Re: [computer-go] Cellular automata

["chrilly" <chrilly@xxxxxxxx>]

>People following the CA thread might be interested in the following
>paper (submitted to the NIPS 2004 conference):
>
Evaluating a position from a high-level human game and evaluating a position
in a computer-game are two different things. Human position have already
some structure in it and it is therefore possible to detect at least parts
of this structure.
E.g. The stones form good shapes and adequatly cover own territory.
Computer games are in contrast more chaotic. This is especially true if one
evaluates at the end of a search. Although chess programm play nowadays a
reasonable game, practically all of the positions an evaluator sees are from
a human point of view completly weired. But the decision on the root is
based on this weired-positions. An evaluator most therefore be able to cope
with these positions. E.g. Does the evaluator adequatly deal with bad shapes
and/or overconcentration of stones? From the human games he has not learned
to deal with this configurations.

To see if the proposed modell is usefull, one should therefore also use:
A) Positions from computer-computer games.
B) Starting from these positions make some random moves and evaluate these
positions again.

Best Regards
Chrilly Donninger


_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/