Copy Link
Add to Bookmark
Report
Machine Learning List Vol. 4 No. 01
Machine Learning List: Vol. 4 No. 1
Tuesday, Jan 21, 1992
Contents:
Learning And Vision
AAAI-92 Workshop on Approximation and Abstraction
AAAI-92 Workshop on Constraining Learning with Prior Knowledge
The Second International Workshop on Inductive Logic Programming
The Machine Learning List is moderated. Contributions should be relevant to
the scientific study of machine learning. Mail contributions to ml@ics.uci.edu.
Mail requests to be added or deleted to ml-request@ics.uci.edu. Back issues
may be FTP'd from ics.uci.edu in pub/ml-list/V<X>/<N> or N.Z where X and N are
the volume and number of the issue; ID: anonymous PASSWORD: <your mail address>
------------------------------
Date: Tue, 7 Jan 92 00:23:53 CST
From: "Douglas H. Fisher" <dfisher@vuse.vanderbilt.EDU>
Subject: learning and vision
See the Dec 1991 IEEE PAMI for instructions on submission to a special
issue on `learning and computer vision.'
------------------------------
From: ellman@cs.rutgers.EDU
Date: Mon, 13 Jan 92 14:40:07 EST
Subject: Workshop on Approximation and Abstraction
Approximation and Abstraction of Computational Theories
AAAI-92 Workshop, San Jose, California, July 1992
Thomas Ellman (Chair) (Rutgers University) ellman@cs.rutgers.edu
Brian Falkenhainer (Xerox PARC) falkenhainer.pa@xerox.com
Fausto Giunchiglia (IRST) fausto@irst.it
Richard Keller (NASA Ames Research Center) keller@ptolemy.arc.nasa.gov
Craig Knoblock (USC Information Sciences Institute) knoblock@isi.edu
Devika Subramanian (Cornell University) devika@cs.cornell.edu
Toby Walsh (University of Edinburgh) T.Walsh@edinburgh.ac.uk
1 Topic of the Workshop
Approximations and abstractions are used frequently to overcome computational
intractability in knowledge-based systems. In planning, scheduling and design
tasks, for example, approximations and abstractions diminish the combinatorial
costs of searching large spaces. In simulation, diagnosis and monitoring
tasks, approximations and abstractions diminish the costs of tracking
interactions among large numbers of variables. Current practice usually
requires human experts to provide approximations or abstractions that are
suitable to the problem at hand. A number of investigators is attempting to
remove this burden from humans. One line of investigation is aimed at
automating the synthesis problem, i.e., given a complete, correct but
intractable domain theory, construct an approximate or abstract version of the
theory. A second line of investigation is directed at the selection problem,
i.e., given a set of different approximate or abstract domain theories, select
one that is most appropriate to the problem at hand. The workshop is intended
to examine both the synthesis problem and the selection problem.
2 Objectives of the Workshop
Although a growing number of investigators is studying approximation and
abstraction of computational theories, the field has yet to adopt a shared
framework for discussing various approaches and research issues. The workshop
is intended to promote such a shared framework. It will introduce participants
to research on approximation and abstraction carried out in fields other than
their own specialties. It will enable participants to compare goals,
techniques, vocabularies and paradigms. It will encourage participants to
identify important research issues and engineering hurdles. Finally, it will
facilitate the process of setting up collaborative research projects.
3 Format and Intended Audience
The need for good approximations and abstractions cuts across all areas of
artificial intelligence. The workshop is therefore aimed at investigators from
a variety of areas within the AI community, including machine learning,
qualitative physics, planning, automatic programming and logic, among others.
It is also intended to include people focused on AI applications in areas such
as scheduling, design, simulation, databases, mechanical engineering, medicine
and biology. Approximately forty to fifty people will be invited to attend.
The workshop will include a limited number of formal presentations as well as a
panel discussion. The format will promote interaction and informal discussion
among participants.
4 Submission Requirements
Persons wishing to attend the workshop should submit five copies of a 1 - 2
page research summary including a list of relevant publications, along with a
phone number and an electronic mail address if possible. Persons wishing to
make presentations at the workshop should submit five copies of a short paper
or extended abstract, in addition to the research summary. All submissions
must be received by March 13, 1992 and must be made in the form of paper
copies. Electronic submissions will not be accepted. Notification of
acceptance or rejection will be mailed to applicants by April 3, 1992. Camera
ready copies of papers accepted for inclusion in the working notes of the
workshop will be due on April 17, 1992.
