you can be as
hard as you like to the loosers :D.

- Tomas Andersson -
