Copy Link
Add to Bookmark
Report
Machine Learning List Vol. 6 No. 09
Machine Learning List: Vol. 6 No. 9
Thursday, March 17, 1994
Contents:
ILPNewsletter Vol. 1 No. 1
MLnet NEWS 2.2
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: Thu, 17 Mar 1994 06:03:46 +0100
From: Ilpnet@ijs.si
Subject: ILPNewsletter Vol. 1 No. 1
[Note: I'm including the first ILPNewsletter in ML-LIST. Please subscribe
as indicated below to receive future copies- Mike]
Dear potential subscriber,
Inductive Logic Programming (ILP) is a research area in the
intersection of inductive machine learning and computational logic.
ILP aims at developing a theoretical framework and practical algorithms
for learning in a first-order (logic programming) formalism.
Enclosed find the first issue of the ILP Newsletter, a newsletter of the
European Inductive Logic Programming Scientific Network ILPNET.
The ILP Newsletter will include material relevant to ILPNET
and ILP in general, including a calendar of ILP events, conference reports
(from the ILP perspective), book reviews, etc.
To subscribe or unsubscribe, send a message to
ilpnet@ijs.si
with a subject heading SUBSCRIBE ILPNEWS or UNSUBSCRIBE ILPNEWS.
We would be glad to receive your contributions to the newsletter.
Send contributions to ilpnet@ijs.si with a subject heading
ILPNEWS CONTRIBUTION.
We hope that you will find the newsletter interesting and
recommend it to your colleagues. Comments and suggestions
are also welcome.
Editors of the ILP Newsletter,
Saso Dzeroski and Nada Lavrac
saso.dzeroski@ijs.si // nada.lavrac@ijs.si
P.S.
Due to mailing problems this issue is late in arriving
(some of you may be receiving this issue for a second time).
The calls for papers for the MLNET workshops are thus out of date.
However, you can still attend the workshops.
Please accept our apologies for the inconvenience.
%------------------------------------------------------------------------------%
ILP Newsletter
Volume 1, Number 1, 22nd February 1994
%------------------------------------------------------------------------------%
Address all communication related to the ILP Newsletter to ilpnet@ijs.si
To subscribe/unsubscribe send email with subject SUBSCRIBE/UNSUBSCRIBE ILPNEWS
Send contributions in messages with subject heading ILPNEWS CONTRIBUTION
Send comments and suggestions under subject heading ILPNEWS COMMENTS
%------------------------------------------------------------------------------%
Contents:
- About ILPNET, The Inductive Logic Programming Scientific Network
- About the GMD repository of ILP publications, data and programs
- SIGART Special issue on Inductive Logic Programming
- List of ILP and ILP-related books
- ILP book announcements
- Call for papers: Workshops on Declarative Bias and Theory Revision
- Call for papers: Fourth International ILP Workshop (ILP'94)
%------------------------------------------------------------------------------%
%------------------------------------------------------------------------------%
About ILPNET, The Inductive Logic Programming Scientific Network
----------------------------------------------------------------
ILPNET is the Inductive Logic Programming Pan-European Scientific
Network, financially supported by the European Community Action for
Cooperation in Science and Technology with Central and Eastern
European Countries (PECO 92), contract no. CIPA3510OCT920044.
ILPNET is being financed for the duration of three years, starting
on July 26, 1993.
Aims and objectives
-------------------
Inductive Logic Programming (ILP) is a research area in the
intersection of inductive machine learning and computational logic. The
goal of ILP is to upgrade the techniques of the classical inductive
machine learning paradigm to a logic programming framework.
The aim of the scientific network ILPNET is to stimulate the
development, coordination, communication and exchange of results and
personnel in European ILP research and to disseminate the research
results to a wider European/world community. In the scientific
sense, the goal of ILPNET is to provide the infrastructure for the
ongoing European research in ILP, most of which is currently
performed within the ESPRIT III Basic Research Project No. 6020
Inductive Logic Programming (coordinated by Luc De Raedt and Maurice
Bruynooghe, Katholieke Universiteit Leuven).
ILPNET has the following objectives:
- Support the existing and build new communication channels
between ILPNET nodes.
- Coordinate and enable joint research activities by supporting
short visits of researchers at other ILPNET nodes.
- Support organizing and attending specialized meetings and workshops.
- Build a common database of ILP scientific publications, data and systems.
- Promote the results of ILP research also outside ILPNET.
Structure of ILPNET
-------------------
ILPNET consists of 19 European research institutions.
The coordinating node of ILPNET is the J. Stefan Institute, Ljubljana,
Slovenia. Table 1 is a list of ILPNET nodes. Given are the
names of Project Managers, their email addresses, and the
ILPNET node acronym, name, town and country.
--------------------------------------------------------------------------------
Table 1. ILPNET nodes and project managers.
--------------------------------------------------------------------------------
|Project Manager | Email | ILPNET Node Acronym | Name, Town | Country
--------------------------------------------------------------------------------
-| N. Lavrac | nada.lavrac@ijs.si | LAI--IJS | J. Stefan Institute, Ljubljana |
| Slovenia |
--------------------------------------------------------------------------------
-| A. Jezernik |jezernik@uni-mb.si | TFM | Faculty of Technical Sciences,
Maribor | Slovenia |
-| Z. Markov | markov@iinf.bg | SO | Bulgarian Academy of Science, Sofia |
Bulgaria |
-| O. Stepankova | step@lab.felk.cvut.cz | CTU | Czech Technical University,
Prague | Czech Republic |
-| G. Tecuci | tecuci@cs.gmu.edu | RA | Romanian Academy of Science,
Bucharest | Romania |
--------------------------------------------------------------------------------
-| I. Mozetic | igor@ai.univie.ac.at | OFAI | Austrian Research Institute for AI
Vienna | Austria |
-| P. Brazdil | pbrazdil@ncc.up.pt | LIACC | University of Porto | Portugal |
-| K. Morik | morik@kimo.informatik.uni-dortmund.de | UDO |
| University of Dortmund, Dortmund | Germany |
-| L. De Raedt | lucdr@cs.kuleuven.ac.be | KUL | Katholieke Universiteit Leuven
| Belgium |
-| C. Rouveirol | celine@lri.fr | LRI | Universite Paris-Sud | France |
-| S. Wrobel | stefan.wrobel@gmd.de | GMD | GMD, Bonn | Germany |
-| B. Tausend | tausend@informatik.uni-stuttgart.de | STU |
| University of Stuttgart | Germany |
-| S. Muggleton | steve@comlab.ox.ac.uk | OUCL | Oxford University | UK |
-| F. Bergadano | bergadan@di.unito.it | TO | University of Torino | Italy |
-| C.G. Jansson | calle@dsv.su.se | STO | University of Stockholm | Sweden |
-| P.A. Flach | flach@kub.nl | KUB | Tilburg University | Netherlands |
-| R. Wirth | wirth@faw.uni-ulm.de | FAW | FAW, Ulm | Germany |
-| M. Kubat | mirek@dpmi.tu-graz.ac.at | TUG | Technical University, Graz |
| Austria |
-| T. Gyimothy | h42gyi@ella.hu | HA | Hungarian Academy of Scence, Szeged |
| Hungary |
--------------------------------------------------------------------------------
The management structure of ILPNET is the following:
------------------------------
- Academic Coordinator:
Nada Lavrac, J. Stefan Institute, Ljubljana.
- Management Board:
Consists of the Academic Coordinator and 18 Project
Managers of ILPNET nodes (Table 1 lists the names of the Project Managers).
- Academic Secretary: Darko Zupanic, J. Stefan Institute, Ljubljana
Contact:
--------
Nada Lavrac, Darko Zupanic
Jozef Stefan Institute, Jamova 39, 61111 Ljubljana, Slovenia
phone +386 61 12 59 199, fax +386 61 12 58 058 and +386 61 219 385
Email: {Nada.Lavrac,Darko.Zupanic}@ijs.si
ILPNET activities
-----------------
The first ILPNET Management Board meeting was held in Paris on October 2, 1993.
The second Management Board meeting will be held at ECML'94, Catania, Italy.
Tentative date: April 8, 1994.
The first Management Board meeting was attended by:
Francesco Bergadano, Pavel Brazdil, Saso Dzeroski, Luc De Raedt, Peter Flach,
Danielle Gunetti, Tibor Gyimothy, Carl Gustaf Jansson, Nada Lavrac,
Igor Mozetic, Stephen Muggleton, Celine Rouveirol, Olga Stepankova,
Birgit Tausend, Steffo Weber, Rudiger Wirth and Stefan Wrobel.
Agenda of the Management Board meeting:
---------------------------------------
1. Formal meaning and contents of the minutes of the ILPNET meeting.
2. Information about the new phone/fax numbers of J. Stefan Institute.
3. ILPNET Newsletter and its relation to ML Net Newsletter.
4. Status of the ILP Human Capital and Mobility proposal.
5. Appointment of ILPNET secretary.
6. ILPNET partners, two new partners (FAW, Ulm and Technical University, Graz)
and proposal for the acceptance of another new partner.
7. ILPNET contract with CEC.
8. Communication infrastructure.
9. Policy of acknowledgement of ILPNET support
10. Workplan
- ILPNET brochure
- Database of ILP publications
- Database of ILP data sets
- Archive of ILP software
- ILP workshops (status of ILP93 workshop Bled, ILP94 workshop)
- Exchanging visits of scientists/planned work on common projects
- Next ILPNET meeting
11. Deliverables.
12. Reports as requested by Annex II.
13. Confirmation of agreements made about contract and financial arrangements.
14. Various.
The minutes of the MB meeting were taken by Saso Dzeroski and were approved
by MB members via Email. Approved minutes act as an operational workplan for
the activities of the ILPNET in the first year of the project.
The following subsections report on ILPNET activities,
---------------------------
performed in accordance with the workplan for the first year.
Workshops
---------
The Third International Inductive Logic Programming Workshop (program
chair: Stephen Muggleton, conference chair: Nada Lavrac) was
organized by the J. Stefan Institute and held at Bled, on April 1-3, 1993.
The Fourth International Inductive Logic Programming Worhskop
(program and conference chair: Stefan Wrobel) will be organized by GMD.
It will be held in Bad Honnef/Bonn in the second year of the project, on
September 12-14, 1994 - the call for papers has already been widely distributed.
Newsletter
----------
ILPNET starts with publishing a newsletter, edited by Saso Dzeroski
and Nada Lavrac. The ILP Newsletter will include material relevant to ILPNET
and ILP in general, including a calendar of ILP events, conference reports
(from the ILP perspective), book reviews, etc. A selection of information
appearing in the ILP Newsletter will be forwarded to the ML Net Newsletter
and the ML mailing list. Initially, the ILP Newsletter will be distributed
only electronically, one issue will be printed out for the archive.
It is planned to distribute the ILP Newsletter also in printed form later on.
Mailing lists
-------------
Three mailing lists will be established:
- ILPNET: a list of ILPNET Project Managers and some other ILPNET node members.
- ILPNEWS: a list of ILP Newsletter subscribers.
- ILPWORLD: a list of individuals interested in ILP.
Databases
---------
One of the main goals of ILPNET is to establish a central database
which will store a list of scientific publications relevant to ILP,
public domain prototypes of ILP systems, and data concerning ILP applications.
The archive of data sets can be used as a testbed for novel ILP systems.
In year 1, all databases will be
collected at GMD. They can be accessed by anonymous ftp to
ftp.gmd.de on directories /MachineLearning/ILP/public/bib,
/MachineLearning/ILP/public/data and /MachineLearning/ILP/public/software.
A uniform environment for accessing ILP references, data and software
will be designed at IJS in the second year of the project.
