--
Discovery in Databases (KDD) community, focusing on the latest research and
applications.
Submissions are most welcome and should be emailed,
with a DESCRIPTIVE subject line (and a URL, when available) to kdd@gte.com
To subscribe, email to kdd-request@gte.com
message with
subscribe kdd-nuggets
in the first line (the rest of the message and subject are ignored).
See http://info.gte.com/~kdd/subscribe.html
for details.
Nuggets frequency is approximately 3 times a month.
Back issues of Nuggets, a catalog of S*i*ftware (data mining tools),
and a wealth of other information on Data Mining and Knowledge Discovery
is available at Knowledge Discovery Mine site http://info.gte.com/~kdd
-- Gregory Piatetsky-Shapiro (editor)
********************* Official disclaimer ***********************************
* All opinions expressed herein are those of the writers (or the moderator) *
* and not necessarily of their respective employers (or GTE Laboratories) *
*****************************************************************************
~~~~~~~~~~~~ Quotable Quote ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Out of Sight, Out of Mind (a proverb)
Absence makes the heart grow fonder (another proverb)
Absence diminishes little passions and increases great ones,
as wind extinguishes candles and fans a fire.
--Francois de la Rochefoucauld
(a reconciliation of the first two) Previous1NextTop
>~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
From: Usama Fayyad (fayyad@MICROSOFT.com)
Subject: followup: first mention of data mining
Date: Fri, 13 Dec 1996 09:18:11 -0800
Content-Length: 1161
In reply to janathan Hoskings 'oldest' reference
to data mining, I can site an even older reference
in stats:
A book published in 1978 by Wiley:
Leamer, Edward, E. Specification searches: ad hoc inference
with nonexperimental data, Wiley, 1978
The book starts with references to 'data mining' and Leamer
actually uses it in a nonderogative sense (but first use is to
acknowledge its use as a derogative term in traditional stats
of that time. See intro chapter.
[Thanks to Chris Meek for pointing me to it].
(the same book was cited by Chris Meek in his message below. GPS)
From: Chris Meek (meek@microsoft.com)
Subject: An earlier mention of data mining (WAS : first mention of data mining)
Date: Mon, 16 Dec 1996 14:58:05 -0800
Content-Length: 1130
An early mention of 'data mining' can be found in Ed Leamer's book
'Specification Searches: Ad Hoc Inference with Nonexperimental Data'
(1978), Wiley. Like the reference given by Jon Hosking in the last
KDD-Nuggets, the topic is Econometrics. However, in this reference,
despite the subtitle, the term 'data mining' is not used pejoratively as
the quote below indicates.
[from the introduction of Leamer (1978)]
....This book is about 'data mining.' It describes how specification
searches can be legitimately used to bring to the surface the nuggets of
truth that may be buried in the data set....
Chris Meek
Microsoft Research
meek@microsoft.com
Previous2NextTop
>~~~Publications:~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Subject: Papers on overfitting and model complexity
Date: Mon, 9 Dec 96 14:15:33 -0400
From: David Jensen (jensen@cs.umass.edu)
The University of Massachusetts' Experimental Knowledge Systems
Laboratory announces the availability of three related papers on
overfitting and model complexity. The papers are:
They will appear in the Preliminary Papers of the Sixth
International Workshop on Artificial Intelligence and Statistics,
January 1997.
The first paper demonstrates that a class of induction algorithms
produce overfitting and relates the problem to the well-known
statistical theory of multiple comparisons. The second paper presents
a tree induction algorithm that adjusts for multiple comparisons, and
evaluates the algorithm's performance on artificial and realistic data
sets. The final paper reports on experiments demonstrating a strong
relationship between training set size and decision tree complexity
for several common pruning methods.
