Copy Link
Add to Bookmark
Report
AIList Digest Volume 8 Issue 055
AIList Digest Friday, 19 Aug 1988 Volume 8 : Issue 55
Today's Topics:
Spang Robinson Report on Supercomputing
Computer Bridge
----------------------------------------------------------------------
Date: Wed, 17 Aug 88 10:44:30 CDT
From: smu!leff@uunet.UU.NET (Laurence Leff)
Subject: bm957
Spang Robinson rEport on Super Computing and Parallel Processing
June 1988, Volume 2, No.6
Lead article is on High Performance Networking:
Scientific Computer System announced a 1.4 gigabit per second token
net. HYPERchannel-DX and Canstar's Super-Network are 100 megabit
per second and Ultra Computer is rumored to have something competative
with Scientific Computer Systems.
______________________________________________________________________________
Next article is on "Network Computing Using Linda"
Sandia showed that 14 VAX processors were twice as powerfull as Sandia's
Cray-1 doing a rocket plume analysis. The VAXEN were over a thousands
a part. Linda coordinates the processes. Similar results
were achieved comparing a CRAY 1 with eleven VAXEN networked
on a Semiconductor application. However a
thermal analysis achieved only six percent of CRAY performance.
((((((((((((((((((((((((((((((((((((((((
Active Memory Technology has announced a 1024 processor system in
a twenty-five inch tall enclosure. The system is based on single
board chips. The DAP510 costs $120,000.
++++++++++++++++++++++++++++++++++++++++
Review ofthe Karteshev Boston Super computer Conference.
It had 3500 registrants and had Gene Amdahl, Alan Perlis,
erich Blockh and Marvin Minsky.
****************************************
SHORTS:
Network Systems Corporation had a 78 percent increase in hardware
sales, 49 percent more revenue and 250 percent increase in profit
over the previous year.
Encore has its third consecutive profitable quarter with revenues
of 9 million and profits of 649,000. They have installed 170
systems. Berita Information Systems will be installing systems
in Malaysia with a system already at New-Straits Time Press.
Stellar Computer announced it has 48 milliion invested since
inception.
Cray had a drop in revenues from 214 million to 145 million from
this first quarter to the first quarter of lastyear. Cray did
get 37 million in two orders, a X-MP/48 for Bettis Atomic
Power Lab and the Korean Advanced Institute of Science and Technology
ordered a Cray model 2.
Convex' first quarter revenues jumped to 22.1 million, double a year ago.
Net income was 1.1 million.
Multiflow delivered 14 TRACE systems this quarter with 33 sold total.
It will be concentrating on computational chemistry.
Concurrent Computer Corporation had increased revenue to 10.9 milllion
from 4.7 year with revenues going from 201.8 million to 179.2 million.
Cray Research announced X-MP Extended architecture.
It will run both the X-MP and Y-MP instruction sets and allows up to
four times as much memory. DEC has software that will allow
development of CRAY software under VMS.
Cydrome has expanded its Cydra 5 supercomputer system. The model 1205 is
priced at $495,000. It ow achieves 14.5 megaflops on the LINPACK benchmark
and 4.5 megaflop on Livermore loops.
Parasoft announced a version of dbx for parallel systems. It runs on
NCUBE, Transputer systems.
Ardent introduced a packaged system for computational chemistry.
The systems are BIOGRAF which handles proteins, nucleic acids, lipids
and carboyhydrades. POLYGRAF is designed for polymer and
material chemistry and handles amorphous or crystallnie solids.
William H. Wallace left Convex to go to Colorado Springs startup
Prisma corporation.
------------------------------
Date: 10 Aug 88 17:50:14 GMT
From: tness7!tness1!nuchat!moray!uhnix1!ceick.cs.uh.edu!ceick@bellcore
.bellcore.com (C. F. Eick)
Subject: COMPUTER BRIDGE
COMPUTER BRIDGE ??
Bridge is one of the few games that is known all over the world, which
is confirmed by the fact that usually about 50 nations participate in
Bridge world championships. The American Contract Bridge League(ACBL)
has about 200000 members.
Computer Science research was always attracted to simulate human
capabilities in computers. Writing computer programs that play "good"
Bridge is a challenging research project, because:
* Bridge requires strategical planning of a high degree of complexity:
a "good" Bridge player selects using complex criteria the plan that is most
likely to succeed in the current situation from a set of applicable plans;
sometimes, dynamic events force the player to refine his plan (for example,
if he didn't get a trick he was hoping for). That is, the game requires
planning in uncertain environments.
Furthermore, two independent players have to cooperate in defense
and bidding; that is, Bridge requires multi-agent planning.
* In Bridge, like in all other card-games, the distribution of the cards
is unknown. However, during the bidding and play additional clues become
available to locate which cards are held by which oponent. That is, a good
Bridge program has to be capable to draw inferences and make guesses
based vague and uncertain knowledge that evolves with time; that is, it
requires reasoning capabilities involving fuzzy knowledge.
* There is a large amount of expert knowledge on how to play Bridge.