Publications
------------
In year 1, a database of bibliographic references relevant to ILP will be
collected at GMD and maintained by IJS. It can be accessed by anonymous ftp to
ftp.gmd.de on directory /MachineLearning/ILP/public/bib.
A new format for establishing uniform labeling of references is currently
being developed at IJS, adopted from KUL.
Datasets
--------
In year 1, the database of ILP data sets is being collected at GMD.
It will be accessible by anonymous ftp to ftp.gmd.de
directory /MachineLearning/ILP/public/data.
Software
--------
In year 1, the archive of ILP software is being kept at GMD.
The directory /MachineLearning/ILP/public/software presently contains MOBAL
with the following ILP learners: RDT, GOLEM, FOIL, mFOIL, MacCLINT and INDEX,
as well as MILES, a flexible environment for tests with ILP methods.
%------------------------------------------------------------------------------%
%------------------------------------------------------------------------------%
GMD MachineLearning-Archive: Announcement and Call for Contributions
The ML-archive ftp.gmd.de:/MachineLearning [129.26.8.90] contains a
growing collection of Machine Learning related papers, articles, tech
reports, data, and software with a particular focus on results
achieved by the European ESPRIT research projects "Machine Learning
Toolbox" (MLT) and "Inductive Logic Programming" (ILP), the European
Network of Excellence in Machine Learning (MLnet) and the Inductive
Logic Programming Pan-European Scientific Network (ILPnet).
For example, the archive presently contains
- the source code of Stephen Muggleton's and Cao Feng's learning sys-
tem Golem (in "/MachineLearning/ILP/public/software/golem"),
- a BibTex file with around 325 entries of articles related to ILP
("/MachineLearning/ILP/public/bib"),
- the knowledge acquisition and machine learning system MOBAL 2.2
for non-commercial academic use in
"/MachineLearning/ILP/public/software/Mobal"), and
- PROLOG implementations of basic machine learning algorithms (e.g.,
COBWEB, ID3, ARCH) ("/MachineLearning/general/ML-Program-Library").
This library is maintained by Thomas Hoppe (for more details, see
the README file in the subdirectory).
Here's how the anonymous FTP server works. To access or store material
on the server, use ftp to ftp.gmd.de, login ID "anonymous" and your
full E-Mail address as password. Change to directory
/MachineLearning, where the ML-related stuff is located. Remember,
when ftping compressed or compacted files (.Z) to use binary mode for
retrieving the files.
The directory structure is subject to change.
Please note: Wherever appropriate and possible, material has been
cross-indexed between the different subdirectories using symbolic
links.
You are invited to contribute your own software, papers etc. to the
ML-archive. If you have ML-related material, which might be relevant
for other researchers or potential users of Machine Learning
techniques, place it in one of the subdirectories of
"/ftp/incoming/Learning" AND also send mail to "ml-archive@gmd.de"
saying what you placed in "incoming". Our ml-archive manager Marcus
Luebbe will read these mails and install all contributions in the
proper place. As for papers, please send them in compressed
PostScript (.ps.Z) form. Please send us also a file with a plain text
bibliographic entry and, if possible, a corresponding BibTeX entry
with names of all authors, title, how and where the paper has been
published. As for software, please send both a compressed tarfile
containing your software and manuals as well as a README file
describing the software and its installation. Please let us know
which name to use for the subdirectory that stores your software.
Please, include the following statement in your mail:
COPYRIGHT CLEARANCE: I understand that the material I have
submitted will be made publicly available worldwide on an anoymous
FTP Server. I have made sure that this does not conflict with any
relevant copyrights on the material.
Please send questions and suggestions to: ml-archive@gmd.de
%------------------------------------------------------------------------------%
%------------------------------------------------------------------------------%
SIGART Bulletin Special Issue on Inductive Logic Programming
(Volume 5, Number 1, January 1994, ACM Press)
Table of contents:
Muggleton S., "Inductive Logic Programming"
Mooney R. and Zelle J., "Integrating ILP and EBL"
Kietz J-U. and Dzeroski S., "Inductive Logic Programming and Learnability"
Cameron-Jones M. and Quinlan R.,
"Efficient Top-down Induction of Logic Programs"
Bratko I. and King R., "Applications of Inductive Logic Programming"
%------------------------------------------------------------------------------%
%------------------------------------------------------------------------------%
ILP BOOKS
Published:
----------
- Shapiro, E. (1983) Algorithmic program debugging. MIT Press.
- Muggleton, S., editor. (1992) Inductive logic programming. Academic Press.
- De Raedt, L. (1992) Interactive theory revision:
an inductive logic programming approach. Academic Press.
- Morik, K. et al. (1993) Knowledge acquisition and machine learning:
theory, methods, and applications. Academic press.
- Lavrac, N. and Dzeroski, S. (1994) Inductive logic programming: techniques
and applications. Ellis Horwood.
Unpublished:
------------
- Muggleton, S., editor. (1991)
Proceedings of the International Workshop on Inductive Logic Programming
(ILP91). Viana de Castelo, Portugal, March 1991. (307 pages).
- Muggleton, S. and Furukawa, K., editors. (1992)
Proceedings of the Second International Workshop on Inductive Logic
Programming (ILP92). Tokyo, Japan, June 1992. Technical Report TM-1182,
ICOT Research Center, Tokyo Japan. (283 pages).
- Muggleton, S., editor. (1993)
Proceedings of the Third International Workshop on Inductive Logic Programming
(ILP93). Bled, Slovenia, April 1993. Technical Report IJS-DP-6707,
Jozef Stefan Institute, Ljubljana, Slovenia. (292 pages).
Forthcoming:
------------
- Bergadano, F. and Gunetti, D. (1995)
Inductive Logic Programming: from Machine Learning to Software Engineering.
MIT Press.
- Muggleton, S. (1995) Foundations of Inductive Logic Programming.
Prentice Hall.
Related to ILP:
---------------
- Bergadano, F. et al. (1991) Machine learning: an integrated framework
and its applications. Ellis Horwood.
- Flach, P. (1994) Simply logical: intelligent reasoning by example.
John Wiley. Forthcoming.
- Michalski, R. and Tecuci, G. (1994) Machine learning: A multistrategy
approach, vol IV. Morgan Kaufmann.
%------------------------------------------------------------------------------%
%------------------------------------------------------------------------------%
BOOK ANNOUNCEMENT
"Knowledge Acquisition and Machine Learning - Theory, Methods and Applications"
K. Morik, S. Wrobel, J.U.Kietz, and W. Emde
Academic Press 1993
ISBN: 0-12-506230-3, 350 pages, 34.95 pounds
The book shows how incorporating learning algorithms into a knowledge
acquisition environment provides new work-share between system and
user, assisting the user in both setting up a learning task using the
knowledge acquisition environment and supporting knowledge acquisition
and knowledge maintenance using learning algorithms. The book reports
on BLIP and MOBAL, fully operational systems which integrate knowledge
acquisition, maintenance and learning in a restricted predicate logic.
The book is practically oriented. Theoretical results have been used
and and tested in real-world applications of different complexity and size.
%------------------------------------------------------------------------------%
%------------------------------------------------------------------------------%
BOOK ANNOUNCEMENT
"INDUCTIVE LOGIC PROGRAMMING: Techniques and Applications"
Nada Lavrac and Saso Dzeroski
Ellis Horwood (Simon & Schuster), 1994
(Ellis Horwood Series in Artificial Intelligence)
ISBN: 0-13-457870-8, 310 pages, 39.95 pounds (67.95 dollars)
Keywords: artificial intelligence, applications, databases, deductive databases,
induction, learning, logic, logic programming, machine learning,
knowledge discovery in databases
The book is an introduction to inductive logic programming (ILP), a
research area at the intersection of inductive machine learning and
logic programming. This field aims at a formal framework and practical
algorithms for inductively learning relational descriptions in the form
of logic programs. ILP is of interest to inductive machine learning
researchers as it significantly extends the usual attribute-value
respresentation and consequently enlarges the scope of machine learning
applications; it is also of interest to logic programming researchers as
it extends the basically deductive framework of logic programming with
the use of induction.
The book consists of four parts. Part I is an introduction to the field
of ILP. Part II describes in detail several empirical ILP techniques and
their implementations. Part III presents the techniques for handling
imperfect data in ILP, whereas Part IV gives an overview of several ILP
applications.
The book serves two main purposes. On the one hand, it can be used as a
course book on ILP since it provides an easy-to-read introduction to ILP
(Chapters 1-3), an overview of empirical ILP systems (Chapter 4),
discusses ILP as search of refinement graphs (Chapter 7), analyses the
sources of imperfect/noisy data and the mechanisms for handling noise
(Chapter 8) and gives an overview of several interesting applications of
ILP (Chapter 14). On the other hand, the book is a guide/reference for
an in-depth study of specific empirical ILP techniques, i.e., using
attribute-value learners in an ILP framework and specialization
techniques based on FOIL (Chapters 5-6,9-10) and their applications in
medicine, mesh design and learning of qualitative models (Chapters 11-13).
The book will be of interest to engineers, researchers and graduate
students in the field of artificial intelligence and database
methodology, in particular in machine learning, logic programming,
software engineering, deductive databases, and knowledge discovery in
databases. Basic knowledge of artificial intelligence and logic would be
helpful, but is not a prerequisite.
%------------------------------------------------------------------------------%
%------------------------------------------------------------------------------%
MLNET FAMILIARISATION WORKSHOP
Declarative Bias
Catania, Italy, april 9 1994
**********************************************************************
The workshop is to be organized after the European Conference on
Machine Learning (ECML 94), april 6-8 1994, in the context of the
second MLNet familiarization workshop Catania, Italy.
**********************************************************************
Call for Contributions
Control of the learning process has always been a fundamental issue in
ML because it strongly affects the complexity of the learning process
and the learning results. There has been lately a strong interest
within the Machine Learning community concerning the elicitation of
this control knowledge, referred to as Declarative Bias. It grows
with the development of real world applications that require more
adaptable learning tools and more complex representation languages.
Representing control knowledge in a declarative way allows an expert
in ML or the ML system itself to easily shift it.
Past experiences with ML applications have demonstrated that
cooperation with the user speeds up the learning process by providing
explicit control information when available. Declarative bias is
therefore a concise and powerful way for the user to explicitly
program the ML system, instead of tuning low level knowledge such as
examples and domain theory representation in order to improve learning
results.
Three types of biases may be characterised. The first class a priori
restricts the initial set of candidate definitions for the target
concept (search space), referred to as language biases, the second
class sets heuristics to improve the search for the best definition(s)
through the search space. The third class defines validation criteria
for learning.
It is obviously a difficult task for the user to find the appropriate
combination of biases to meet her/his expectations. Shifting biases
with respect to discrepancies observed between actual learning results
and expected ones is a promising research issue. Validation of
results and shift of bias may be incrementally performed, after each
learning step by submitting intermediate results to the user or at the
end of the learning process. This cyclic task will be as easy as the
relationship between biases and her/his learning goals are stated
clearly.
Authors are encouraged to submit papers describing their favorite
learning system(s) in terms of elementary learning steps and biases
belonging to each of the three above classes.
SUBMISSION REQUIREMENTS
Authors should submit a paper or an extended abstract (not less than 2
pages) fully explaining the relevance of their work to the
workshop. Persons wishing to participate but who do not wish to give a
presentation should submit an abstract (1 page) describing their
research and/or interest in the subject area and their expected
contributions to the workshop. Papers / abstracts should be sent in
five copies by March 1 to:
Celine Rouveirol
LRI Bat 490
Universite Paris-Sud
F-91405 Orsay, France
Tel : +33 (1) 69 41 64 62
Fax : +33 (1) 69 41 65 86
e-mail : celine@lri.fr
Notification of acceptance will be e-mailed or faxed by March 15
(please specify your email or fax on the submitted paper).