Previous3NextTop
>~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
From: th03%bacon@gte.com
Date: Tue, 17 Dec 1996 09:30:11 -0500
Subject: New book on Rough Sets and Data Mining
Content-Length: 2485
New book on Rough Sets and Data Mining
======================================
Title: Rough Sets and Data Mining : Analysis of Imprecise Data
edited by T.Y. Lin and N. Cercone
Publisher : Kluwer Academic Publishers
ISBN number 0-7923-9807-6
Table of Contents
Part I: Expositions
1. Rough Sets Z. Pawlak
2. Data Mining: Trends in Research and Development, J. Deogun, V.
Raghavan, A. Sarkar, and H. Sever
3. A Review of Rough Set Models, Y. Y. Yao, S. K. M. Wong, T. Y. Lin
4. Rough Control: A Prospective, T. Munakata
Part II Applications
5. Machine Learning & Knowledge Acquisition, and The English Semantic
Code, J. Grzymala-Busse, S. Y. Sedelow, W. A. Sedelow
6. Generation of Multiple Knowledge from Databases Based on Rough Set
Theory, X. Hu, N. Cercone, W. Ziarko
7. Fuzzy Controllers:An Integrated Approach Based on Fuzzy Logic,
Rough Sets, and Evolutionary Computing, T. Y. Lin
8. Rough Real Functions and Rough Controllers, Z. Pawlak
9. A Fusion of Rough Sets, Modified Rough Sets, and Genetic
Algorithms for Hybrid Diagnostic Systems, R. Hashemi, B. Pearce, R Arani,
W. Hinson, M. Paule
10. Rough Sets as a Tool for Studying Attribute Dependencies in The
Urinary Stones Treatment Data Set J. Stefanowski, K. Slowinski
Part III. Related Areas
11. Data Mining Using Attribute-Oriented Generalization and Information
Reduction, N. Cercone, H. Hamilton, X. Hu and N. Shan
12. Neighborhoods, Rough Sets, and Query Relaxation in Cooperative
Answering, B. Miachael, T. Y. Lin
13. Resolveing Queries Through Cooperation in Multi-Agent Systems, Z. Ras
14. Synthesis of Decision Systems From Data Tables, A. Skowron and L.
Polkowski
15. Combination of Rough and Fuzzy Sets Based on (-level Sets, Y. Y. Yao
16. Theories That Combine Many Equivalence and Subset Relations, J.
Zytkow, R. Zembowicz
Part IV Generalization
17. Generalized Rough Sets in Contextual Spaces, E. Bryniarski and U.
Wybraniec-Skardowska
18. Maintenance of Reducts in The Variable Precision Rough Set Model,
M. Kryszkiewicz
19. Probabilistic Rough Classifiers with mixture of Discrete and
Continuous Attributes, A. Lenarcik and Z. Piasta
20. Algebraic Formulation of Machine Learning Methods Based On Rough
Sets, Matroid Theory, and Combinatorial Geometry, S. Tsumoto and H.
Tanaka
21. Topological Rough Algebras, A. Wasilewska
Previous4NextTop
>~~~Siftware:~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
From: bissantz@forwiss.uni-erlangen.de
Date: Wed, 11 Dec 96 14:28:21
Subject: new award-winning data mining software
MIS AG introduces new Data-Mining-Tools.
BETREX awarded as one of the 'Best Business Management Solutions'
This year the American trade journal 'BYTE' awarded for the first time
in the competition 'Best of Systems Award' the best products in
various categories. In the category 'Best Business Management
Solution' the Data-Mining-Tool BETREX from Bissantz Kueppers and
Company, Erlangen was presented as one of the three best new Products.
BETREX expands as 'Delta Miner' the solution path Delta Solutions of
MIS AG and at the beginning of 1996 will be introduced onto the
market.
- Better Information for the Controller Delta Miner is a navigation
instrument developed specifically for controlling problems. It guides
the user automatically through complex data to the most important
unusual data.
In contrast to the drill-down-possibilities of common EIS and MIS
tools, Delta Miner is equipped with helpful automations. The system
suggests independently the best analysis path and thus relieves the
controller of tedious activities. In addition to identifying the main
business developments, the tool indicates their causes. A
contribution margin flow analysis locates and informs the user of
possibilities for initiating preventive measures.
In the middle of next year the product line will be expanded. The
primary characteristic will be a new approach for Data Mining. The
Delta-Miner production line refrains consistently from using
procedures whose mechanisms are difficult for the user to deduce.
Instead typical routine activities, which are tedious and time
consuming for the user, are handled entirely by the computer. Thus
more time remains for the actual tasks of the analysts: the
interpretation of results and the preparation of decisions.
- Successful Research Transfer
The basic procedures of the
Delta-Miner-Tools were developed at FORWISS, the Bavarian Research
Center for Knowledge Based Systems. There, since 1993 intensive
research has been done in the research group Information Systems lead
by Prof. Peter Mertens on algorithms for Data Mining.
