Re: [computer-go] Fractals, was Search = Bad!

[Robin Kramer <robin@xxxxxxxx>]

My AI text book calls that a search, in its very broad definition of the 
term.

Searching backwards resembles "End Game Table Bases" EGTB in chess.  
Which I can only summarize as, for all the combinations of pieces on the 
board at the end of a chess game(King and a queen vs, a king and two 
Kinghts a sophisticated program will have as many as all 7 peiece 
combinations at its disposal),  and a garunteed shortest sequence of 
moves for the winner.   So during the middle game the chess program 
tries to convert the board to one of the board states which it has 
calculated to be a win for its side, and avoid the certain losses.  The 
table base data base is constructed before the game.  I am not sure how 
the program goes about making the two meet in the middle.  Is there 
anyone who has more experience with this?

EGTB is esspecially useful in chess, because of the way the search space 
narrows near the end of the game.  I believe some algorithm such as this 
EGTB will be able to be employed in Go as well, because of the fact that 
the number of possible moves in Go narrows as well, although the 
configuration space of the go end game is probably larger, so the table 
will probably have to be more sophisticated.

Cheers,

Robin

Frank de Groot wrote:

> ----- Original Message ----- From: "David G Doshay" <ddoshay@xxxxxxxxxxxxxxxxx>
>
> > So, all in all, I doubt that a program that "just jumps the
> > right way in the fractal space" will be written any time soon.
> > But some of the ideas close to that one are still interesting.
>
> Yes.
>
> Like, catapult a bunch of autonomous agents a bit further away in 
> Fractal space and let them find their way back.
> The ones that come back (have not been killed along the way) have 
> something important to tell about the fractal space further on. Like, 
> how to get there without getting killed (they made it back the same 
> path..)
>
> So you don't *search*, you reverse the process, making it converge 
> from an x-number of "tries" back to 1, instead of from 1 to just about 
> infinite.
> _______________________________________________
> 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/