ORGANISATION
All attendees will receive before the workshop a list of topics and
open questions that have emerged from the accepted papers. To
stimulate discussions, presentations are strongly encouraged to refer
to these. The worshop will start with a panel presenting the key
issues that will be discussed during the sessions. Schedule will
leave time for open discussions and syntheses.
PROGRAM COMMITTEE
Rouveirol C. Univ. Paris-Sud, France
Bergadano F. Univ. Catania, Italy
Esposito F. Univ. Bari, Italy
Lavrac N. JSI, Ljubljana, Slovenia
Mozetic I. Techn. Univ. Vienna - ARIAI, Austria
Nedellec C. Univ. Paris-Sud, France
Plaza E. IIIA-CSIC, Spain
Popelinsky L. Univ. Brno, Czech Republic
Sleeman D. Univ. Aberdeen, U.K.
Van de Merckt T. Free Univ. of Brussels, Belgium
Van Someren M. Univ. Amsterdam, The Netherlands
%------------------------------------------------------------------------------%
%------------------------------------------------------------------------------%
ECML MLNet Workshop on
Theory revision and restructuring
Catania, Italy, April 9 or 10, 1994
Call for Contributions
With the growing complexity of applications being tackled
by Machine Learning, it has become increasingly clear
that besides approaches for the initial acquisition of
knowledge bases we also need techniques for theory revision
and restructuring, i.e., techniques that can use existing
learned or human-supplied domain theories and can modify
them to improve their correctness, completeness, efficiency
or understandability.
This ECML MLNet workshop intends to bring together
the various approaches to revision and restructuring that
have been developed under different perspectives within
Machine Learning. Traditionally, revision has been a
part of incremental or hill-climbing learning systems which
keep only one current hypothesis and modify it whenever
new examples arise. More recently, revision has been
identified has an important part of approaches that learn
multiple predicates simultaneously, and incorporated as
a central component of integrated multi-strategy learning
systems. Moreover, revision and restructuring are also
important topics in several neighboring fields such as
knowledge representation, logic programming or deductive
databases.
The workshop invites submissions on all topics related
to theory revision and restructuring, including but not
limited to:
o multiple-predicate learning
o selection of preferred revisions, bias, constraints
o revision as a part of multi-strategy learners
o relationships to neighboring fields (e.g. revision
work in knowledge representation and deductive database
communities)
o scientific theory revision
o debugging techniques for revision
o theory restructuring
o applications of these techniques
Authors intending to present their work should submit a
two-page abstract of their talk until March 1st, 1994, to:
Stefan Wrobel
GMD, I3.KI, Schloss Birlinghoven
53757 Sankt Augustin, Germany
stefan.wrobel@gmd.de
preferably by E-Mail in LaTeX. Authors will be notified of
acceptance until March 14, 1993. A handout of accepted
abstracts will be made available to participants.
Organizing Committee:
Hilde Ade, Carl-Gustav Jansson, Stefan Wrobel.
%------------------------------------------------------------------------------%
%------------------------------------------------------------------------------%
The Fourth International Workshop on Inductive Logic Programming (ILP94)
September 12 -- 14, 1994, Bad Honnef/Bonn, Germany
Call for Papers (Short Version)
General Information
Originating from the intersection of Machine Learning and Logic
Programming, Inductive Logic Programming (ILP) is an important
and rapidly developing field that focuses on theory, methods,
and applications of learning in relational, first-order logic
representations. ILP94 is the fourth in a series of international
workshops designed to bring together developers and users of ILP in
a format that allows a detailed exchange of ideas and discussions.
Reflecting the growing maturity of the field, ILP94 for the first
time will offer a systems and application exhibit as an opportunity
to demonstrate the practical results and capabilities of ILP.
ILP94 will take place in Bad Honnef, a small resort town close to
Bonn in the Rhine valley and adjacent to the Siebengebirge nature
park. Participants will be able to take advantage of Bad Honnef's
vicinity to medieval castles and of the new wine season that starts
at the time of the workshop.
Submission of papers
Reflecting the broadening scope of the field, ILP94 invites papers
covering the three main aspects of ILP, namely inductive data
analysis and learning in first-order representations, inductive
synthesis of non-trivial logic programs from examples, and inductive
tools for software engineering. Please submit four paper copies of
your paper to the workshop chair
Stefan Wrobel
GMD, I3.KI, Schloss Birlinghoven, 53757 Sankt Augustin, Germany.
E-Mail: ilp-94@gmd.de, Fax:+49/2241/14-2889 Tel: +49/2241/14-2670
to be received on or before May 31, 1994. Length of papers should
be reasonable and adequate for the topic, but no more than 20 pages.
Please use LaTeX if at all possible. Authors will be notified of
acceptance or rejection until July 15, 1994, and camery-ready copy
will be due on August 9, 1994. Accepted papers will be published
as a GMD technical report to be distributed at the workshop and
officially available to others from GMD afterwards. Publication of
an edited book is planned for after the workshop.
Program Committee
Francesco Bergadano (Italy) Ivan Bratko (Slovenia)
Wray Buntine (USA) William W. Cohen (USA)
Luc de Raedt (Belgium) Koichi Furukawa (Japan)
Jorg-Uwe Kietz (Germany) Nada Lavravc (Slovenia)
Stan Matwin (Canada) Stephen Muggleton (UK)
Celine Rouveirol (France) Claude Sammut (Australia)
Further Information
A full call for papers can be obtained via anonymous
FTP from the ML Archive at GMD (server ftp.gmd.de, file
/MachineLearning/general/CallsForPapers/ilp94.ascii or .ps). To
receive the complete registration brochure as soon as it is
available, please send E-Mail to ilp-94@gmd.de, specifying your name
and address, E-Mail, Fax, and (if you know) whether you intend to
submit a paper.
%------------------------------------------------------------------------------%
------------------------------
Date: Tue, 15 Mar 94 16:50:27 0000
From: MLnet Admin <mlnet@computing-science.aberdeen.ac.uk>
Subject: MLnet NEWS 2.2
**********************************
MLnet NEWS 2.2 February 1994
Electronic version
**********************************
(If you want to be added to the mailing list
of the "printed" newsletter please contact us at :
mlnet@csd.abdn.ac.uk
Tel: +44 224 272304
Fax: +44 224 273422
***********************************
CONTENTS:
- News From the Technical Committees
- MLnet Familiarization Workshops, Call for Participation
- ECML 94, Call for Participation
- ECML 95, Call for Proposals
- Network Meeting @ CEC 27/9/93
- IJCAI-93, Review
- Reinforcement Learning, by L.P.Kaelbling
- Current Academic Research in Europe, Questionnaire
- Focus on Machine Learning Research at the University of Aberdeen
- ECAI-94, Call for Participation
- MLnet Workshop on Industrial Applications of Machine Learning,
Call for Participation
- MLnet Summer School, Call for Participation
- Junior scientist Fellowship, Call for Application
- Procedures for Joining MLnet
- List of Main and associate nodes of MLnet
***********************************
NEWS From the Technical Committees
Electronic Communication Technical Committee
Bob Wielinga (the convener) reported on the meeting with other
Networks held in November '93 where the role of the Networks was
discussed (see page 6).
Regarding the infrastructure for electronic communication, the
decision was taken to run a small scale experiment with the ANDREW
distributed file system. The general feeling was, however, that
modern ftp services could match the requirements of MLnet at a much
lower cost than ANDREW.
The access facilities to the data bases that MLnet is constructing
were discussed and, as the expected sizes of those DBs are quite
small, the decision was taken to simply download the DBs to local
machines and to access them locally.
Industrial Liaison Technical Committee
The progress on the construction of a data base on active researches
in ML/KA in European IT industries was discussed.
The proposal for an Industrial Liaison Workshop, to be held near
Paris in September, was accepted (see page 13 for the provisional
programme and a first call for participation).
Research Technical Committee
The convener (Lorenza Saitta) reported on the status of the data base
on ML/KA academic researchers in Europe; the data base will be made
available in Summer '94 on Macintoshes (FileMaker Pro), Unix machines
(LaTeX files), and PC DOS (simple text files). Please, if
appropriate, complete the questionnaire on page 10.
A "State of the Art" document is planned and will be made available
shortly.
Training Technical Committee
The proposal for a Summer School, to be held near Paris in September,
was accepted (see page 13 for the provisional programme and a first
call for participation).
Written Communication Technical Committee
Future issues of MLnet NEWS have been planned for:
- May (including reports on the Catania Conference and
Familiarization Workshop)
- October (including reports on the Industrial Liaison Workshop and
the Summer School).
Management Board
Funding was approved for ECML-94 and a series of Familiarization
Workshops to be held immediately after the conference. An important
aspect of the workshop will be to discuss policy-related matters.
**********************************
MLnet Familiarization Workshops
9-10 APRIL 1994 - CATANIA, ITALY
Call for Participation
MLnet is organizing four one-day workshops on selected Machine
Learning topics, to be held at the University of Catania, on April
9th and April 10th, 1994.
Expenses (APEX air fare, two nights of Hotel at a maximum of 100,000
liras per night, and workshop registration fees) will be met by MLnet
for approved participants. If car is used, then first class rail fare
can be charged by the driver only. In these circumstances, car
drivers will only be reimbursed if they present, in Catania, a note
from a travel agency saying what the first class rail fare is. In
general, all receipts, photocopies of plane tickets, and other
claims for reimbursement must be ready in Catania and will be
collected by Davide Roverso.
As attendance is limited to 80 at this time, at most 3 participants
per main or associate node should ask bergadan@di.unito.it for
approval, mentioning their APEX flight cost to Catania. Requests will
be ordered chronologically in a file and answered before March 1st,
1994.
Important:
- Late requests may not be satisfied (after a total number of about
80 participants is reached)
- No more than the APEX cost you declare will be refunded
- Hotel reservation (see ECML forms below) must be made before
February 5th
- People outside the EC or not belonging to MLnet need special
approval, and will be handled separately, probably with some delay.
The workshops are open to other MLnet members and to all ECML
participants, but payment of the registration fee is required.
The planned workshops and contact persons are:
Theory Revision and Restructuring
Stefan Wrobel
GMD (The German Natl. Research Center for Comp. Science)
FIT.KI, Schloss Birlinghoven, 53757 Sankt Augustin
Fed. Rep. of Germany
Tel.: +49 2241 14 2670, Fax +49 2241 14 2889,
email: wrobel@gmd.de
Knowledge Level Models of Machine Learning
Walter Van de Velde
Artificial Intelligence Laboratory
Vrije Universiteit Brussel, Pleinlaan 2, B-1050 Brussels
Tel: +32 2 641 37 00, Fax: +32 2 641 37 29,
email: walter@arti.vub.ac.be
Declarative Bias
Celine Rouveirol
LRI Bat 490, Universite Paris-Sud
F-91405 Orsay, France
Tel: +33 1 69416462, Fax: +33 1 69416586,
email: celine@lri.fr
Machine Learning and Statistics
Gholamreza Nakhaeizadeh and Charles Taylor
Daimler-Benz AG, Research and Technology
Postfach 2360, 89013 Ulm, Germany
Tel: +49 731 505 2860, Fax: +49 731 505 4210,
email: reza@fuzi.uucp
Those interested in giving a talk should send an extended abstract to
the workshop organizer by March 1st, 1993. Workshop organizers,
however, may distribute separate calls and prospective authors are
encouraged to ask them for more specific instructions. Workshop-
dependent "Specializations" and "Refinements" on deadlines and format
are in fact possible.
Other enquiries should go to the familiarization workshop organizers:
Francesco Bergadano and Luc De Raedt
Dipartimento di Matematica
Universita' di Catania
Via Andrea Doria 6, Citta' Universitaria
Catania, Italy
Fax: +39 95 330094
email: bergadan@di.unito.it, Luc.DeRaedt@cs.kuleuven.ac.be
Anyone interested in participating at the workshops is invited to
fill the relevant parts of the ECML-94 registration and reservation
forms (on pages 4-5) and return them to La Duca Viaggi before the
indicated deadlines.
