Re: [computer-go] question regarding Hydra Chess Supercomputer

[Don Dailey <drd@xxxxxxx>]

Hi John,

I wasn't even  thinking primarily of evaluation, I  was thinking about
hash table implementation.

Do you believe that Deep Blue was able to utilize global transposition
tables  throughout the  whole search?   I believe  it would  have been
possible, but stupid to do this  in hardware.  I have to assume it was
a  better engineering  decision to  either not  use  any transposition
tables in  hardware, or use tiny  local ones which  wouldn't have been
nearly as good as global ones.

I agree it's POSSIBLE to have a superior evaluation in hardware due to
the fact that you could probably  do something that would take far too
long in software.   But actually being able to pull  this off ...  I'm
not a believer that Deep Blue actually did.    

Deep Blue used an automated method to tune evaluation parameters based
on master games which many people think  is not a good way to build an
evaluation function.  The argument that  Bob made was that if you have
thousands of evaluation  features compared to the few  that PC's have,
you  MUST have  a far  superior evaluation.   That's not  the  same as
saying it's possible.  Of course I know YOU are not saying this, but I
still  seriously  doubt Deep  Blue  had  a  "far superior"  evaluation
function to  those tweaked  and tuned  over many years  by the  top PC
authors and masters.

- Don




   Date: Sun, 24 Oct 2004 21:21:34 +0200
   From: John Tromp <John.Tromp@xxxxxxxxxxxxxxxxx>


   Don Dailey wrote:

   > Bob was  so strong in  his opinion  that I just  had to shut  up.  His
   > conclusion was  that Deep Blue  had no compromises whatsoever  in it's
   > search and was  in fact more efficient and that  the evaluation was of
   > much higher quality than the PC programs.  I still don't believe this.

   I remember Feng Hsiung Hsu discussing this point in his book "Behind Deep Blue"
   (see http://www.chessbase.com/newsdetail.asp?newsid=551 for a review)

   Before reading that book, I also held the believe that the chip's evaluation
   had to be limited in its sophistication. But I found Feng Hsiung to be quite
   convincing at arguing the opposite. It's a shame he never got to develop
   the chip further after the famous matches...

   regards,
   -John

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