In order to facilitate communication among participants, authors are
encouraged to address the following questions in their papers or research
summaries: What types of computational theories are you investigating? How are
the theories represented? What types of approximations or abstractions are you
investigating? What purpose do they serve? What tradeoffs do they incorporate?
How are they synthesized or selected? What knowledge, criteria and mechanisms
are used to find good approximations or abstractions? What aspects of the
synthesis or selection process appear useful and feasible to automate? What
hurdles must be overcome in order to apply your approach to important problems?
What are good research goals and challenge problems for the field?
5 Submission Address
Thomas Ellman
Department of Computer Science
Rutgers University
Hill Center, Busch Campus
New Brunswick, NJ 08903
Phone: (908) 932-4184
FAX: (908) 932-5530
Email: ellman@cs.rutgers.edu
------------------------------
Date: Mon, 20 Jan 92 11:41:24 -0800
From: Marie desJardins <marie@erg.sri.COM>
Subject: AAAI-92 workshop on Constraining Learning with Prior Knowledge
CALL FOR PARTICIPATION:
AAAI-92 Workshop on Constraining Learning with
Prior Knowledge
DESCRIPTION OF WORKSHOP
Scaling up machine learning techniques to large domains will not be
feasible unless we can find ways to reduce the complexity of the
learning task. One of the primary ways this can be done is by
constraining learning using domain knowledge. By bringing specific
knowledge to bear, the complexity of a learning problem can be greatly
reduced. Furthermore, ignoring available knowledge may cause a system
to form incorrect beliefs.
Explanation-based learning techniques use a complete and correct domain
theory to provide an extremely strong constraint on learning. In this
workshop, we will be exploring how other forms of knowledge can be used
to constrain learning to a lesser degree.
The objective of this workshop is to bring together researchers from a
variety of fields to discuss the problem of using explicit
representations of prior knowledge to constrain learning. Specific
research areas include:
1. How can prior knowledge be used to constrain, or bias, the
learning problem? (E.g., using a half-order theory as in
Meta-DENDRAL, selecting a feature set, or suggesting relevant
experiments.)
2. How can the bias be represented so that it may be integrated
into a general learning model? (E.g., as search control
heuristics or as linguistic templates.)
3. What form does the prior knowledge take? (E.g., monotonicity
constraints, partial domain theories, or probability distribution
information.)
4. How can techniques for using different types of knowledge be
integrated into a single learning system?
5. Where does the knowledge come from? (E.g., from prior learning
or expert advice.)
This is not an exhaustive list of topics of interest. In particular,
we hope to encourage a broad interpretation of the term ``prior
knowledge.'' For example, knowledge about the computational complexity
of the concept to be learned and of the learning process itself, and
resource limitations of the learning system, can be treated as prior
knowledge; therefore, research on resource-bounded learning methods
that use explicit models of time and space requirements is relevant to
the workshop.
Many existing systems embody knowledge that is treated as an implicit
assumption, such as the initial weights in a connectionist system. We
encourage submission of theoretical or empirical work that explicitly
considers the sources and effects of these initial assumptions.
We are especially interested in views on this problem from other
research areas, including but not limited to:
- Cognitive and developmental psychology.
- Knowledge representation.
- Computational learning theory.
- Neural networks.
- Heuristic search.
- Constraint-based reasoning.
- Qualitative reasoning.
We are not looking for lists of constraints for specific problem
areas. Rather, the focus is on innovative approaches to the research
questions enumerated above that will lead to domain-independent ways to
incorporate domain-specific knowledge into the learning process.
In order to encourage an open forum for discussion, the workshop will
be limited to approximately 30 to 40 participants. Preference will be
given to those who have done research on explicitly using prior
knowledge to constrain learning. However, we will also invite
researchers who have done work in related fields with demonstrable
relevance to the topic of the workshop, and some graduate students who
have an interest in the field. Please feel free to contact any of the
program committee members to discuss the relevance of your research to
the workshop.
The workshop will include ten to twelve technical presentations, an
invited talk by Bruce Buchanan, and a wrap-up discussion session at the
end of the day. If there are a large number of high-quality
submissions, we may consider an additional poster session.
Those who wish to attend the workshop should send a two- or three-page
research summary following the submission guidelines below. The
summary should describe your research interests, and discuss their
connection to the workshop topic.
