Copy Link
Add to Bookmark
Report
AIList Digest Volume 6 Issue 004
AIList Digest Saturday, 9 Jan 1988 Volume 6 : Issue 4
Today's Topics:
Seminars - Recovery From Incorrect Knowledge In SOAR (GMR) &
Open-Ended Learning Through Machine Evolution (Siemens),
Conference - 2nd Workshop on Qualitative Physics &
Neural Controls Session at ACC
----------------------------------------------------------------------
Date: Mon, 4 Jan 88 11:38 EST
From: "R. Uthurusamy" <SAMY%gmr.com@RELAY.CS.NET>
Subject: Seminar - Recovery From Incorrect Knowledge In SOAR (GMR)
Seminar at the General Motors Research Laboratories in Warren, Michigan.
Wednesday, January 20, 1988 at 10 a.m.
RECOVERY FROM INCORRECT KNOWLEDGE IN SOAR
JOHN E. LAIRD
Assistant Professor, Electrical Engineering and Computer Science Dept.
The University of Michigan
ABSTRACT:
In previous work, we have demonstrated some of the generality of Soar's
problem solving and learning capabilities. We even gone so far as to
hypothesize that the simple learning mechanism in Soar, chunking, combined
with its general problem solving capabilities, is sufficient for all
cognitive learning. This is a radical hypothesis especially when we
consider Soar's difficulty with recovery from incorrect knowledge.
Soar acquires incorrect knowledge whenever it chunks over invalid
inductive inferences made during problem solving. Recovery requires
some form of identification and correction of the incorrect knowledge.
Recovery is complicated in Soar by the fact that we have made the following
assumptions: chunking is the only learning mechanism; long-term knowledge,
represented as production rules, is only added, never forgotten, modified
or replaced; and the productions are not open for direct examination by the
learning mechanism or the problem solver.
In this talk I will review chunking in Soar and present recent results in
developing a domain-independent approach for the recovery from incorrect
knowledge in Soar. This approach does not require any change to the Soar
architecture, but uses chunking to learn rules that overcome the incorrect
knowledge. The key is to use the problem solving to deliberately reconsider
decisions that might be in error. If a decision is found to be incorrect,
the problem solving corrects it and a new chunk is learned that will correct
the decision in the future.
Non-GMR personnel interested in attending this seminar please contact
R. Uthurusamy [ samy@gmr.com ] 313-986-1989
------------------------------
Date: 7 Jan 88 00:39:29 GMT
From: siemens!hudak@princeton.edu (Michael J. Hudak)
Subject: Seminar - Open-Ended Learning Through Machine Evolution
(Siemens)
Speaker: Peter Cariani
Systems Science Dept., Thomas J. Watson School of Engineering
State University of New York at Binghamton
Title: Structural Preconditions for Open-Ended Learning
through Machine Evolution
Location: Siemens Corporate Research & Support, Inc.
3rd floor Multi-Purpose Room
Princeton Forrestal Center
105 College Road East
Princeton, NJ 08540-6668
Date: Thursday, 14 January 1988
Time: 10:00 am (refreshments: 9:45)
For more information call Mike Hudak: 609/734-3373
Abstract
One of the basic problems confronting artificial life simulations is
the apparent open-ended nature of structural evolution, classically known
as the problem of emergence. Were it possible to construct devices with
open-ended behaviors and capabilities, fundamentally new learning tech-
nologies would become possible. At present, none of our devices or models
are open-ended, due to the nature of their design and construction.
The best devices we have, in the form of trainable machines, neural net
simulations, Boltzmann machines and Holland-type adaptive machines,
exhibit learning within the categories fixed by their feature spaces.
Learning occurs through the performance dependent optimization of alter-
native I/O functions. Within the adaptive machine paradigm of these
devices, the measuring devices, feature spaces, and hence the real world
semantics of such devices are stable. Such machines cannot create new
primitive categories; they will not expand their feature and behavior
spaces.
Over phylogenetic time spans, however, organisms have evolved new sensors
and effectors, allowing them to perceive more and more aspects of their
environments and to act in more and more ways upon those environments.
This involves a whole new level of learning: the learning of new primitive
cognitive and behavioral categories. In terms of constructible devices,
this level of learning encompasses machines which construct and select
their own sensors and effectors, based upon their real world performance.
The semantics of the feature and behavior spaces of such devices thus
changes so as to optimize their effectiveness as categories of perception
and action. Such devices construct their own primitive categories, their
own primitive concepts. Evolutionary devices could be combined with
adaptive ones to both optimize primitive categories and I/O mappings
within those categories.
Evolutionary machines cannot be constructed through computations alone.
New primitive category construction necessitates that new physical
measuring structures and controls come into being. While the behavior of
such devices can be represented to a limited degree by formal models,
those models cannot themselves create new categories vis-a-vis the real
world, and hence are insufficient as category-creating devices in their
own right. Computations must be augmented by the physical construction
of new sensors and effectors implementing processes of measurement and
control respectively. This construction process must be inheritable and
replicable, hence encodable into symbolic form, yet involving the autono-
mous, unencoded dynamics of the matter itself.
The paradigmatic example of a natural construction process is protein
folding. A one-dimensional string of nucleotides, itself a discrete,
rate-independent symbolic structure, is transformed into continuous, rate
dependent dynamics having biological function through the action of the
physical properties inherent in the protein chain itself. The functional
properties of speed, specificity, and reliability of action are thus
achieved with symbolic constraints but without the explicit direction of
rules.
