Copy Link
Add to Bookmark
Report
AIList Digest Volume 5 Issue 234
AIList Digest Monday, 12 Oct 1987 Volume 5 : Issue 234
Today's Topics:
Seminar - Systems with Multiple Expertise (BBN) &
Very Large Traveling Salesman Problems (Stanford) &
AI in Manufacturing (SU) &
Pengi: An Implementation of a Theory of Activity (SU) &
Persistence, Intention, and Commitment (AT&T),
Conference - AI and Statistics at JSM
----------------------------------------------------------------------
Date: Mon, 28 Sep 1987 10:34 EDT
From: Marc Vilain <MVILAIN at G.BBN.COM>
Subject: Seminar - Systems with Multiple Expertise (BBN)
[Forwarded from the IRList Digest.]
BBN Science Development Program
AI/Education Seminar Series
SYSTEMS WITH MULTIPLE EXPERTISE
Alexander Makarovitsch
Computer Science Department, West Catholic University
(Angiers, France)
BBN Laboratories Inc.
10 Moulton Street
Large Conference Room, 2nd Floor
10:30 a.m., Tuesday, September 29, 1987
Abstract: By a system with multiple expertise, we mean one containing
two or more expert systems, all working on a common information domain.
Three such systems under current development will be briefly reviewed.
(1) Syclope, a system being developed for the French Board of
Electricity for aiding agents working under risky conditions to
improve their task behavior;
(2) Sonar, a system being developed for the Bull Computer Group for
aiding in the design, operation, and maintenance of computer
networks;
(3) Link, a system being developed for the Bull Group for aiding
decision makers in the networked computer area in the processes of
product planning.
The review will focus on the difficulties encountered in key aspects of
the development process: the acquisition of expertise, knowledge
representation, man/machine interface, and interactions amopng the
system and multiple users and experts.
------------------------------
Date: Fri 2 Oct 87 15:50:59-PDT
From: Anil R. Gangolli <GANGOLLI@Sushi.Stanford.EDU>
Subject: Seminar - Very Large Traveling Salesman Problems (Stanford)
15-October-87: David Johnson (AT&T Bell Labs)
Near-Optimal Solutions to
Very Large Traveling Salesman Problems
Most experimental studies of algorithms for the Travel-
ing Salesman Problem (TSP) have concentrated on relatively
small test cases, instances with 100 cities or less. In
practice, however, much larger instances are frequently
encountered, both in engineering and scientific applica-
tions. This talk begins by surveying complexity results
about the TSP and the status of algorithms for finding
optimal solutions to small instances. It then goes on to
report the results of experiments with standard TSP heuris-
tics on large instances, from 500 cities to 100,000, examin-
ing the trade-offs obtainable between running time and qual-
ity of solution. Most of the standard heuristics will be
compared, including such new approaches as ``simulated
annealing,'' but particular emphasis will be placed on the
acknowledged ``champion,'' the sophisticated Lin-Kernighan
algorithm. Using various programming tricks, we have imple-
mented a version of this heuristic for the Euclidean case
that remains feasible even for 10,000 city instances (8
hours on a minicomputer), and continues to find tours that
are within 2% of optimal. For 20,000 or more cities, we
could still obtain tours that were within 5% of optimal
using Lin-Kernighan as a subroutine in a partitioning scheme
suggested by Karp. If one is willing to settle for slightly
worse tours, an approximate version of the Christofides
heuristic seems to stay within 20% of optimal and has quite
acceptable running times even for 100,000 cities.
------------------------------
Date: Fri, 9 Oct 1987 15:16 PDT
From: Marty Tenenbaum <TENENBAUM@SPAR-20.ARPA>
Subject: Seminar - AI in Manufacturing (SU)
FIRST-CUT: A Knowledge-based CAD/CAM System
for Concurrent Product and Process Design
Prof. Mark Cutkosky (ME)
Friday, Oct. 16 at 3:30 pm.
Terman 556
Abstract: FIRST-CUT is a novel CAD/CAM system for rapid prototyping of
mechanical parts. Parts are designed interactively by graphically
composing a high-level plan for producing them. A plan consists of
abstract machining operations, such as "make hole" or "make pocket".
As each operation is added, expert systems check feasibility and fill
in details (e.g., whether a hole should be drilled, milled, or bored.)
Also, a solid modeler incrementally simulates the plan to detect
geometric interference problems and to enable the designer to
visualize the part taking shape. Completed plans are compiled into
NC-code and run on a table-top milling machine to physically
instantiate the design.