**********************************
ECML 94
7th EUROPEAN CONFERENCE ON MACHINE
LEARNING
6-8 APRIL 1994 - CATANIA, ITALY
Call for Participation
GENERAL INFORMATION
ECML-94 is the 7th meeting of this kind, continuing the tradition of
EWSL conferences, and the second under this name, after ECML-93 in
Vienna. ECML will continue to provide a major occasion for
presenting the latest and most significant results in the area of
Machine Learning.
PROGRAM
The scientific program will include presentation of selected
papers and two invited talks. The invited speakers are Michael
Kearns (Bell Labs) and Lorenza Saitta (University of Torino). An
invited panel on industrial applications will be chaired by Yves
Kodratoff (CNRS). Conference Proceedings will be published by
Springer-Verlag, Lecture Notes in Artificial Intelligence.
PROGRAM CHAIRS:
Francesco Bergadano
(University of Catania, Italy)
Luc De Raedt
(Katholieke Universiteit Leuven, Belgium)
PROGRAM COMMITTEE:
Ivan Bratko (Slovenia), Pavel Brazdil (Portugal),
Wray Buntine (USA), Floriana Esposito (Italy),
Jean-Gabriel Ganascia (France), Igor Kononenko (Slovenia),
Yves Kodratoff (France), Nada Lavrac (Slovenia),
Stan Matwin (Canada), Katharina Morik (Germany),
Igor Mozetic (Austria), Stephen Muggleton (UK),
Enric Plaza (Spain), Lorenza Saitta (Italy),
Derek Sleeman (UK), Paul Vitanyi (Netherlands),
Gerhard Widmer (Austria), Stefan Wrobel (Germany)
ORGANIZING COMMITTEE:
H. Ade, V. Cutello, G. Gallo, D. Gunetti, G. Sablon.
CONFERENCE VENUE
ECML-94 will take place at the Department of Mathematics of the
University of Catania, situated five minutes from the city
centre.
Address: Citta' Universitaria, Via Andrea Doria 6,
tel. +39 95 330533 (then ask for "Dipartimento di Matematica"),
fax[+39 95 330094.
TRAVEL INFORMATION
Catania Fontanarossa International airport, at 7 km from the
city centre, is served by Ati, Meridiana, Alitalia, Lufthansa and
several charter companies from the main European gateways. The
airport is connected to the railway station by city bus number
24. A taxi from the airport to downtown should cost about 50,000
ITL. The railway station is situated at 1 km from downtown and is
served by city buses and taxis.
REGISTRATION FEE
The registration fee covers conference participation, one copy of the
Proceedings, the Wednesday evening reception, coffee breaks and
working lunches.
The fee is as follows:
BEFORE MARCH 5th, 1994 ITL 350,000
AFTER MARCH 5th, 1994 ITL 450,000
Please complete the Registration Form (on Page 4) and return it to
LA DUCA VIAGGI SRL (as indicated on the form).
ACCOMMODATION
Hotel accomodation will be arranged by La Duca Viaggi srl,
Congress Department. Rooms have been reserved until February 5th,
1994 at the Grand Hotel Exclesior in Catania, a four stars hotel
centrally located, at very special conference rates. A shuttle
bus will connect the hotel to the Conference site according to the
working sessions. For hotel reservation, please use the reservation
form (on Page 5) and send it directly to La Duca Viaggi srl before
February 5th, 1994.
AIRPORT SHUTTLE SERVICE
A shuttle service will be organized from Catania airport to the
hotel and back on April 5th and 11th. It is possible to book the
transfer at the price of Itl 50,000 per person, round trip. You are
kindly requested to advice your arrival and departure flights on the
reservation form.
SOCIAL PROGRAMME
A welcome reception will be organized on Wednesday, April 6th. The
conference dinner will take place at the Grand Hotel
Excelsior on April 7th at the price of ITL 70,000.
EXCURSIONS
On April 5th, an afternoon excursion to Taormina is scheduled for all
participants and accompanying persons. Taormina is one of the major
tourist attractions in the whole of the Mediterranean; famous
worldwide for its charming position, and for its Greek, Roman and
Medieval remains. Price per person, Itl 40,000. A minimum of 30
participants is required.
PAYMENT
All payments, in Italian Lire (Itl) , may be effected by:
1. Bank transfer in Italian Lire, free of charge for the
benificiary, to La Duca Viaggi srl - bank account n.
1598423/01/68, Banca Commerciale Italiana, Catania Headquaters (Abi
2002.4, Cab 16900.3), reference ECML-94 and participant name.
2. Bank draft, not endorsable, payable to La Duca Viaggi srl.
3. Authorisation to charge on an American Express Credit Card,
specifying Card details and expiration date.
CANCELLATION/REPAYMENT
Registration fees can be refunded at a rate of 50% at any time.
Hotel reservation or any other payment related to subsisence can be
refunded at a 50% rate if cancelled no later than March 6th, 1994.
For further information, please send mail to one of the following
addresses:
ecml@cs.kuleuven.ac.be, bergadan@mathct.cineca.it, or
gunetti@di.unito.it. For other organizational issues, please
contact La Duca Viaggi (address enclosed with the forms).
REGISTRATION FORM
ECML94 RETURN TO
7TH EUROPEAN CONFERENCE LA DUCA VIAGGI SRL
ON MACHINE LEARNING CONGRESS DEPARTMENT
CATANIA, 6-8 APRIL 1994 Via Don Bosco, 39
I-98039 TAORMINA
Phone [+39] 942 625255
Fax [+39] 942 625256
Surname____________________________ Name______________________
Affiliation _____________________________________________________
Address____________________________________ Town-Country_________
Phone_________________ Fax________________ email________________
Accompanying person______________________________________________
VAT (if applicable) _____________________________________________
N.____ Early Registration (ITL 350.000) ITL______________
N.____ Late Registration (ITL 450.000) ITL______________
N.____ Accompanying person (ITL 40.000) ITL______________
N.____ MLnet Workshops (ITL 200.000) ITL______________
Bank Fees ITL 15,000
T O T A L ITL______________
PAYMENT: Please return this form together with:
1.( ) copy of the bank transfer, on La Duca Viaggi srl's account
N. 1598423/01/68, Banca Commerciale Italiana, Corso Sicilia
n. 55 - 95100 CATANIA (ABI-2002.4, CAB-16900.3), reference
ECML-94 and participant name.
2.( ) bank draft, not endorsable, payable to La Duca Viaggi srl;
3.( ) I authorize La Duca Viaggi srl to charge the amount of
Itl___________________________ on my AMERICAN EXPRESS CARD
N.________________________________________________________
Name ________________________Expiration Date_______________
CANCELLATION/REPAYMENT
Registration fees can be refunded at a rate of 50% at any time.
Hotel reservation or any other payment related to subsisence can be refunded at a 50% rate if cancelled no later than March 6th, 1994.
Date___________________________ Signature________________________
RESERVATION FORM
ECML94 RETURN TO
7TH EUROPEAN CONFERENCE LA DUCA VIAGGI SRL
ON MACHINE LEARNING CONGRESS DEPARTMENT
CATANIA, 6-8 APRIL 1994 Via Don Bosco, 39
I-98039 TAORMINA
Phone [+39] 942 625255 Fax [+39] 942 625256
Surname____________________________ Name______________________
Affiliation _____________________________________________________
Address____________________________________ Town-Country_________
Phone_________________ Fax________________ email________________
Accompanying person______________________________________________
VAT (if applicable) _____________________________________________
Date of Arrival_____________________________ Flight N.___________
Date of Departure___________________________ Flight N.___________
EXCELSIOR GRAND HOTEL (FOUR STARS), Piazza Verga, Catania
SPECIAL CONFERENCE RATES:
SINGLE ROOM ITL 100.000, DOUBLE ROOM ITL 160.000
Rates are quoted in Italian Lire, per night and per room, including continental breakfast and taxes.
PLEASE BOOK:
N.____ single room(s), N.____ double room(s) ITL_______________
N.____ airport shuttle service at ITL 50,000 ITL_______________
N.____ excursion to Taormina at Itl 40,000 ITL_______________
N.____ Conference dinner at Itl 70,000 ITL_______________
Bank Fees (not required if done with
registration fee in only one payment) ITL 15.000
TOTAL ITL________________
PAYMENT: Deposit of one night accomodation and full payment for the other services is requested. Please note that no reservation can be made without payment. Please enclose:
1.( ) copy of the bank transfer, on La Duca Viaggi srl's account
N. 1598423/01/68, Banca Commerciale Italiana, Corso Sicilia
n. 55 - 95100 CATANIA (ABI-2002.4, CAB-16900.3), reference
ECML-94 and participant name.
2.( ) bank draft, not endorsable, payable to La Duca Viaggi srl;
3.( ) I authorize La Duca Viaggi srl to charge the amount of
Itl___________________________ on my AMERICAN EXPRESS CARD
N.________________________________________________________
Name ________________________Expiration Date_______________
CANCELLATION/REPAYMENT: Registration fees can be refunded at a rate of 50% at any time. Hotel reservation or any other payment related to subsisence can be refunded at a 50% rate if cancelled no later than March 6th, 1994.
Date___________________________ Signature________________________
**********************************
ECML 95
8th EUROPEAN CONFERENCE ON MACHINE
LEARNING
Call for Proposal
In view of the MLNet Meeting in Catania, where the issue will be
discussed, it is time to think of the next European Conference on
Machine Learning, to be held in 1995.
Any site willing to host the Conference shall present a proposal,
specifing:
- Chairperson
- Format of the conference
- Location of the conference and hosting institution
- Accommodation facilities and costs
- Travelling information
- Preliminary budget for the conference
The proposals shall be sent to the Convener of the Research Technical
Committee of the ML Network, Prof. L. Saitta. The proposals will be
examined at the MLNet's Research Technical Committe Meeting in April
1994 (to be held in Catania).
**********************************
Network Meeting @ CEC,
27/9/93
Summary Report
B J Wielinga (University of Amsterdam)
Introduction
The meeting was attended by representatives from eight networks and a
number of CEC officers from the BRA office. Later during the evening
a visit was made to the European Parliament where the network
representatives met the subcommittee on Energy. The networks were
introduced to the members of that committee through posters and
informal discussions.
During the discussions of various points some clarification of
important questions and issues was given by Simon Bensasson of the
BRA office. His remarks form the basis for the summary below.
Role of the Networks
At various points during the meeting it was made clear that the
networks are viewed as a strategic vehicle to coordinate research in
Europe, but the CEC does not consider itself "owner" of the networks.
Existence and funding of a network does not imply that the area is a
priority area in the CEC's research programme. There is currently no
clear mechanism of how a network can influence the CEC programmes;
however some networks have influenced programmes through state-of-
the-art reports and analyses of their fields' strong and weak points.
In particular assessing the state of the art and plans in Japan and
the US are considered important.
My personal conclusion from this is that we should consider writing
such a report before the workplans for Framework 4 are finalised.
Another role for networks that was emphasised was technology transfer
to industry, in particular SMEs. It was even suggested that SMEs
should become part of networks.
It was also stressed that networks should act as a taskforce to
develop and assess strategies. Networks should produce reports about
what needs to be done and who is going to do it. Unlike the US and
Japan, European industry appears not to have such strategic views,
and the networks should play a role in developing them. In addition,
strategies for human resource management are called for: reports
addressing questions such as what kind of IT staff is lacking in
Europe, would be most welcome.
Framework 4
The next round of R&D programmes will start early in 1995 and this is
currently being discussed in political circles. The total requested
funding is 5% higher than Framework 3. Long Term Research is
currently 10% of the total, and maybe it would rise a little in
Framework 4.
