[computer-go] Local Search

["Robin Kramer" <robin@xxxxxxxx>]

There is a principal for search.  As the heuristic for searching for the
answer becomes more complicated, the less deep one can search.  The
benefits of life and death are that the evaluation heuristic is simple as
opposed to playing in an open section of the board.  Since the evaluation
heuristic is simple one can potentially search deeply.

For computers, is life and death necessarily the solution? Obviously
professional go players train with life and death, but I am not certain as
to if this is an actuall stratgey or a method to train themselves to store
future board positions in their mind while they search for possibilities. 
The analogy is that a baseball player uses weights to train specific
muscle groups, but weight lifting is not part of the game, and making a
machine lift weights will not likely improve the matchines efficiancy at
playing baseball.

http://sheng-lab.psych.umn.edu/pdf_files/fMRI_GO_inpress.pdf.
If you read the paper and look at the pictures, you will see that the
regions at the occipital lobe are ative, which is in the areas of visual
processing rather than the frontal cortex.

In short maybe life and death problems should be avoided in games, but as
the game progresses, and there are fewer positions to play, making the
heuristic simpler, would allow the computer to search deeper.  In the end
game the search space has been constrained, unlike the 361! it is more
like 100! or 50!

This is one of the things that computers have as an advantage in Go as
opposed to chess.  The search space actually increases in chess as the
game progresses, whereas more often than not the search space decreases in
Go.

If you believe that life and death is not in fact a necessary component of
Go, then the question becomes how does one avoid life and death situations
without actually searching?

Regards,

Robin Kramer


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