A second part of the project is concerned with monitoring the actual
execution of process plans on specially instrumented machine tools,
and using the results to refine the knowledge base and produce better
plans.
A live demonstration will follow the talk.
Students seeking an exciting real-world domain for AI research (in
areas such as planning, spatial reasoning, knowledge-acquisition and
learning, intelligent agents, signal understanding and man-machine
communication) are especially invited.
------------------------------
Date: Fri 9 Oct 87 15:47:14-PDT
From: Anne Richardson <RICHARDSON@Score.Stanford.EDU>
Subject: Seminar - Pengi: An Implementation of a Theory of Activity
(SU)
Daniel Weise is hosting Phil Agre here at Stanford on Tuesday, October 27
who will be giving the following talk in Bldg. 200, Rm. 303 at 4:15:
***For any questions, please contact Daniel@mojave.***
Pengi: An implementation of a theory of activity
Phil Agre
MIT Artificial Intelligence Laboratory
AI has typically sought to understand the organized nature of human activity
in terms of the making and execution of plans. There can be no doubt that
people use plans. But before and beneath any plan-use is a continual
process of moment-to-moment improvisation. An improvising agent might use a
plan as one of its resources, just as it might use a map, the materials on a
kitchen counter, or a string tied round its finger. David Chapman and I
have been studying the organization of the most common sort of activity, the
everyday, ordinary, routine, familiar, practiced, unproblematic activity
typified by activities like making breakfast, driving to work, and stuffing
envelopes. Our theory describes the central role of improvisation and the
inherent orderliness, coherence, and laws of change of improvised activity.
The organization of everyday routine activity makes strong suggestions about
the organization of the human cognitive architecture. In particular, one can
get along remarkably well with a peripheral system much as described by Marr
and Ullman and a central system made of combinational logic. Chapman has
built a system with such an architecture. Called Pengi, it plays a
commercial video game called Pengo, in which a player controls a penguin to
defend itself against ornery and unpredictable bees. The game requires both
moderately complex tactics and constant attention to opportunities and
contingencies. I will outline our theory of activity, describe the Pengi
program, and indicate the directions of ongoing further research.
------------------------------
Date: Thu 8 Oct 1987 12:40:14
From: dlm@allegra.att.com
Subject: Seminar - Persistence, Intention, and Commitment (AT&T)
Title: Persistence, Intention, and Commitment
Speaker: Philip R. Cohen
Affiliation: SRI International Menlo Park, CA
Place: AT&T Bell Laboratories Murray Hill 3D-473
Date: October 8, 1987 1:30 PM.
(work done jointly with Hector Levesque)
Abstract:
This talk explores principles governing the rational balance among an
agent's beliefs, goals, actions, and intentions. Such principles provide
specifications for artificial agents, and approximate a theory of human
action (as philosophers use the term). By making explicit the conditions
under which an agent can drop his goals, i.e., by specifying how the
agent is _committed_ to his goals, the formalism captures a number of
important properties of intention. Specifically, the formalism provides
analyses for Bratman's three characteristic functional roles played
by intentions, and shows how agents can avoid intending all the foreseen
side-effects of what they actually intend. Finally, the analysis shows
how intentions can be adopted relative to a background of relevant beliefs
and other intentions or goals. By relativizing one agent's intentions in
terms of beliefs about another agent's intentions (or beliefs), we
derive a preliminary account of interpersonal commitments.
------------------------------
Date: 7 Oct 87 12:26:16 GMT
From: ihnp4!erc3bb!may@ucbvax.Berkeley.EDU (M.A.Yousry)
Subject: Conference - AI and Statistics at JSM
At the August 22-25, 1988 Joint Statistical Meetings (American
Statistical Association, Biometric Society, Institute of Mathematical
Statistics) in New Orleans, I'll be chairing an invited session, titled
"Bridging the Gap, Artificial Intelligence and Statistics," on solving
problems using combinations of AI and statistical techniques.
Both applied and theoretically oriented papers will be considered.
Potential areas might include fault diagnosis, process control,
reasoning with uncertainty, ... If you are interested in giving a talk,
please send a short abstract to:
...!{ihnp4, allegra}!erc780!may
or
Mona Yousry, (609) 639-2405
AT&T - ERC
PO Box 900
Princeton, NJ 08540
------------------------------
End of AIList Digest
********************