BRA will have two types of projects:
- Long term research projects.
These projects investigate pure research issues, are usually short (1
year), small consortia and many projects. This category will amount
to 1/3 of the budget for BRA. The response rate will be short, calls
are expected every six months. The idea is to have a regular flow of
money.
- Advanced Research projects.
These projects are more result oriented, "initiative generating"
and can be compared to B-type ESPRIT projects. These projects must
provide bridges to industry. This type of project will also have a
fast response time (3 month between submission and signature of
contract).
Infrastructure
A long discussion took place on (electronic) infrastructure. There
appeared to be three tpositions. Cabernet, Elsnet and Compulog have
decided to go ahead with the Andrew File System (AFS)* . Several
networks (including MLnet) are convinced that electronic
communication is an essential medium, but hesitate to invest in AFS
since other services are also becoming available. The third group of
networks - mainly hardware oriented - was not convinced by the need
for extensive electronic communication at all.
No common policy was adopted.
___________
* The arguments given in favour of AFS were: use latest technology possible,
shared information, efficiency and additional functionality.
**********************************
IJCAI-93
Chambery, France, 28 Aug-3 Sep 1993
reviewed by Celine Rouveirol
(LRI, Paris)
IJCAI 93 has been held in Chambery (France) from August 28th until
September 3rd 1993; local organization was by the University of
Savoie.
The three first days were devoted to domain specific workshops and
tutorials. Two workshops related to ML were : Inductive Logic
Programming (F. Bergadano, L. De Raedt, S. Matwin, S. Muggleton) and
Machine Learning and Knowledge Acquisition (S. Kedar, Y. Kodratoff,
G. Tecuci). Machine Learning tutorials were on Case Based Reasoning
(K. Ashley, E. Simoudis and K. Sycara), Applications of Machine
Learning to Problems in Robotics (Y. Kodratoff and G. De Jong) and
Multi Strategy Learning (R. Michalski and G. Tecuci).
In the actual conference, Machine Learning was fairly well
represented (36 papers out of 116, 31% of the total number of
papers). Among them, 8 authors were European, a majority of them from
the Inductive Logic Programming community.
The papers in Machine Learning were divided into the following areas.
- Formalisation, which was quite well represented with three sessions
on Inductive Logic Programming (Lapointe & al, De Raedt & Bruynooghe,
De Raedt et al, Bergadano & Gunetti, Cameron - Jones & Quinlan, Ali &
Pazzani), algebraic formalisation of top down inductive methods with
the paper of Ganascia, and PAC oriented issues (Heath, Kasif &
Salzberg).
- Theory Revision, with two American papers (Wogulis & Pazzani,
Baffes & Mooney) of groups quite active in this field.
- Concept Acquisition (Ragavan et al, Rendel & Ragavan) and numeric
concept acquisition in the framework of Induction with Continuous
Attributes (Van de Merckt, Fayyad & Irani).
- Combinatorial problems, with the two papers of Minton and Ellman
whose system learn heuristics for Constraint Satisfaction Problems.
- Genetic algorithms : (Paredis) uses constraints for restricting GA
search, (Iba et al).
- Numeric techniques, with papers of Langley & Iba, Bailey & Iba ,
Cohen et al, Koyabashi et al.
- Combined Learning methods; a paper by Thrun & Mitchell describes
the integration of Neural Networks and Explanation Based Learning;
Botta & Giordana which integrates inductive learning with GA for
optimising learning on numeric attributes.
- Sequence Learning and production, with one paper describing
sequence learning with a Neural Networks approach - Case Based
Reasoning - Learning from the Environment - Search control -
Improving Behaviour - Learning and Statistics
From that list of subjects, a large range of ML aspects were covered
at IJCAI. One can notice that well established areas of learning
(induction - numeric techniques, concept acquisition in a
propositional or numeric framework, Explanation Based Learning) are
active. More recent and potential research directions are also
confirmed, namely: Machine Learning integration with other
disciplines (such as Constraint Logic Programming, Logic Programming,
Knowledge Acquisition) and, within Machine Learning, combination of
symbolic learning techniques with sub-symbolic ones (NN - GA).
Lorenza Saitta gave an invited talk entitled "Machine
Learning, Dream or Reality".
**********************************
Reinforcement Learning
Leslie Pack Kaelbling
Computer Science Department, Brown University
Box 1910 Providence, RI 02912-1910 USA
lpk@cs.brown.edu
1. Introduction
There is an apparent and growing need for computerized, intelligent,
autonomous systems. Such systems could be used to explore planets,
clean up hazardous sites, perform reconnaissance on land and under
water, and serve as useful assistants and transports in the home and
in industry. This need has been recognized and a great deal of effort
has gone into the construction and programming of a variety of
autonomous agents. But researchers and industrial developers have
found that the task of programming autonomous agents is more
problematic than other programming tasks.
Autonomous agents must react in real time to their changing external
environment. They cannot, in general, be thought of as executing a
linear or slightly conditional set of instructions. Rather, they must
choose actions based on the instantaneous state of the world,
sometimes abandoning entire courses of action in favour of new ones.
More recently, there has been a great deal of work in developing new
programming paradigms for autonomous systems [Brooks86, Firby87,
Kaelbling88a].These methods ameliorate the problems of programming
agents, but do not solve them.
Even with appropriate programming methodologies, human programmers
find it difficult to articulate their knowledge about how to take
action in the world, and in most cases, even if they could articulate
the knowledge, it would not be appropriate to the suite of sensors
and effectors of the autonomous system being programmed. To solve
this problem, we must build autonomous systems that learn the details
of their behaviour from high-level feedback about the world.
The need for learning is even more pronounced when we wish to build
autonomous agents that can operate in a variety of different
environments or in environments for which we do not have complete
specifications. In such cases, a fixed strategy cannot be obtained,
because correct behaviour will vary from run to run and occasionally
even from within a single run.
2. Reinforcement Learning
One way to view the problem of constructing learning behaviours for
agents is as a reinforcement learning problem. In reinforcement
learning, the goal of the agent's designer is for the agent to learn
what actions it should perform in which situations, in order to
maximize an external measure of success. All of the information the
agent has about the external world is a variety of inputs from the
environment. These inputs may encode information ranging from the
output of a vision system to a robot's current battery voltage. The
agent can be in many different states of information about the
environment, and it must map each of these information states, or
situations, to a particular action that it can perform in the world.
The agent's mapping from situations to actions is referred to as an
action map. Part of the agent's input from the world encodes the
agent's reinforcement, which is a scalar measure of how well the
agent is performing in the world. The agent should learn to act in
such a way as to maximize the total reinforcement it gains over its
lifetime.
As a concrete example, consider a simple robot with two wheels and
two photo-sensors. It can execute five different actions: stop, go
forward, go backward, turn left, and turn right. It can sense three
different states of the world: the light in the left eye is brighter
than that in the right eye, the light in the right eye is brighter
than that in the left eye, and the light in both eyes is roughly
equally bright. Additionally, the robot is given high values of
reinforcement when the average value of light in the two eyes is
increased from the previous instant. In order to maximize its
reinforcement, this robot should turn left when the light in its left
eye is brighter, turn right when the light in its right eye is
brighter, and move forward when the light in both eyes is equal. The
problem of learning to act is to discover such a mapping from
information states to actions.
Thus, the problem of learning to act can be cast as a function-
learning problem: the agent must learn a mapping from the situations
in which it finds itself, represented by streams of input values, to
the actions it can perform. In the simplest case, the mapping will be
a pure function of the current input value, but in general it can
have state, allowing the action taken at a particular time to depend
on the entire stream of previous input values.
In the past few years there has been a great deal of work in the
artificial intelligence (AI) and theoretical computer science
communities on the problem of learning pure Boolean-valued functions
[Haussler88a, Michalski83, Mitchell82, Quinlan83, Valiant84].
Unfortunately, this work is not directly relevant to the problem of
reinforcement learning because of the different settings of the
problem. In the traditional function-learning work, often referred to
in the AI community as "concept learning", a learning algorithm is
presented with a set of input-output pairs that specify the correct
output to be generated for that particular input. This setting allows
for effective function learning, but differs from the situation of an
agent trying to learn an action map. The agent, finding itself in a
particular input situation, must generate an action. It then receives
a reinforcement value from the environment, indicating how valuable
the current world state is for the agent. The agent cannot, however,
deduce the reinforcement value that would have resulted from
executing any of its other actions. Also, if the environment is
noisy, as it will be in general, just one instance of performing an
action in a situation may not give an accurate picture of the
reinforcement value of that action.
Reinforcement learning reduces to concept learning when the agent has
only two possible actions, the world generates Boolean reinforcement
that depends only on the most recently taken action, there is exactly
one action that generates the high reinforcement value in each
situation, and there is no noise. In this case, from performing a
particular action in a situation, the agent can deduce that it was
the correct action if it was positively reinforced; otherwise it can
infer that the other action would have been correct.
Reinforcement learning has its name because of its similarity to
models used in psychological studies of behaviour-learning in humans
and animals [Estes50]. It is also referred to as "learning with a
critic" [Widrow73], in contrast with the "learning with a teacher" of
traditional supervised concept learning.
3. Current State of the Art
The problem of learning from reinforcement has been studied by a
variety of researchers: statisticians studying the "two-armed bandit"
problem, psychologists working on mathematical learning theory,
learning-automata theorists, and researchers in artificial
intelligence. They have generated good solutions, both theoretical
and practical, to a number of problems within reinforcement learning.
We will consider the state of the art in particular areas of
reinforcement learning in the sections below.
3.1 Exploration
One of the most interesting facets of the reinforcement learning
problem is the tension between performing actions that are not well
understood in order to gain information about their reinforcement
values, and performing actions that are expected to have good results
in order to increase overall reinforcement. If an agent knows that a
particular action works well in a certain situation, it must trade
off performing that action against performing another one that it
knows nothing about but which might prove to be better. Or, as Ashby
[Ashby60] put it: the process of trial and error can thus be viewed
from two very different points of view. On the one hand it can be
regarded as simply an attempt at success; so that when it fails we
give zero marks for success.
From this point of view it is merely a second-rate way of getting to
success. There is, however, the other point of view that gives it an
altogether higher status, for the process may be playing the
invaluable part of gathering information, information that is
absolutely necessary if adaptation is to be successfully achieved.
The longer the time span over which the agent will be acting, the
more important it is for the agent to be acting on the basis of
correct information. Acting to gain information may improve the
expected long-term performance while causing short-term performance
to decline.
The trade-offs of exploration are modelled in a branch of statistics
called the theory of bandit problems [Berry85]. This work is largely
of theoretical interest, as the results require distributional
assumptions that are very difficult to justify. More practical
methods are given in the area of learning automata [Narendra89]. In
these methods, the choice of action is controlled by a vector of
probabilities, one associated with each action, which sums to 1. As
actions are taken in the world, the results are used to adjust the
probability vector. When it is time to take an action, it is
generated according to the probability distribution specified by the
vector. Fairly good results can be obtained in this way.
Better results have been obtained in more recent work [Lai87,
Kaelbling93], which uses the statistical notion of confidence
intervals to construct exploration algorithms. For each potential
action, the agent computes the upper bound of a confidence-interval
estimate of the expected value of taking that action, then takes the
action that has the highest upper bound. These methods are quite
effective: Lai's algorithm is computationally complex, but has been
shown to be asymptotically optimal; Kaelbling's algorithm is
efficient and empirical studies show its behaviour to be similar to
Lai's algorithm. These methods are an improvement over the
probability-vector methods because they incorporate an estimate of
the amount of information the agent has about each action as well as
about its expected reinforcement. Another important aspect of the
reinforcement-learning problem is that the actions that an agent
performs influence the input situations in which it will find itself
in the future. Rather than receiving an independently chosen set of
input-output pairs, the agent has some control over what inputs it
will receive and complete control over what outputs will be generated
in response. In addition to making it difficult to make
distributional statements about the inputs to the agent, this degree
of control makes it possible for what seem like small "experiments"
to cause the agent to discover an entirely new part of its
environment.
