RE: [computer-go] Addressing the root of the problem

["Daniel Burgos" <Daniel_Burgos@xxxxxxxxx>]

I think my approach is the same as yours.

I'm not using neural networks, just patterns. I see them more easier
(and faster) to evolve.

The algorithm I use to discover whether a set of rules is worth does not
know anything about Go. It just makes two set of rules compete and the
best is taken for generating new aspirants, a very usual approach.

I would like to see something interesting emerge of this experiment.
But, who knows...

Daniel

-----Original Message-----
From: computer-go-bounces@xxxxxxxxxxxxxxxxx
[mailto:computer-go-bounces@xxxxxxxxxxxxxxxxx] On Behalf Of Anton Safonov
Sent: Thursday, October 07, 2004 2:36 PM
To: computer-go@xxxxxxxxxxxxxxxxx
Subject: RE: [computer-go] Addressing the root of the problem

>
> From:
> "Daniel Burgos" <Daniel_Burgos@xxxxxxxxxxxxxxxxx>
> Date:
> Thu, 7 Oct 2004 13:53:17 +0200
> To:
> "computer-go" <computer-go@xxxxxxxxxxxxxxxxx>
>
>
>
>I'm following that approach to solve Go: find a small set of
>rules/patterns (as small as possible) to evaluate a board in order to
>decide where is worth to play. I want to cut down the decision tree.
>
>I think my hypothesis is the same as yours: because the complexity of
go
>is coming from its small set of rules (but very well designed), its
>solution should be something similar.
>
>Something inspired in the small rules that allow creating complex
>figures as the fractals.
>  
>
I guess, it could be some mix of cellular automata and neural
networks...

>But, what is this set of rules? This is the winning answer, if our
>hypothesis are correct I'm afraid...
>
>I'm implementing evolution algorithms for that, but they are only in
>design stage by now.
>  
>
I tried to establish some rules which will grow the needed "fractal", 
but suddenly came to the conclusion, that all my constructions repesent 
a less powerfull system.
Now trying to design a meta-system with even less rules (one could be 
tracking that produced end-system must be "at the edge of chaos"), which

is given a set of building blocks (for rules) and evolves the rules, 
which in turn will grow the end-system.
I.e. "meta-system" --(evolution)->  "rules" --(grow)-> "solution system"

Anton.


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