Copy Link
Add to Bookmark
Report
AIList Digest Volume 4 Issue 023
AIList Digest Wednesday, 12 Feb 1986 Volume 4 : Issue 23
Today's Topics:
Seminars - Systems of Actors (USC) &
Artificial Concept Formation (Edinburgh) &
Parallelism in Production Systems (SU) &
A Storage Manager for Prolog (SU) &
Statistical Theory of Evidence (SRI),
Conference - Compcon Spring 86
----------------------------------------------------------------------
Date: 7 Feb 1986 08:21-EST
From: gasser@usc-cse.usc.edu
Subject: Seminar - Systems of Actors (USC)
USC DISTRIBUTED PROBLEM SOLVING GROUP
MEETING
"Formalizing the Development of
Systems of Actors"
Ed Ipser
Ph.D Student, USC
A formalization of the process of specifying and developing
distributed systems is presented, with the emphasis on the description
of multiple robot environments. The general scheme is a recursive
reduction of behaviors with constraints to actors with pre-determined
behaviors by showing that the behaviors of the actors satisfy the
behavior and constraint requirements of the system. Possible
applications of this scheme are presented, including automatic
programming, planning, theorem proving, and the description of
non-computable functions. This work is based on the work of Goldman
and Wile on GIST, and Georgeff's work on the theory of processes.
Time: 3:00 PM Wednesday, Feb 12, 1986
Place: Seaver Science Bldg., Room 319, USC
Questions: Dr. Les Gasser, CS Dept., USC (213) 743-7794
------------------------------
Date: Mon, 10 Feb 86 14:56:48 GMT
From: Gideon Sahar <gideon%edai.edinburgh.ac.uk@cs.ucl.ac.uk>
Subject: Seminar - Artificial Concept Formation (Edinburgh)
EDINBURGH AI SEMINARS
Date: Wednesday, 12th February l986
Time: 2.00 p.m.
Place: Department of Artificial Intelligence
Seminar Room - F10
80 South Bridge
EDINBURGH.
Professor Donald Michie, The Turing Institute, Glasgow will give a
seminar entitled - "Artificial Concept Formation".
The approach develops from a position taken in the 1950's by H.A. Simon.
He proposed, in essence, a new criterion for the adequacy of a theory
(he considered economic theory), namely that in explaining the flux of
transactions a theory must take full account of the resource-limited
nature of the calculations performed by the participating agents. Is
economic man rational in the sense of making fully rational choices
whatever the computational cost (as in the von Neumann and Morgenstern
theory of economic behaviour), or does he exhibit at most the level of
rationality which human brains can feasibly compute in the time
available for each choice? By implication Simon also requires that
such a theory should be feasibly interpretable by its human user:
runnability on the machine is not enough.
This leads to the idea that what is run on the machine should be
human-oriented in a very strong sense, unprecedented in conventional
software technology even as an aspiration: if a program is to be not
just an operationally effective description or prescription, but a
machine representation of a concept and hence an eligible component of
a Simon-type theory, it must be not only human-intelligible but also
human-interpretable. This entails that the human expert skilled in
the given area must be able mentally to check it against trial data in
his head, just as he can in the case of his own professionally acquired
concepts.
------------------------------
Date: Mon 10 Feb 86 09:28:13-PST
From: Sharon Gerlach <CSL.GERLACH@SU-SIERRA.ARPA>
Subject: Seminar - Parallelism in Production Systems (SU)
On Friday, Feb 21, Anoop Gupta, a CSL faculty candidate from CMU, will
be speaking on "Parallelism in Production Systems" in MJH 352 at 3:15.
Parallelism in Production Systems
Anoop Gupta
Department of Computer Science
Carnegie-Mellon University
Pittsburgh, PA 15213
Production systems (or rule-based systems) are widely used in Artificial
Intelligence for modeling intelligent behavior and building expert systems.
Most production system programs, however, are extremely computation intensive
and run quite slowly. The slow speed of execution has prohibited the use of
production systems in domains requiring high performance and real-time
response. The talk will elaborate on the role of parallelism in the high-speed
execution of production systems.