3.2 Temporal Credit Assignment
In many realistic environments, an agent will have to carry out a
number of actions before arriving at a state that has high
reinforcement value. How can that reinforcement be propagated
backward in order to reward temporally distant decisions? This
question is referred to as the problem of temporal credit assignment
or sequential decision making. It was originally solved in the
context of dynamic programming, with complex iterative algorithms
that run after a complete state-transition model of the world has
been built.
More recent methods [Barto89a, Sutton88, Watkins92] allow credit to
be propagated backward incrementally during the activity of the
agent. Watkins' Q-learning method works quite reliably, but must be
carefully coupled with a good exploration strategy.
3.3 Generalization
Many of the early algorithms for reinforcement require the
enumeration of all possible inputs to the agent. For interesting
real-world agents, the number of inputs can become enormous, causing
a combinatorial explosion in the time and space requirements for the
algorithms. In addition, such algorithms completely compartmentalize
the information they have about individual input situations. If they
learn to perform a particular action in a particular input situation,
that has no influence on what they will do in similar input
situations. In realistic environments, an agent cannot expect ever to
encounter all of the input situations, let alone have enough
experience with each one to learn the appropriate response. Thus, it
is important to develop algorithms that will generalize across input
situations. It is important to note, however, that in order to find
algorithms that are time and space efficient and that have the
ability to generalize over input situations, we must give up
something. What we will be giving up is the possibility of learning
any arbitrary action mapping. In the worst case, the only way to
represent a mapping is as a complete look-up table, which is what the
early reinforcement-learning algorithms do. There are many useful and
interesting functions that can be represented much more efficiently,
and the continuing research in this area must rest on the hope and
expectation that an agent can learn to act effectively in interesting
environments without needing action maps of pathological complexity.
Input generalization can be added to reinforcement-learning
algorithms by adopting function-approximation methods, such as radial
basis [Poggio89], CMAC [Albus81], and error back-propagation
[Rumelhart86b]. A more direct statistical approach, which creates a
decision tree by finding the most "relevant" input bits, was taken by
Chapman and Kaelbling [Chapman91] and shown to be effective in the
face of large numbers of irrelevant features with a wide range of
noise distributions.
3.4 Mappings With State
So far, we have considered learning only actions map that are pure,
instantaneous functions of their inputs. It is more generally the
case, however, that an agent's actions must depend on the past
history of input values in order to be effective. By storing
information about past inputs, the agent is able to induce a finer
partition on the set of world states, allowing it to make more
discriminations and to tailor its actions more appropriately to the
state of the world. Perhaps the simplest way to achieve this finer-
grained historical view of the world is to simply remember all input
instances from the last k time steps and present them in parallel to
the behaviour-learning algorithm. This method has two drawbacks: it
is not possible for actions to depend on conditions that reach back
arbitrarily far in history and the algorithmic complexity increases
considerably as the length of the available history is increased.
There have been no completely successful alternative approaches. One
suggestion, made by Kaelbling [Kaelbling93] and implemented with
limited success, is to learn mappings described by sequential boolean
networks, containing set-reset flip-flops for storage. This approach
is difficult, because there seem to be no good heuristics to drive
the construction of candidate networks. Another approach, used by
Drescher [Drescher91], is to learn mappings with state by first
learning world models with state, then using the world models to
construct action maps. More recent work by Chrisman and others
[Chrisman92] uses the techniques of partially observable markov
processes to build an update function for internal state variables,
then uses Q learning to choose actions based on inputs and the
internal state.
4. The Future of Reinforcement Learning
Initial experiments with tabula rasa learning systems have shown that
reinforcement learning methods, taken alone, are likely to be
impractical. They require an enormous amount of data to learn what to
do from a position of having no knowledge at all. To build practical
embedded agents, we will have to incorporate a priori knowledge
supplied by human programmers with on-line learning of situation-
specific knowledge by the agent. To do so will require the
investigation of the types of information that humans can most
easily, efficiently, and correctly supply and the development of
programming paradigms and languages that allow the easy intermixing
of programmer-supplied knowledge with learning algorithms.
References
- James S. Albus. Brains, Behavior, and Robotics. BYTE Books,
Subsidiary of McGraw-Hill, Peterborough, New Hampshire, 1981.
- W. Ross Ashby. Design for a Brain: The Origin of Adaptive
Behaviour. John Wiley and Sons, New York, New York, second edition,
1960.
- A. G. Barto, R. S. Sutton, and C. J. C. H. Watkins. Learning and
sequential decision making. Technical Report 89-95, Department of
Computer and Information Science, University of Massachusetts,
Amherst, Massachusetts, 1989. Also published in Learning and
Computational Neuroscience: Foundations of Adaptive Networks, Michael
Gabriel and John Moore, editors. The MIT Press, Cambridge,
Massachusetts, 1991.
- Donald A. Berry and Bert Fristedt. Bandit Problems: Sequential
Allocation of Experiments. Chapman and Hall, London, 1985.
- Rodney A. Brooks. A robust layered control system for a mobile
robot. IEEE Journal of Robotics and Automation, RA-2:14Q23, 1986.
- David Chapman and Leslie Pack Kaelbling. Input generalization in
delayed reinforcement learning: An algorithm and performance
comparisons. In Proceedings of the International Joint Conference on
Artificial Intelligence, Sydney, Australia, 1991.
- Lonnie Chrisman. Reinforcement learning with perceptual aliasing:
The perceptual distinctions approach. In Proceedings of the Tenth
National Conference on Artificial Intelligence, pages 183Q188, San
Jose, California, 1992. AAAI Press.
- Gary L. Drescher. Made-up Minds: A Constructivist Approach to
Artificial Intelligence. The MIT Press, Cambridge, Massachusetts,
1991.
- William K. Estes. Toward a statistical theory of learning.
Psychological Review, 57:94Q107, 1950.
- R. James Firby. An investigation into reactive planning in complex
domains. In Proceedings of the Sixth National Conference on
Artificial Intelligence, volume 1, pages 202Q206, Seattle,
Washington, 1987. Morgan Kaufmann.
- David Haussler. Quantifying inductive bias: AI learning algorithms
and Valiant's learning framework. Artificial Intelligence, 36(2):177Q
222, 1988.
- Leslie Pack Kaelbling. Goals as parallel program specifications. In
Proceedings of the Seventh National Conference on Artificial
Intelligence, Minneapolis-St. Paul, Minnesota, 1988.
- Leslie Pack Kaelbling. Learning in Embedded Systems. The MIT Press,
Cambridge, Massachusetts, 1993. Also available as a PhD Thesis from
Stanford University, 1990.
- Tze Leung Lai. Adaptive treatment allocation and the multi-armed
bandit problem. The Annals of Statistics, 15(3):1091Q1114, 1987.
- Ryszard S. Michalski. A theory and methodology of inductive
learning. In Ryszard S. Michalski, Jaime G. Carbonell, and Tom M.
Mitchell, editors, Machine Learning: An Artificial Intelligence
Approach, Chapter 4. Tioga, 1983.
- Tom M. Mitchell. Generalization as search. Artificial Intelligence,
18(2):203Q226, 1982.
- Kumpati Narendra and M. A. L. Thathachar. Learning Automata: An
Introduction. Prentice-Hall, Englewood Cliffs, New Jersey, 1989.
- Tomaso Poggio and Federico Girosi. A theory of networks for
approximation and learning. Technical Report AIM-1140, MIT Artificial
Intelligence Laboratory, Cambridge, Massachusetts, 1989.
- J. Ross Quinlan. Learning efficient classification procedures and
their application to chess end games. In Ryszard S. Michalski, Jaime
G. Carbonell, and Tom M. Mitchell, editors, Machine Learning: An
Artificial Intelligence Approach, chapter 15. Tioga, 1983.
- D. E. Rumelhart, G. E. Hinton, and R. J. Williams. Learning
internal representations by error propagation. In David E. Rumelhart
and James L. McClelland, editors, Parallel Distributed Processing,
volume 1, chapter 8. The MIT Press, Cambridge, Massachusetts, 1986.
- Richard S. Sutton. Learning to predict by the method of temporal
differences. Machine Learning, 3(1):9Q44, 1988.
- L. G. Valiant. A theory of the learnable. Communications of the
ACM, 27(11):1134Q1142, 1984.
- C. J. C. H. Watkins and P. Dayan. Q-learning. Machine Learning,
8(3):279Q292, 1992.
- Bernard Widrow, Narendra K. Gupta, and Sidhartha Maitra.
Punish/reward: Learning with a critic in adaptive threshold systems.
IEEE Transactions on Systems, Man, and Cybernetics, SMC-3(5):455Q465,
1973.
**********************************
Current Academic Research in Europe:
QUESTIONNAIRE
Dear Colleague,
at the Research Technical committee of the Machine Learning Network
of Excellence has been decided that it would be useful to build a
data base of people and groups working in machine learning in Europe.
Subsequently I have prepared the following questionnaire, which is
divided into two parts: Part A refers to the research group as a
whole; Part B refers to individual members of the group.
Completed questionnaires can be returned to me by regular mail, fax
or e-mail.
Lorenza Saitta
Universit di Torino
Dipartimento di Informatica
Corso Svizzera 185
10149 TORINO (Italy)
Tel: +39 11 742 9214; Fax: +39 11 751 603
e-mail: saitta@di.unito.it
The results of the questionnaires will be inserted into a data base,
available on Macintosh and Unix to all nodes of the network.
PLEASE FILL THE QUESTIONNAIRE AND DISTRIBUTE IT TO ANYONE YOU THINK
IS INTERESTED IN MACHINE LEARNING !!!
PLEASE RETURN AS SOON AS POSSIBLE
QQQQQQQQQQQQQQQQQQQQ
Part A - Research Team
QQQQQQQQQQQQQQQQQQQQ
1) Affiliation (Organization which the team officially belongs to)
2) Team Leader
- Name
- Position
3) List of permanent components of the team
4) Average number per year of temporary members of the team (undergraduate or
graduate students or others).
List of temporary members, if appropriate, indicating their position and the
period of charge inside the team.
5) Research themes (Less than 5 line of description for each one)
6) Implemented systems
For each one:
- Main characteristics
- Status (under development, prototype, commercial tool)
- Availability (Yes/No, Free, Equipment required, Distribution means)
7) Data sets
For each one:
- Main characteristics
- Availability (Yes/No, Free, Equipment required, Distribution means)
8) Relevant publications ( 2 20 for the whole team)
9) National and International projects related to ML field in which the team has
participated (Few line of description for each one)
QQQQQQQQQQQQQQQQQQ
Part B - Individual Researchers
QQQQQQQQQQQQQQQQQQQQ
1) Name
2) Position
3) Affiliation
4) Brief curriculum vitae
4) Research interest (list of keywords)
QQQQQQQQQQQQQQQQQQQQQ
Part C - Agreement to publication
QQQQQQQQQQQQQQQQQQQQQ
I undersigned, ............., am aware that the above information is public and
declare to agree to its diffusion into the academic and industrial milieu for non
profit purposes.
Date .......... Signature
......................
**********************************
Focus on Machine Learning Research
at the
University of Aberdeen
by Derek Sleeman
Aberdeen's Machine Learning group is part of a wider AI
activity (including Intelligent Robotics, Vision,
Qualitative Reasoning and Object Oriented Databases) in the
Computing Science Department. The group was formed in
1986, when Derek Sleeman moved to the Department; Peter
Edwards joined the group in 1988 as a Research Fellow and
became a lecturer in 1991; currently the group includes one
Research Fellow (Robin Boswell), one Honorary Research
Fellow (Susan Craw), one Research Assistant (Leo Carbonara)
and 12 Research Students.
