Re: Fuseki and Joseki Database with Neural Network

[Juergen Kahnert <Kahnert@xxxxxxxxxxxxxxxx>]

JG> I've spent quite some time on using GP for GO, so I understand your
JG> difficulties.

Cool. What have you done? What's the result?
I'm very interested in your experiences with GP for go, because i think,
it's the wrong way to solve the problem. GA is very good, but only for
some problems - and go is not such a problem...
But maybe i'm wrong. So, please, tell me more.

JG> If you want to use a simple function that determines whether
JG> a move is or isn't in the joseki database

Yea, but you know the problem with that...

JG> Hybridizing a neural net with Go & Gp is a good compromise, since
the ANN seems
JG> to acquire complexity a bit more efficiently than Go or Gp at this
point in
JG> time.  Other GA/GP hybrids may also be appropriate.

Maybe, i'm not sure ANN is useful for go. Ok, it would be able to know
intuitively where to place a stone or where you have to set the focus
for further calculations. But if it's wrong, you have to train it
completely(!) new.

JG> Another problem you are facing is: you need to create a joseki
individual that
JG> is exactly correct as opposed to having one that is correct on
average;
JG> otherwise, it can be exploited.  In other words, robustness.  How do
you propose
JG> to train?  How do you propose to capture sequences of moves?  How
about deep
JG> ladders which beg for loops and recursion?

That's a non trivial problem. I think to solve this problem we need
something new wether ANN nor GP seems to be the right way.

JG> Trying to synthesize joseki in a global context would be
challenging.

Yes, after finding the right way to do so...

Juergen