Re: Fuseki and Joseki Database with Neural Network

[David Mechner <mechner@xxxxxxxxxxxxxxxxx>]

Juergen Kahnert wrote:
> But by using GA, we approximate to the maxima or a local maxima. 
> GA find this quickly while vary the step width - in the beginning 
> big steps and near a maxima reducing it.
> We have to use a step width of one and have no longer benefits 
> from GA. Thats my understanding of GA or am i wrong?

and

> I can't follow you, sorry.
> To get the fitness of a generation I need a continuous 
> function. ;-) [...]
> I can't approximate step by step to the maxima while 
> having "jumps" in

I think you're confused about how GA works. There is no "step size". It
sounds like you're thinking of some gradient descent search technique
like simulated annealing. There are lots of sites on the web that
describe GA, GP, ANNs, etc., check them out.  Also, none of these ML
techniques require a very smooth (let alone continuous) error surface -
if they did, they would be useless in all but the simplest toy domains. 

> i can't imagine the improvement of this knowledge - it's not 
> really new. So, why not take this move directly out of the 
> database? Sorry, i couldn't find a way to make a use of it.
> How to find a good alternative sequence - which is not a part 
> of the database - if all move not contained in the database 
> are bad?

The point of using joseki as a domain for machine learning research
would (or should) not be to store precisely the josekis you have in your
dictionary, but to GENERALIZE; to find and represent patterns or
regularities in the domain (in the josekis you start with) that will
allow the system to play good (joseki-like) moves in novel situations.
Actually, it's not clear that joseki is a good domain for this kind of
research because it's not much simpler than general go - to evaluate a
joseki move you have to know about territory, influence, tactical
fighting, life and death, etc. 

-David

-- 
David A. Mechner            Center for Neural Science
mechner@xxxxxxxxxxxxxxxxx         4 Washington Place, New York, NY 10003
212.998.3580                http://cns.nyu.edu/~mechner/