The three main thrusts of the group have been:
- the development of the Machine Learning Toolbox project
(ESPRIT Project 2154)
- the development of Knowledge Base Refinement Systems
- the development of Scientific Aides.
Support for these activities has been provided by the EEC's
Science Program, ESPRIT R & D (MLT and VIVA projects), and
studentships from the UK's SERC.
Although MLT was an R & D project, [1], it did highlight a
number of important research issues, including the need to
view the application of Machine Learning to a task as a
series of phases, including pre-processing when the data is
massaged to fit the requirements, and a post-processing
phase where the domain expert often wishes to pose a series
of "what-if" queries. These phases can be conveniently
organized in a Workbench, eg WILA [2]. Secondly, we
realized that the usage of the inferred knowledge, should
have a major influence on the Machine Learning tool used.
This argues for a stronger coupling between Machine
Learning, Knowledge Acquisition and Problem Solving sub-
systems, and is the focus of Nicolas Graner's thesis work,
[3].
As a second activity, the group has been involved in
building Knowledge Base Refinement systems, since its
formation. The initial system implemented was REFINER,
which is able to detect inconsistencies in a series of
cases, and makes suggestions to the expert, as to how these
inconsistencies might be removed, [4]. KRUST, a further
refinement system, [5] represents its knowledge as a set of
production rules, and has the ability to refine the KB when
the expert indicates how the wrongly worked task should be
solved. To contain the potentially very large search
space, KRUST uses heuristics, background knowledge and a
series of "chestnut" cases1. The focus of the CREATOR
system by Ian Ellery was to infer new rules, hence it was a
successor of INFER [6] and complements KRUST which has a
simplistic mechanism for inferring new rules [7].
KRUST was originally developed using the wine selection
task. Currently it is being extended to work in
conjunction with several demanding real world tasks, as
part of the Viva project (Esprit project 6125).
Additionally, the enhanced KRUST system will form part of a
toolset and will need to interface and cooperate with a
variety of V&V2 tools.
The third thrust has been to apply the Knowledge Base
refinement techniques developed above to the demanding task
of building systems which can refine Scientific Knowledge
Bases; the challenges arise because of the size and
diversity of the domain knowledge. Current activities
include:
- building a relational learning system to refine a theory which
predicts coupling constants in NMR3 spectroscopy (Simos Metaxas,
[8]);
- Learning Apprentice Systems to help the expert interpret 2-D NMR
Spectra of proteins, [9] (Charalambos Tsatsarakis);
- a study to determine how professional botanists cope with surprise
when classifying specimens (Eugenio Alberdi, [10]);
- developing a framework for Informal Qualitative Models (Structural
Models)[11] to explain the historical evolution of theories for the
colligative properties of liquids (Adrian Gordon [12]);
- Exploratory Discovery in Science: a study which explores the
strengths of both CBR and Discovery in a single system; the domains
used are the hydrostatic model of fluids and electrical circuits
(Ruediger Oehlmann [13,14]).
- a study of the relationship between abstraction and the access and
mapping stages of analogical reasoning (Davide Roverso [15]).
Additionally, we are investigating the application of learning
techniques in the context of Distributed AI systems (Peter Edwards
[16]). Current projects include a study into the use of refinement
techniques in distributed problem-solvers, and the development of
methods for learning models of agent behaviour.
References:
1. Y KODRATOFF, D SLEEMAN, M USZYNSKI, K CAUSSE, S CRAW, 1992. Building a machine
learning toolbox. In: Enhancing the knowledge engineering process. L Steels, B
Lepape (Eds). North-Holland, Elsevier Science Publishers, pp 81 - 108.
2. C TSATSARAKIS, D SLEEMAN, 1993. A Prototype Workbench for the ID3 Algorithm.
In: Proceedings of the 4th Hellenic Conference on Informatics. Vol 1, pp 205 -
218.
3. N GRANER, D SLEEMAN, 1993. MUSKRAT: A Multistrategy Knowledge Refinement and
Acquisition Toolbox. In Proceedings of the Second International Workshop on
Multistrategy Learning, edited by R S MICHALSKI & G TECUCI, pp 107 - 119.
4. S SHARMA & D SLEEMAN, 1988. Refiner: A Case-based Differential Diagnosis Aide
for Knowledge Acquisition and Knowledge Refinement. In Proceedings of EWSL-88 D
Sleeman (Ed). Pitman: London, pp 201-210.
5. S CRAW & D. SLEEMAN, 1990. Automating the Refinement of Knowledge-Based
Systems. In Proceedings of ECCAI-90. Luigia Aiello (Ed). London: Pitman, pp 167-
172.
6. D SLEEMAN, H HIRSH, I ELLERY & IN-YUNG KIM, 1990. Extending Domain Theories:
Two Case Studies in Student Modelling. Machine Learning Journal, 5, pp 11-37.
7. D SLEEMAN, S CRAW, I ELLERY, S SHARMA, P EDWARDS, 1992. Machine Learning,
Knowledge Base Refinement & Knowledge Acquisition: systems which cope with
incomplete and inconsistent Knowledge Bases. Proceedings of EKAW '91, GMD Studien
NR. 211, pp 332 -342.
8. S METAXAS, 1993. The Prediction of Physical Properties with CRITON. In:
Working Notes of MLnet Workshop on Machine Discovery. P Edwards (Ed). Blanes,
Spain, pp 61 - 65.
9. P EDWARDS, D SLEEMAN, G C K ROBERTS & L Y LIAN, 1993. An AI Approach to
Interpretation of NMR Spectra of Proteins. In: AI & Molecular Biology, L Hunter
(Ed), AAAI/MIT Press, pp 396 - 432.
10. E ALBERDI, D SLEEMAN, 1993. Theory refinement & scientific classification: a
study in the domain of plant taxonomy. In: Working Notes of MLnet Workshop on
Machine Discovery. P Edwards (Ed). Blanes, Spain, pp 51-55.
11. D H SLEEMAN, M K STACEY, P EDWARDS & N A B GRAY, 1989. An Architecture for
Theory-Driven Scientific Discovery. In Proceedings of the Fourth EWSL. K Morik
(Ed). London: Pitman, pp 11-23.
12. A GORDON, 1993. Informal Qualitative Models and the Depression of Freezing
Points of Solutions. In: Working Notes of MLnet Workshop on Machine Discovery. P
Edwards (Ed). Blanes, Spain, pp 56 - 60
13. R OEHLMANN, D SLEEMAN, P EDWARDS, 1992. Self-Questioning & Experimentation in
an Exploratory Discovery System. Proceedings of the ML-92 Workshop on Machine
Discovery (MD-92), pp 41 - 50.
14. R OEHLMANN, D SLEEMAN, P EDWARDS, 1993. Learning plan transformations from
self-questions: a memory-based approach. In: Proceedings of the Eleventh National
Conference on Artificial Intelligence. Menlo Park, California: AAAI/MIT Press,
pp 520-525.
15. D ROVERSO, P EDWARDS, D SLEEMAN, 1992. Machine Discovery by Model-Driven
Analogy. In : Proceedings of the ML-92 workshop on Machine Discovery (MD-92), pp
87-89.
16. P EDWARDS & W H E DAVIES, 1993. A Heterogeneous Multi-Agent Learning System,
in Proceedings of the CKBS-SIG Workshop on Cooperating Knowledge-Based Systems,
DAKE Centre, University of Keele.
For further information contact:
Derek Sleeman
Department of Computing Science
King's College
University of Aberdeen
Aberdeen, Scotland, UK
AB9 2UE
Tel: +44 224 272288/2304
Fax: +44 224 273422
email: {sleeman, mlnet}@csd.abdn.ac.uk
_______________________
1 i.e. cases whose results are known and which all refined KBs must solve correctly.
2 Validation and Verification.
3 Nuclear Magnetic Resonance.
**********************************
ECAI'94 11th EUROPEAN CONFERENCE ON ARTIFICIAL
INTELLIGENCE
C A L L F O R P A R T I C I P A T I O N
AUGUST 8 - 12, 1994
AMSTERDAM RAI INTERNATIONAL EXHIBITION AND CONGRESS CENTRE, THE
NETHERLANDS
Organized by the European Coordinating Committee for Artificial
Intelligence (ECCAI); hosted by the Dutch Association for Artificial
Intelligence (NVKI). In cooperation with the AAAI WORKSHOPS.
A full workshop programme is planned for ECAI '94. This will take
place in the two days immediately before the main technical
conference, i.e., on August 8 and 9, 1994. They will give
participants the opportunity to discuss specific technical topics in
small, informal groups, which encourages interaction and exchange of
ideas.
Details of all workshops will be available by anonymous FTP from
cs.vu.nl, directory /pub/ecai94 by January 31, 1994; or via
electronic mail to ecai94-workshops@cs.vu.nl. It should be noted that
registration for the main conference will be required in order to
attend an ECAI '94 workshop.
TUTORIALS
A full tutorial programme will take place on August 8 and 9, 1994.
Extended tutorial information can be obtained by anonymous FTP from
swi.psy.uva.nl, directory/pub/ecai94.
SCHEDULE
Monday August 8, 1994
morning: T1, T2, T3 afternoon: T4, T5, T6
Tuesday August 9, 1994
morning: T7, T8, T9, T13a afternoon: T10, T11, T12, T13b
T1: MODELS OF UNCERTAINTY AND GRADUALITY IN AI
Instructors: Didier Dubois (National Center for Scientific Research, France)
Philippe Smets (Institut de Recherches Interdisciplinaires et de Developpements en
Intelligence Artificielle, France)
T2: THE KNOWLEDGE MEDIUM: THE USE OF FORMAL KNOWLEDGE REPRESENTATION FOR
INSTITUTIONAL MEMORY AND COMMUNICATION
Instructors: Thomas Gruber (Stanford University) Luc Steels (Free University
Brussels)
T3: INTELLIGENT MULTIMEDIA INTERFACES
Instructors: Mark Maybury (MITRE Corporation, Badford, MA, USA) Yigal Arens
(University of Southern California)
T4: REASONING WITH CASES: THEORY AND PRACTICE
Instructors: Klaus-Dieter Althoff (Technical University of Aachen) Michel Manago
(AcknoSoft, France) Stefan Wess (University of Kaiserslautern)
T5: THE ART AND THE SCIENCE OF MODELLING: CRUCIAL ISSUES IN BUILDING SECOND
GENERATION KNOWLEDGE-BASED SYSTEMS
Instructors: Peter Struss (Technical University of Munich) Bert Bredeweg
(University of Amsterdam)
T6: VALIDATION OF KNOWLEDGE-BASED SYSTEMS
Instructors: Pedro Meseguer (Technical University of Catalonia, Barcelona) Alun
Preece (University of Savoie, France)
T7: MANAGING MACHINE-LEARNING APPLICATION DEVELOPMENT AND ORGANISATIONAL
IMPLEMENTATION
Instructors: Yves Kodratoff (University of Paris-Sud) Vassilis Moustakis
(Technical University of Crete, Greece)
T8: KNOWLEDGE-BASED PRODUCTION MANAGEMENT
Instructors: Norman M. Sadeh (Carnegie Mellon's Robotics Institute, Pittburgh)
Stephen F. Smith (Carnegie Mellon's Robotics Institute, Pittburgh)
T9: MULTI-AGENT SYSTEMS AND DISTRIBUTED AI
Instructors: Les Gasser (University of Southern California) Jeffrey S. Rosenschein
(Stanford University)
T10: TEMPORAL REASONING IN AI
Instructors: Han Reichgelt (University of the West Indies in Mona, Jamaica) Lluis
Vila (Institute for Research in AI of Blanes, Spain)
T11: RULES IN DATA- AND KNOWLEDGE BASES
Instructors: Ulrike Griefahn (University of Bonn) Rainer Manthey (University of
Bonn)
T12: PRINCIPLES AND PRACTICE OF KNOWLEDGE ACQUISITION
Instructors: Angel R. Puerta (Stanford University) Henrik Eriksson (Linkoping
University)
T13a and T13b: ARTIFICIAL LIFE AND AUTONOMOUS ROBOTS
Instructors: Luc Steels (Free University Brussels) Dr McFarland (University of
Oxford)
ECCAI GRANT
The ECCAI Board has established a grant for East European researchers. Persons
interested in a grant are invited to contact Prof. J. Cuena, ECCAI Secretary,
Departamento de Intelligencia Artificial, Campus de Montegancedo s/n, E-28660
Boadilla del Monte [Madrid], Spain, fax:[+34] 1 352 4819, phone: [+34] 1 352 4803,
e-mail: jcuena@mayor.dia.fi.upm.es for details on the submission procedure.