- Enormous Time Savings
In several application tests with industry
partners a dramatic acceleration of common analysis activities was
proven. For example it was possible to reduce the study of market
research data from six days to three hours. In addition, new findings
were made, which remain when using common procedures as a rule
unknown. For further information please refer to Dr. Nicolas
Bissantz, Bissantz Kueppers & Company, Erlangen, Germany
(bissantz@forwiss.uni-erlangen.de).
best regards
nick
-------------------------------------------------------------------------------------------
Dr. Nicolas Bissantz, Geschaeftsfuehrung
BISSANTZ KUEPPERS & COMPANY
Advanced Information Management
Previous5NextTop
>~~~Positions:~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Date: Fri, 13 Dec 1996 05:56:17 -0800
From: Emmanuel Fourakis (Manos.Fourakis@worldnet.att.net)
To: kdd@gte.com
Subject: Positions
Content-Length: 876
I would like to announce the following positions:
a. One-two openings for graduate level interns with Annapolis, MD
software company for algorithm and code development. Positions are
initially part-time, however the company is looking for a long-term
association.
b. One opening for a graduate level intern with Herndon, VA
business consulting company for market research and marketing plan
development. Need business school background but also understanding of
knowledge tools and methodologies. This position is also part-time at
first and it is with my company, The AEF Group. We provide business
development consulting to technology companies.
If anyone is interested, please contact me at:
manos.fourakis@worldnet.att.net
or
at 703/404-1481, fax: 703/404-3782.
Thank you,
Manos Fourakis
The AEF Group, Inc.
12418 Willow Falls Drive
Herndon, VA 20170
Previous6NextTop
>~~~Meetings:~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
From: Sven.Koenig@A.GP.CS.CMU.EDU
Date: Wed, 11 Dec 96 03:42:02 EST
To: genetic-programming@CS.STANFORD.EDU,
hybrid-list@CS.UA.EDU,
ilpnet@IJS.SI,
kdd@gte.com
Subject: AAAI-97 Workshop on On-Line Search
==================================
AAAI-97 Workshop on On-Line Search
==================================
The AAAI-97 Workshop on On-Line Search is part of the Fourteenth
National Conference on Artificial Intelligence. The conference will be
held from July 27-31, 1997 in Providence, Rhode Island, and is is
sponsored by the American Association for Artificial Intelligence
(AAAI). The workshop will be held on either July 27 or July 28.
Description:
On-line search is driven by the need to commit to 'actions' before
their complete consequences are known. An 'action,' in this context,
can correspond to such diverse things as making a move in a two-player
game, moving a robot, or allocating some resource (such as a page in a
cache). On-line search can be necessary for a variety of reasons:
there may be missing domain knowledge that has to be acquired
actively, the domain may be known but so large that it cannot be
searched completely in a reasonable amount of time, or it may simply
be that the consequences of one's actions depend on the behavior of
some other entity. On-line search can also reduce the sum of planning
and execution time.
The on-line search paradigm underlies many applications and has been
independently investigated in
- artificial intelligence (single-agent search and two-player games),
- robotics (path planning and execution), and
- theoretical computer science,
among others. This has resulted in the development of a variety of
on-line search approaches including
The AAAI-97 workshop on On-Line Search is intended to bring together
researchers from different fields who investigate on-line search
approaches. Our goal is to learn about the different methods,
assumptions, and results, and to enable the transfer of ideas between
the different fields.
Topics:
Questions that we hope to see addressed during the workshop include,
but are not limited to
- what information to gather in the limited time available,
- when to stop collecting information and commit to an action, and
- what action to commit to given the information collected.
We are especially interested in the trade-offs between
- thinking versus acting, and
- acting to exploit existing knowledge versus acting to acquire further knowledge.
We are also interested in applications of on-line search algorithms as
well as empirical and formal results that explain how properties of
the tasks and domains influence the behavior and efficiency of on-line
search algorithms.
Workshop Format:
The workshop will consist of two invited talks, followed by short
presentations and longer discussions, in an atmosphere that encourages
the interaction of researchers with different backgrounds. There will
be plenty of opportunity to discover common ground between different
fields and to speculate on how methods from one field could be applied
to another.
Attendance:
Researchers from all fields are encouraged to submit papers, including
researchers from artificial intelligence, robotics, and theoretical
computer science. Workshop participants will be selected based on the
quality of their submissions and relevance to the topic.
Submission Requirements:
We are soliciting both papers about original research and papers that
give overviews of classes of on-line search algorithms or application
areas (up to eight pages in AAAI format). Persons desiring to
participate, but not to give a presentation, are encouraged to submit
position papers (up to two pages). If you inform us well before the
submission deadline of your intentions, you help us to plan better. We
encourage e-mail submissions in Postscript. Hardcopy submissions
should be provided in five copies.