On the surface, production system programs appear to be capable of using
large amounts of parallelism -- it is possible to perform match for each
production in a program in parallel. Our research shows that in practice,
however, the speed-up obtainable from parallelism is quite limited, around
10-fold as compared to initial expectations of 100-fold to 1000-fold. The main
reasons for the limited speed-up are: (1) there are only a small number of
productions that are affected (require significant processing) as a result of a
change to working memory and (2) there is a large variation in the processing
requirement of these productions. Since the number of affected productions is
not controlled by the implementor of the production system interpreter (it is
governed mainly by the author of the program and the nature of the problem),
the solution to the problem of limited speed-up is to somehow decrease the
variation in the processing cost of affected productions. We propose a
parallel version of the Rete algorithm which exploits parallelism at a very
fine grain to reduce this variation. We further suggest that to exploit the
fine-grained parallelism, a shared-memory multiprocessor with 32-64 high
performance processors should be used. For scheduling the fine-grained tasks
consisting of about 50-100 instructions, a hardware task scheduler is proposed.
The results presented in the talk are based on simulations done for a large
set of production systems exploiting different sources of parallelism. The
simulation results show that using the suggested multiprocessor architecture
(with individual processors performing at 2 MIPS), it is possible to obtain
execution speeds of 5000-27000 working memory element changes per second. This
corresponds to a speed-up of 5-fold to 27-fold over the best known sequential
implementation using a 2 MIPS processor. This performance is also higher than
that obtained by other proposed parallel implementations of production systems.
------------------------------
Date: Tue 11 Feb 86 16:30:05-PST
From: Karin Scholz <SCHOLZ@SU-SUSHI.ARPA>
Subject: Seminar - A Storage Manager for Prolog (SU)
this is a correction to the colloquium notice for this week:
Database Seminar CS 545, Friday Feb 14, 3:15pm, mjh352
Persistent Prolog: A Secondary Storage Manager for Prolog
Peter M D Gray
University of Aberdeen, Scotland
ABSTRACT OF TALK
The talk will describe a general purpose "tight coupling" system based on a
C-Prolog interpreter interfaced to a "Persistent Heap" database, which
can store a wide variety of data types and objects. We are
currently extending Prolog to allow definitions of modules and Abstract
Data Types. This provides a disciplined way of accessing frame structures,
bit maps, attached procedures and other non-Prolog objects.
With this system we are able to use Prolog to maintain an evolving
knowledge base on disc. Prolog clauses and data structures are
manipulated in memory in the usual way, but migrate to disc on a
"commit" step.
This work is part of the U.K. "Alvey" program in IKBS
------------------------------
Date: Wed 12 Feb 86 08:55:19-PST
From: FIRSCHEIN@SRI-AI.ARPA
Subject: Seminar - Statistical Theory of Evidence (SRI)
Bob Hummel will be giving a talk on Tuesday, Feb. 18 at 10:30,
Conf room EK242 (the "old" conf room). An abstract of his talk follows:
A Statistical Viewpoint on the Theory of Evidence
Robert Hummel
Courant Institute, New York University
Abstract
The Dempster/Shafer "Theory of Evidence" can be regarded as an alge-
braic space with a combination formula that combines the opinions of
"experts". This viewpoint, which is really the origin of the theory, will
be explained by introducing spaces with simple binary operations, giving
these spaces intuitive interpretations, relating them to Bayesian updating,
and showing that the spaces are (in a homomorphic sense) equivalent to the
Dempster/Shafer theory of evidence space.
The viewpoint allows us to remark on limitations of the theory. By
making compromises in a different manner, an alternative combination method
can be introduced. This representation of states of belief by "Parameter-
ized Statistics of Experts" will be described.
------------------------------
Date: Fri, 31 Jan 86 07:40:54 pst
From: Doug Coffland <coffland@lll-crg.ARPA>
Subject: Conference - Compcon Spring 86
Register for Compcon Spring 86 now and attend the year's
only broad based computing conference sponsored by the
IEEE Computer Society. Compcon will be held in San Fran-
cisco, March 3-6, 1986.
Key topics include: supercomputers, SDI software reliability,
AI applications, Japanese software practices, RISC vs. CISC,
and more. Four full day tutorials will be given on Monday,
March 3. Topics include silicon compilation, issues in expert
systems, complex computer graphics, and high performance com-
puting.
The advanced registration deadline is February 14. For further
information, contact Robert M. Long, Lawrence Livermore National
Labratory, P. O. Box 808, MS L130, Livermore, Ca. 94550.
The telephone number is 415-422-8934. Telephone registrations
will be accepted with Visa, MasterCard, or American Express.
------------------------------
End of AIList Digest
********************
