Re: Fuseki and Joseki Database with Neural Network

[Juergen Kahnert <Kahnert@xxxxxxxxxxxxxxxx>]

JG> Some earlier work I did is on my web pages at:
JG>     www.cris.com/~jgberg

The last time i have a look the pages are unavailable yet. :-(

JG> The gist of it is that I tried to see how the combination of GP
solving
JG> power and GO complexity would play against each other. [...]

Sounds interesting. Do you want to enhance your GP language and try it
again?

JG> My goal in all of this is to find a way to develop software without
requiring
JG> the programmer to be so much of a domain expert.

Me too. The problem is, how to represent the knowledge and how to let it
learn...?
I think it had to be a combination of pattern matching and some kind of
learning algorithm.

For example: The programme plays and lose a lot of stones because of
running into a ladder.
Now, it had to go back step by step and analyse why it could happen and
find the starting pattern. Now this pattern is marked as bad. But all of
us know, it's not generally bad. So, in some later games it use this
'bad' pattern to catch a lot of enemy stones but now the enemy did not
lose many stones... on the contrary the programme lose itself many
stones.
In this situation it had to look back and try to find out why this could
happen. It must be able to realize that stones in the way of the ladder
would change the effect. To do so, it must try out some variations to
find the reason when it works and not works.

Assumed we have such an algorithm, how would we able to save the
received knowledge?

	Juergen