Deep Blue??? PLEASE
Jason M. Rotenberg
project? Is there any other similar projects taking place? Since the
match in 1989, not much has been cooking in computer that play slow
chess. When are we going to see games/matches between Deep Blue and
leading Grandmasters? have any been taking place? Is Deep Blue almost
up to 1 billion moves a second? What about its estimated rating? Is
this estimate (2700 I hear) based on games? Are they keeping Deep Blue
in the closet? Is it just faster than Deep thought or are they using an
improved algorithm? Do I need to make an outrageous opinioneted remark
about the subject to get some answers (like the Fisher treads)? Answers
PLEASE!
Geoffrey P. Wyatt
: Can anyone write a littlee bit about the current state of the Deep Blue
: project? Is there any other similar projects taking place? Since the
: match in 1989, not much has been cooking in computer that play slow
: chess. When are we going to see games/matches between Deep Blue and
: leading Grandmasters? have any been taking place? Is Deep Blue almost
: up to 1 billion moves a second? What about its estimated rating? Is
: this estimate (2700 I hear) based on games? Are they keeping Deep Blue
: in the closet?
Deep Blue almost certainly does not play slow chess at a 2700
FIDE level. Programmers continue to underestimate the difficulty of
creating a machine that could actually win a match from *any* super-GM.
In the case of Deep Blue, I don't believe 1 billion moves a second has
been achieved. But I think the real problem is that the Deep Blue team
doesn't want to see its baby be butchered. So we wait, and wait, and...
--
geo...@teleport.COM Public Access User --- Not affiliated with TECHbooks
Public Access UNIX and Internet at (503) 220-1016 (2400-14400, N81)
Stefan Hahndel
In article <39aqa6$k...@elaine.teleport.com>, geo...@teleport.com (Geoffrey P.
Wyatt) writes:
...
|> FIDE level. Programmers continue to underestimate the difficulty of
|> creating a machine that could actually win a match from *any* super-GM.
|> In the case of Deep Blue, I don't believe 1 billion moves a second has
|> been achieved. But I think the real problem is that the Deep Blue team
I agree. I think to obtain big speedup values in a parallel tree search is
much more difficult than most programmers think. Now, we have very good
search algorithms for tree search (improvement of alpha-alpha beta as PVS,
NegaScout, etc.). But, all these efficient tree search algorithms go through
the search tree in sequence node by node. Observe on of the best commercial
chess programs as Fritz, King, etc. When they search the n-th ply within
the iterative deepening, the spend most of the time to determine the
correct minimax value of the principal variation. Then, with this information,
they are pruning very big parts of the rest of the search tree.
But, if you do parallel tree search the information for a cut in the search
tree often is not available. When the search process in one subtree determines
a new [alpha,beta]-window, this information must be communicated to other
search processes. So, there is a big communication overhead. But, even if
such information becomes available in other search processes searching other
subtrees it't often to late to obtain the best cuts. For this reasons, a
parallel approach must visit much more nodes to reach ply n than other
approaches. Of, course they are visiting much more nodes within a time intervall.
I assume, that many work already has been done in parallel tree search.
Are there any good recent publications about the best speedups that can
be obtained by parallel tree search.
Stefan Hahnde;
--
Office: Technische Universitaet Muenchen Phone: +49-89-48095-210
Institut fuer Informatik Fax: +49-89-48095-203
Orleansstr.34 Telex: 3457 tumue d
D-81667 Muenchen, Germany Room: 236
Email: hah...@informatik.tu-muenchen.de [نؤِضüـك]
PGP public key available on request !