------------------------------
Date: Tue, 5 Jan 88 15:50:05 CST
From: forbus@p.cs.uiuc.edu (Kenneth Forbus)
Subject: Conference - 2nd Workshop on Qualitative Physics
CALL FOR PARTICIPATION
SECOND WORKSHOP ON QUALITATIVE PHYSICS
PARIS, JULY 26-28, 1988
Following last year's success of the first workshop on Qualitative
Physics organized by the Qualitative Reasoning Group at the University
of Illinois (with AAAI sponsorship), the second workshop on
Qualitative Physics will be organized by the European Group on
Qualitative Physics and the IBM Paris Scientific Center. The
workshop, sponsored by the Commission of the European Community
(JRC-Ispara) and in cooperation with AAAI, will be held in Paris on
July 26-28, 1988. It is intended as a forum for discussion of
ongoing research in Qualitative Physics and related areas.
To develop interaction and exchange of ideas, a number of panels will
be organized. We invite proposals for panels on ongoing debates in
the area, such as:
-- Causal Reasoning
-- Mathematical Aspects of Qualitative Models
-- Naive Physics versus Qualitative Physics
Another suggested panel format is to pose a particular problem which
panelists must use to focus discussion. Proposers for panels should
obtain the agreement of the panelists and submit the proposal,
including an outline of the suggested discussion, to the program
chairman by March 8, 1988.
ATTENDANCE:
To encourage lively discussion, attendance will be by invitation only.
If you are interested in attending, please submit five (5) copies of
an extended abstract, up to 6 pages long, to the program chairman:
Francesco Gardin
Dipartimento di Scienze dell'Informazione,
Universita degli Studi di Milano
Via Moretto da Bresica, 9
20133 Milano, ITALY
tel. +39-2-2141230
The deadline for submissions is MARCH 8th, 1988 and invitations will be
mailed APRIL 5th, 1988. Abstracts will be reviewed by an international
scientific committee. Results already submitted for publication elsewhere
are acceptable since no proceedings of the workshop will be published.
A subset of the authors may be asked to contribute to a book based on the
workshop. Besides presenters of papers, a limited number of observers
may be accepted. For further information about the organization of the
workshop, contact any member of the organizing committee, or:
Olivier Raiman
IBM Paris Scientific Center
3/5 Place Vendome,
75001 Paris, FRANCE
tel. +33-1-4296-1475
====================
ORGANIZING COMMITTEE
Johan De Kleer (Xerox PARC)
Ken Forbus (University of Illinois, Urbana)
Pat Hayes (Xerox PARC),
Ben Kuipers (University of Texas, Austin)
and all the members of the European Qualitative Physics Committee:
Flavio Argentesi (JRC-Ispra)
Ivan Bratko (University of Ljubljana)
John Campbell (Univ. College of London)
Jean-Luc Dormoy (EDF)
Boi Faltings (E.P.F. Lausanne)
Francesco Gardin (University of Milan)
Bernd Hellingrath (Fraunhofer-Institute ITW)
Roy Leitch (Heriot-Watt University)
Nicools J. Mars (Univ. of Twente)
Pierre Van Nypelseer (AITECH, Brussels)
Olivier Raiman (IBM Paris Scientific Centre)
Peter Struss (Siemens)
------------------------------
Date: Fri, 8 Jan 88 07:27 PST
From: nesliwa%nasamail@ames.arc.nasa.gov (NANCY E. SLIWA)
Subject: Conference - Neural Controls Session at ACC
In response to requests for information about the ACC session in
neural applications to robotics, about which I recently solicited names,
I am posting the current status of the session, along with minimal
conference information. Registration information can probably be obtained
from the general chair.
1988 American Controls Conference
June 15-17, 1988
The Atlanta Hilton and Towers
Atlanta, Georgia
General Chair: Wayne Book
The George W. Woodruff School of Mechanical Engineering
Georgia Institute of Technology
Atlanta, Georgia 30332
(404) 894-3247
Invited Session on Neural Networks in Control
(A 4-hour session, 8 regular papers)
Chairs: Moshe Kam, Drexel University
Don Soloway, NASA Langley Research Center
"How Neural Networks Factor Problems of Sensory Motor Control"
Danial Bullock, Boston University
"Neural and Adaptive Control: Similarities and Differences"
A. Sideris, D. Psaltis, A. Yamakura, California Inst. of Technology
"On State Space Analysis for Neural Networks"
Moshe Kam, Roger Cheng, Allon Guez, Drexel University
"Adaptive Neural Model for Hand-Eye Coordination"
M. Kuperstein, Wellesley College
"A Neural Network for Planning Preshape Postures of the Human Hand"
Thea Ibarall, University of Southern California
"Strategy Learning with Multilayer Connectionist Representations"
Charles Anderson, GTE Labs Inc.
"Neural-Networks-Based Learning Systems for Material Handling
Using Multiple Robots"
D-Y Yeung, George Bekey, University of Southern California
"Using Neural Networks to Characterize Complex Systems"
Philip Daley, A. Thornbrugh, Martin-Marietta Astronautics Group
Nancy Sliwa
NASA Langley Research Center
804/865-3871
nesliwa%nasamail@ames.arpa
------------------------------
End of AIList Digest
********************