EXHIBITION
An industrial and academic exhibition will be held from August 9 - 11, 1994.
Detailed information can be obtained at the Conference Office.
INFORMATION
For more information and registration fees please contact
CONFERENCE OFFICE:
Erasmus Forum c/o ECAI '94 Marcel van Marrewijk, Project Manager Mirjam de Leeuw,
Conference Manager Erasmus University Rotterdam P.O. Box 1738 3000 DR Rotterdam
The Netherlands Tel: (+31)-10-408.2302 Fax: (+31)-10-453.0784 E-mail:
M.M.deLeeuw@apv.oos.eur.nl
Andre' Nijenhuis, Expo Manager Phone: [+31] 1806 18314 Fax: [+31] 1806 17592
**********************************
Provisional Program of MLnet's
Workshop on Industrial Applications of Machine
Learning
FIRST CALL FOR PARTICIPATION
2nd-3rd September 1994
Dourdan (south neighbourhood of Paris)
Organizer: Yves Kodratoff (LRI & CNRS Orsay)
Registration fee: FF. 800. Bursaries may be available.
- Friday 2 Sep 1994: Overview presentations:
Ivan Bratko (Univ. Ljubljana): "On the state-of-the-art of industrial applications
of ML"
Gregory Piatetsky-Shapiro (GTE): "DDB: data mining in data bases"
Speaker to be announced: "Industrial applications of KADS2" .
Attilio Giordana (Univ. Torino): "Applications of ML to robotics"
Franz Schmalhofer (DFKI): "Unifying KA and ML for applications"
Vassilis Moustakis (Univ. Crete): "An overview of applications of ML to medicine"
- Saturday 3 Sep 1994
Special address: Setsuo Arikawa "Knowledge Acquisition from Protein Data by
Machine learning System BONSAI"
Results of ESPRIT projects:
- Nick Puzey (BAe) :"Industrial applications of MLT",
- Pavel Brazdil (Univ. Porto): "Industrial applications of STATLOG"
- Attilio Giordana (Univ. Torino): "The results of BLEARN"
Reports on current projects:
- "Application of GAs at Syllogics" (not yet confirmed)
- "ML at CSELT" (not yet confirmed)
- "Learning Rules about VLSI-Design" (not yet confirmed)
- "ML at Daimler-Benz" (not yet confirmed)
- "CBR at Matra-Space" (not yet confirmed)
Demos will take place during the evenings.
All participants in the Workshop are invited also to attend part or all of the
1994 Summer School which will take place at the same site in the following week.
Note however, that one needs also to register for the separate event; see next
column for details.
Request for information and registration forms are to be addressed
to:
Lola Canamero,
LRI, Bat. 490, Univ. Paris-Sud, F-91405, Orsay, France,
Tel. +33 1 6941 6462, Fax. +33 1 6941 6586,
email : lola@lri.fr
**********************************
Provisional Program of MLNet's
Summer School
FIRST CALL FOR PARTICIPATION
5th - 10th September 1994
Dourdan (south neighbourhood of Paris)
Organizer: Celine Rouveirol (LRI)
Monday Sept 5th
- Morning : Case base reasoning (A. Aamodt)
- Afternoon : Learning and Probabilities (W. Buntine)
Tuesday Sept 6th
- Morning : Learning and Noise (I. Bratko)
- Afternoon : Knowledge Acquisition (B. Wielinga)
Wednesday Sept 7th
- Morning : Integrated Architectures (L. Saitta)
- Morning : Knowledge Revision (S. Wrobel), (D. Sleeman)
Thursday Sept 8th
- Morning : Knowledge Acquisition and Machine Learning (M. van
Someren)
- Afternoon : ILP (S. Muggleton)
Friday Sept 9th
- Morning : ILP (F. Bergadano), (C. Rouveirol)
- Afternoon : Conceptual Clustering (G. Bisson)
Saturday Sept 10th
- Morning : Reinforcement Learning (L. Kaebling)
Invited seminars and demonstrations of software will be organised in
the evening.
Bursaries will be available for the Summer School.
Requests for information and registration forms are to be addressed
to
Lola Canamero,
LRI, Bat. 490, Univ. Paris-Sud, F-91405, Orsay, France,
Tel : +33 1 69 41 64 62, Fax : +33 1 69 41 65 86,
email : lola@lri.fr
**********************************
JUNIOR SCIENTIST FELLOWSHIPS - CALL FOR
APPLICATIONS
Scientific Programme "Learning in Human and Machines"
European Science Foundation
The goal of this recently launched programme is to advance our knowledge about
learning from an interdisciplinary perspective, bringing together - on a European
scale - researchers from cognitive science, in particular from the areas of
psychology, computer science, and educational science. The LHM initiative features
a Junior Scientist Fellowship programme, among other things. Scientists from all
European countries who are either working on their PhD now or have it completed
not longer than four years ago can apply for these fellowships. A fellowship will
allow a scientist to join LHM workshops and to visit a senior scientist's
laboratory. Twenty fellowships are available for the period 1994-1995. (Another
twenty fellowships are avai-lable for the period 1996-1997; a call for
applications for this second batch will be posted in due time.)
The LHM research programme is organzied into five Task Forces on the following
topics:
Task Force 1: Representation changes in learning
Task Force 2: Multi-objective learning
Task Force 3: Learning strategies to cope with sequencing effects
Task Force 4: Situated learning and transfer
Task Force 5: Collaborative learning
If you want to apply for a fellowship, please send in the following materials:
- curriculum vitae - course of studies - abstract of PhD thesis (if applicable) -
list of publications - one letter of recommendation.
Please also state which of the Task Forces mentioned above you wish to join, and
indicate how your research relates to the themes of this Task Force. (If
appropriate several Task forces may be specified.)
These materials should be sent to:
Prof. Dr. Hans Spada OR Prof. Dr. Lorenza Saitta
University of Freiburg Universita' degli Studi di Torino
Psychologisches Institut Dipartimento di Informatica
Niemensstr. 10 Corso Svizzera 185
D-79085 Freiburg/Germany 10149 Torino/Italy
For additional information concerning the programme, please inquire either by fax
+49 761 203 2490 or by electronic mail (esf-lhm-info@psycholo-gie.uni-
freiburg.de). You can also go to the directory /home/cogsys/ftp/pub/esf-lhm on the
machine esf-lhm.psychologie.uni-freiburg.de and access the (short) file ESF-LHM-
Summary.txt and/or the (long) postscript file ProgrammeProposal.ps by anonymous
ftp.
The deadline for applications is 10 May 94. Only scientists from European
countries (including eastern Europe) can apply.
**********************************
Procedures for joining MLnet
Initial enquiries will receive a standard information pack (including a copy of
the Technical Annex)
All centres interested in joining MLnet are asked to send the following to MLnet's
Academic Coordinator:
- A signed statement on Institutional notepaper saying that you have read and
agreed with the general aims of MLnet given in the Technical Annex;
- One hard-copy document listing the Machine Learning (and related
activities) at the proposed node and three copies of any enclosures; the
document should include a list of scientists involved in these field(s),
half page curriculum vitae for each of these senior scientists, current
research students, lists of recent grants and relevant publications over
the last 5 year period;
- A statement of the Technical Committees which the Centre would be
interested in joining, and a succinct statement of the potential
contributions of the Centre to the Network and its Technical Committees.
Two members of the Management Board will be asked to look at the material in
detail and will present the proposal at the next Management Board meeting.
Through the Network's Coordinator, the members may ask for additional information.
The Academic Coordinator will be in touch with the Centre as soon as possible
after the Management Board meeting.
The Management Board is not planning to set a fixed timetable for applications,
but advises potential nodes that it currently holds Management Board meetings in
November, April and September, and that papers would have to be received at least
six weeks before a Management Board meeting to be considered.
**********************************
Main Nodes
Professor D Sleeman, Aberdeen University (GB)
Tel No: +44 224 27 2288/2304
Fax No: +44 224 27 3422
Professor B J Wielinga, University of Amsterdam (NL)
Tel No: +31 20 525 6789/6796
Fax No: +31 20 525 6896
Professor R Lopez de Mantaras, IIIA AIRI,
Blanes (ES)
Tel No: +34 72 336 101
Fax No: +34 72 337 806
Professor K Morik, Dortmund University (DE)
Tel No: +49 231 755 5101
Fax No: +49 231 755 5105/2047
Dr L DeRaedt, Leuven Katholieke Universiteit (BE)
Tel No: +32 16 20 10 15
Fax No: +32 16 20 53 08
Dr Y Kodratoff, Paris Sud University, Orsay (FR)
Tel No: +33 1 69 41 69 04
Fax No: +33 1 69 41 65 86
Professor L Saitta, Torino University (IT)
Tel No: +39 11 742 9214/5
Fax no: +39 11 751 603
Mr C. de Maindreville, Alcatel Alsthom Recherche (FR)
Tel No: +33 1 6449 1476
Fax No: +33 1 6449 0695
Mr T Parsons, British Aerospace plc (GB)
Tel No: +44 272 363 458
Fax no: +44 272 363 733
Mr F Malabocchia, CSELT S.p.A. (IT)
Tel No: +39 11 228 6778
Fax No: +39 11 228 5520
Dr D Cornwell, CEC Project Officer
Tel No: +32 2 296 8664/8071
Fax No: +32 2 296 8390/8397
Associate Nodes
- ARIAI, Vienna (AT) - Bari University (IT) - Bradford University
(GB) - Catania University (IT) - Coimbra University (PT) - CRIM-ERA,
Montpellier (FR) - FORTH, Crete (GR) - Frankfurt University (DE) -
GMD, Bonn (DE) - Kaiserslautern University (DE) - Karlsruhe
University (DE) - Ljubljana AI Labs (SL) - Nottingham University (GB)
- Oporto University (PT) - Paris VI University (FR) - Pavia
University (IT) - Reading University (GB) - Savoie University,
Chambery (FR) - Stockholm University (SE) - Tilburg University (NL)
-JTrinity College, Dublin (IE) - Ugo Bordoni Foundation, Roma (IT) -
VUB, Brussels (BE) - ISoft (FR) - Matra Marconi Space (FR) - Siemens
AG (DE)
Academic Coordinator:
Derek Sleeman
Department of Computing Science
University of Aberdeen
King's College
Aberdeen AB9 2UE
Scotland, UK
Tel: +44 224 27 2288/2304
Fax: +44 224 27 3422
email: {mlnet, sleeman}@csd.abdn.ac.uk
Documents available from Aberdeen:
State of the Art Overview of ML and KA
Recently Announced projects (ESPRIT III)
MLnet Flyer
**********************************
MLnet NEWS 2.2 END
**********************************