Submission Deadline: March 11, 1997
Notification Date: April 1, 1997
Final Date for Camera-Ready Copies: April 22, 1997
Submission Address:
Sven Koenig
Carnegie Mellon University
Department of Computer Science
Pittsburgh, PA 15213-3891, USA
phone: (412) 268-3076
fax: (412) 268-5576
e-mail: skoenig=search@cs.cmu.edu
Organizing Committee:
Avrim Blum, Carnegie Mellon University, avrim@cs.cmu.edu
Sven Koenig, Carnegie Mellon University, skoenig@cs.cmu.edu
Richard Korf, University of California at Los Angeles, korf@cs.ucla.edu
Toru Ishida, Kyoto University, ishida@kuis.kyoto-u.ac.jp
The 9th European Conference on Machine Learning (ECML-97) will be held
in Prague, Czech Republic, during April 23-26, 1997, with a day of
workshops on April 26.
The conference proceedings will be published by Springer Verlag, Berlin.
PROGRAM CHAIRPERSONS:
Maarten van Someren (University of Amsterdam)
Gerhard Widmer (Austrian Research Institute for AI).
LOCAL CHAIR:
Radim Jirousek (University of Economics, Prague).
INVITED SPEAKERS:
Stuart Russell (University of California at Berkeley)
Paul Vitanyi (Universiteit van Amsterdam)
Luc Steels (Vrije Universiteit Brussel)
ECML-MLNET WORKSHOPS:
The ECML-MLNET workshops will be held on 26 April 1997. There is a uniform
time table for all workshops:
Paper submissions: 15 February 1997
Acceptance: 8 March 1997
Camera ready copy: 1 April 1997
WS1: Empirical Learning of Natural Language Processing Tasks
Organisers:
Walter Daelemans (Chair); Tilburg University; Walter.Daelemans@kub.nl
Ton Weijters; Universiteit Maastricht; weijters@cs.unimaas.nl
Antal van den Bosch; Universiteit Maastricht; antal@cs.unimaas.nl
WS3: Case-Based Learning: Beyond Classification of Feature Vectors
Organizers:
Dietrich Wettschereck (GMD; German National Research Center for
Information Technology, Germany)
David Aha (Navy Center for Applied Research in AI, USA)
Contact person:
Dietrich Wettschereck (dietrich.wettschereck@gmd.de)
For call for papers of a workshop please use the workshop web page or
contact the contact person of a workshop. For general information about the
ECML-MLNET workshops please contact Maarten van Someren
(maarten@swi.psy.uva.nl).
Previous8NextTop
>~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Date: Thu, 12 Dec 1996 14:24:24 +0100 (MET)
From: Hendrik Blockeel (Hendrik.Blockeel@cs.kuleuven.ac.be)
To: compunode@ecrc.de,
ml@ics.uci.edu,
kdd@gte.com,
mlnet@csd.abdn.ac.uk
Subject: CFP IJCAI97 Workshop on Frontiers of ILP
Cc: Luc.DeRaedt@cs.kuleuven.ac.be
Content-Length: 7259
The IJCAI-97 one day workshop on 'Frontiers of ILP' in Nagoya, Japan,
will take place in the period August 23-25, immediately prior to
the start of the main IJCAI conference. The exact date is to be determined.
TECHNICAL DESCRIPTION
Inductive logic programming (ILP) is a recent subfield of
artificial intelligence that studies the induction of first order formulae
from examples. The purpose of this workshop is twofold:
on the one hand, we wish to widen the scope of ILP
by investigating its relations to neighboring fields,
and on the other hand, we wish to make ILP more accessible
for researchers from neighboring fields.
The workshop therefore solicits papers
that lie at the frontiers of ILP with neighboring fields.
A non-exclusive list of interesting topics for the workshop includes :
* ILP and Software Engineering:
what has ILP to offer to Software Engineering ?,
and in what way can Software Engineering help to design ILP systems
and applications ?
* ILP for Knowledge Discovery in Databases : ILP aims
at learning complex rules involving multiple relations from small
databases, whereas KDD typically induces simple rules about a
single relation from a large database. Furthermore, ILP allows to
exploit background knowledge in a variety of ways. Can KDD and ILP be
succesfully combined ?
* ILP and Computational or Algorithmic Learning Theory :
though many results have been obtained concerning the learnability
of inductive logic programming, most of the results are negative
and most of the positive results are reducible to propositional learning
methods. Is there a mismatch of COLT with ILP ? and if so,
what can be done about it ?
* ILP versus propositional learning methods :
Since the very start of ILP, researchers and practioners of
machine learning have wondered about the relation between
ILP and propositional learning methods. Theoretical and experimental
questions that arise include:
when to use ILP and when to use propositional learning methods ?
under what circumstances can ILP be reduced to propositional learning ?
what is the price to pay for using first order logic in
terms of efficiency ?
* ILP and Knowledge Representation : ILP has traditionally employed
computational logic to represent hypotheses and observations.
Alternative well-founded knowledge representation formalisms have received
little attention (with the exception of CLASSIC).