Fortunately, this knowledge can be accessed easily: it is described
in about 1000 Bridge books and a even larger number of Bridge magazines. To
represent and organize this huge amount of knowledge in such a way that it
can be easily processed, changed, retrieved and refined by computers is a
very challenging knowledge engineering task.
* Bridge knowledge is usually represented in forms of rules. Therefore,
a rule based programming style seems to be very attractive when automating
the game.
* Brute force algorithms -- very popular in computer chess -- do not seem
to be suitable for card games, because of the large number of potential
hands (that additionally have a different probabilities to occur).
Therefore, different approaches have to be chosen, especially for the early
stages of the game, when the distribution of the cards is still uncertain.
In summary, research automating Bridge has to address the following topics that
are of major intrest for Artificial Intelligence Research:
reasoning and planning in uncertainty, multi-agent planning, rule based
programming paradigms, efficient search algorithms for incompletely specified
problems.
QUESTIONS:
Who is interested in the topic? What can we do to increase the popularity of
Computer Bridge? Who is currently developping programs for Computer Bridge?
Do you think Computer Bridge is a good paradigmatical example application for
AI-research. Will computers be successful in Bridge?
Development of Bridge Programs at the University of Houston
We are working on Bridge programs since 1986. Currently, 6 students are involved
in the developmemt of Bridge programs. A first prototype of a Bridge
Bidding program Cougar has been finished in March 1988. The program uses (more
or less) Kantar's version of the Standard American Bidding System. The program
is complete. It also includes defensive bidding, competive bidding, cuebidding
in Slam tries. The program uses a rule-based approach. Rule-sets are selected
to make the appropiate bid in a given context. Writing a Bridge bidding program
that can compete at club level is a quite challenging task. Some figures
will illustrate this point: The program is written in LISP and consists of
about 19000 lines of symbolic code. About 9000 lines are required by the 900
Bridge Bidding rules currently used by the program; the interpretation of
partner's and oponent's bids and hand-evaluation functions and I/O
require about 8000 lines; finally the rule-based inference
engine requires 1000 lines of code. These numbers might look frightening;
however, if we would rewrite the program -- in my opinion -- less than
12OOO lines of symbolic code would be required. Furthermore, the project is a
good test for knowledge engineering techniques and rule-based programming.
Currently, the Cougar program has already about 50% more rules than the
MYCIN system.
The program's offensive bidding part works quite well. It has already been
tested a lot and outperforms the existing commercial Bridge programs.
The program's component for defensive and competive bidding
still has a lot of bugs. It is -- because of the size of the program and the
large number of combinations -- very timeconsuming to fix these bugs. The
programs hand evaluation component works quite reasonable, however many
improvements can be made. A weakness of the current program is its incapability
make inferences from partner's and openents bids (if I have 16 HCP; each openent
has at least 10 HCP, then my partner has at most 4 HCP). Since beginning of 1988
we have also started to automate Bridge play. We are using a rule-based
approach for defensive play and a planning approach for declarer play. We hope
to finish a first prototype of a complete Bridge playing program through Spring
1989.
According to my knowledge a number of students at US universities are
writing programs automating special aspects of Bridge. However, most of these
students are working quite isolated. According to my experience, at least 3 to 4
students a needed to develop a moderately playing and bidding Bridge program. In
order to be successful, such a project needs support by the corresponding
Computer Science Departments.
In Summer 1989 there will be a Computer Games Olympiad in London. Furthermore,
very likely, an IEEE Computer Bridge Contest for students will be scheduled
for the second half of 1989.
If you have any questions concerning our projects or suggestions for
the future development of Computer Bridge. Please, let me know!
Christoph F. Eick
Assistant Professor
Department of Computer Science
University of Houston
email: ceick@ceick.cs.uh.edu
------------------------------
Date: 14 Aug 88 09:45:41 GMT
From: csli!rustcat@labrea.stanford.edu (Vallury Prabhakar)
Subject: Re: COMPUTER BRIDGE
In article <834@uhnix1.uh.edu> ceick@ceick.cs.uh.edu (C. F. Eick) writes:
#
# COMPUTER BRIDGE ??
#
[...Overview of bridge and related material deleted...]
I recall having played a few games of computer bridge at Dartmouth College
during 1986-87. A few things that I remember about this program are:
1) It was written in Basic for a Honeywell machine which had an operating
system called DCTS.
2) It could take upto 4 players. Meaning that even 1 player could
play if so desired, in which case the other team would be handled
by the computer.
3) The multi-playing capability was handled by using an inter-terminal
communication program. I don't know what or how.
4) The user-interface was not outstanding, but it did the job quite
well. There were some neat features like auto-playing, obvious
moves, and maybe even a post-game analysis (I'm not sure).
4) The program did not seem terribly intelligent in playing, but did
a reasonably good job when playing with an on-and-off not-too-good
player like me.
5) There was some talk during that time about porting it over to a Unix
machine. I left soon after that so I don't have an update.
Hopefully this will help in some way. Perhaps someone currently at
Dartmouth and familiar with this program could provide more information.
Enjoy.
-- Vallury Prabhakar
-- rustcat@csli.stanford.edu
------------------------------
End of AIList Digest
********************