Those who wish to present a paper should send a paper of no more than
4000 words, again following the guidelines below. You may, if you
wish, also submit a research summary which discusses your research
program in more breadth than would be appropriate in a technical
paper.
Submission guidelines: e-mail submissions are preferred, to
marie@erg.sri.com. Electronic submissions must be in ``standalone''
LaTeX format (i.e., no included PostScript files). If e-mail
submission is not possible, please send four hard copies to the
workshop chair (address is given below). All submissions must reach
the chair by March 13.
Submissions will be reviewed by at least two reviewers. Notification
of acceptance or rejection of papers, and invitations to attend the
workshop, will be sent by April 3.
The workshop notes will include presented papers (ten-page limit) and
one-page research summaries from all participants. The papers and
summaries will be due, in camera-ready or electronic form, by April
17.
WORKSHOP CHAIR:
Marie desJardins
SRI International
333 Ravenswood Ave.
Menlo Park CA 94025
marie@erg.sri.com
(415) 859-6323
PROGRAM COMMITTEE:
William Cohen, AT&T Bell Laboratories (wcohen@research.att.com)
Marie desJardins, SRI International (marie@erg.sri.com)
Haym Hirsh, Rutgers University (hirsh@pei.rutgers.edu)
Doug Medin, University of Michigan (Doug_Medin@um.cc.umich.edu)
------------------------------
Subject: Int. Workshop on Inductive Logic Programming (ILP92)
Date: 21 Jan 92 02:59:25 GMT
From: ueda@icot31.icot.or.jp (Ueda Kazunori)
--- First Announcement and Call for Participation ---
The Second International Workshop on
Inductive Logic Programming
(ILP92)
June 6-7, 1992
ICOT, Tokyo, Japan
in conjunction with
International Conference on
Fifth Generation Computer Systems 1992
June 1-5, 1992
This workshop is the second in a series of International Workshops on
Inductive Logic Programing. The workshop will bring together an
international group of researchers in the field. Inductive Logic
Programming is an emerging research area, spawned by Machine Learning
and Logic Programming. While the influence of Logic Programming has
encouraged the development of strong theoretical foundations, this new
area is inheriting its experimental orientation from Machine Learning.
To restrict the workshop to a manageable size, participation is on an
invitational basis. However, invitations could be extended slightly if
the demand exists. (Please contact the Program Chair.) Participants are
welcome either to present new results or simply to take part in the
discussions. Papers are encouraged in, though not restricted to, the
following areas.
[Theory.] Papers should either (1) prove new results concerning programs
which use inductive learning to construct first or higher order
logic descriptions or (2) discuss the relationship of Inductive
Logic Programming to other theoretical areas such as Non-Monotonic
Logic or Abductive Reasoning. Learnability results and theoretical
investigations of predicate invention are especially welcome.
[Implementation.] Details of inductive algorithms. Time complexity
results should be included.
[Experimentation.] Experimental results should be tabulated with
appropriate statistics. Sufficient details should be included to
allow reproduction of results. Comparative studies of different
algorithms running on the same examples, using the same background
knowledge, are especially welcome.
The workshop will immediately follow the International Conference on
Fifth Generation Computer Systems (FGCS92) which will be held in Tokyo
(Tokyo Prince Hotel), June 1-5, 1992. Attendees are thus encouraged
to take part in both ILP92 and the larger FGCS92 conference (contact:
fgcs92@icot.or.jp, fax:+81-3-3456-1618). Participants must find their
own funds for travel.
ORGANIZATION
Program Chair: Stephen Muggleton
Turing Institute
George House
36 North Hanover Street
Glasgow, G1 2AD, U.K.
e-mail: steve@turing.ac.uk
fax: +44-41-552-2985, tel: +44-41-552-6400
Local Chair: Koichi Furukawa
Institute for New Generation Computer Technology (ICOT)
Mita Kokusai Bldg. 21F
4-28 Mita 1-chome, Minato-ku
Tokyo 108, Japan
e-mail: furukawa@icot.or.jp
fax: +81-3-3456-1618, tel: +81-3-3456-3195
PAPER SUBMISSIONS
Printed papers to be included in the proceedings of the workshop must be
received by the program chair no later than February 16, 1992. To
ensure that papers in the proceedings are readable, only hard-copy
papers will be accepted, not papers sent by email or fax. Papers
received after the deadline will not be included in the proceedings.
------------------------------
END of ML-LIST 4.1
