[computer-go] Cellular Automata

[Peter Drake <drake@xxxxxxxxxx>]

On Thursday, July 22, 2004, at 05:16  PM, Chris Fant wrote:

> I might as well throw my two cents in.  It seems like CA, by it's 
> nature, is going to have a hard time with non-local concepts.  How 
> would a CA-based engine (I assume it is really just a CA-based 
> evaluation, correct me if I'm wrong) take non-local features into 
> account?
One answer, as other have mentioned, is that this information can be 
propagated.

One of the conclusions we drew from our neural network research last 
summer was, "If you know how to do it, don't ask the machine learning 
algorithm to figure it out."  Thus, since we know many features are 
sensible at the level of blocks rather than at the level of points, we 
should include a mechanism for this.  We're thinking of maybe a 
two-level CA, where points know about (i.e., receive information from) 
neighboring points, blocks know about neighboring blocks, and each 
block and the points within it know about each other.

One wonders if the CA experts have already played with this in some 
other domain.  To the literature!

Peter Drake
Assistant Professor of Computer Science
Lewis & Clark College
http://www.lclark.edu/~drake/

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