What can ILP learn from Knowledge Representation ?
and in what well-founded Knowledge Representation formalisms
is induction feasible ?
* ILP in multistrategy learning : Multistrategy learning
combines multiple learning strategies. What role can ILP
play for multistrategy learning ?
* ILP and Probabilistic reasoning: in contrast to
propositional learning methods, ILP has not used
probabilistic representations. How can ILP incorporate
such representations ? and how can it interact with
methods such as Bayes nets or Hidden Markov Models ?
* ILP for Intelligent Information Retrieval:
The rapid development of
the World Wide Web has spawned significant interest in intelligent
information retrieval. In particular, the need for algorithms for
reliably classifying textual documents into given categories (like
interesting/uninteresting) be useful for a wide variety of tasks.
Currently, most learning algorithms are not able to make use of
structural information like word order, succesive words, structure of
the text, etc. Can ILP algorithms offer advantages over conventional
information retrieval or machine learning algorithms for this sort of
tasks?
* Applications of ILP in subfields of AI : ILP has been applied
to other subfields of AI, including natural language processing,
intelligent agents and planning.
Further applications of ILP within AI are solicited.
Both position papers about the relation of ILP to other fields, as well
as research papers that make specific techical contributions
are solicited. However, to stimulate discussion, it is expected
that each technical paper also clarifies the position
of ILP with regard to the neighboring field(s) it addresses.
Except for the presentation of position and technical papers,
the workshop will also feature a panel discussion
on the frontiers of ILP and possibly an invited talk.
ORGANISERS
Luc De Raedt (chair and primary contact)
Saso Dzeroski
Koichi Furukawa
Fumio Mizoguchi
Stephen Muggleton
PROGRAMME COMMITTEE
Francesco Bergadano (Italy)
Luc De Raedt (co-chair, Belgium)
Saso Dzeroski (Slovenia)
Johannes Furnkranz (Austria)
Koichi Furukawa (Japan)
David Page (U.K.)
Fumio Mizoguchi (Japan)
Ray Mooney (U.S.A.)
Stephen Muggleton (co-chair, U.K.)
CALL FOR PARTICIPATION
Participation is open to all members of the AI Community.
However, to encourage interaction and a broad exchange of ideas
the number of participants will be strictly limited
(preferably under 30 and certainly under 40).
Participants will be selected on the basis of submissions.
Three types of submissions will be considered :
1) technical contributions (ideally, a 3 to 5 page extended abstract,
in the IJCAI Proceedings Format, 3000-4000 words),
2) position papers (ideally, a 1 to 3 page abstract
in the IJCAI Proceedings Format, 1000 - 3000 words)
3) a statement of interest (ideally, a one page motivation of why you
would like to participate, 300- 500 words)
Only submissions of type 1) and 2) will be considered
for presentation at the workshop and inclusion in the workshop notes.
Submissions should be received no later than April 1, 1997,
and must include first author's complete contact information,
including address, email, phone, and fax number.
Double submissions with the ILP-97 Workshop (which is to take
place in Prague, September 1997) are allowed.
SUBMISSIONS
Submit papers by email (postscript) and surface mail (2 copies) to
Luc De Raedt
Dept. of Computer Science
Katholieke Universiteit Leuven
Celestijnenlaan 200A
B-3001 Heverlee
Belgium
Email : Luc.DeRaedt@cs.kuleuven.ac.be
IMPORTANT DATES
- Paper submission : 1 April
- Notification to Authors : 21 April
- Camera ready copy : the submissions themselve
will serve as camera ready copy
(submissions in the IJCAI Proceedings Style are strongly preferred,
see http://www.ijcai.org/ijcai-97/
for details)
PUBLICATION
The accepted submissions will be included in the workshop notes
to be distributed at the workshop.
Post-conference publication of a selection of the workshop papers
will be considered and discussed at the workshop.
COSTS
To cover costs, a fee of $US 50 will be charged,
in addition to the normal IJCAI-97 conference registration fee.
Attendees of IJCAI workshops will be required to register
for the main IJCAI conference.
Tutorials for both the International Conference on Machine Learning
and the Conference on Computational Learning Theory will take
place during ICML and COLT, which will be co-located at Vanderbilt
University during the week of July 6-12, 1997.
We invite tutorials in all areas related to the theoretical and/or
experimental study of machine learning. We also invite tutorials from
other disciplines engaged in research in closely related fields such
as statistics, information theory, pattern recognition, statistical
physics, and information retrieval. Tutorials should be introductory
in nature and geared towards the broadest audience possible. Further,
tutorials are expected to introduce a broad body of research, rather
than focusing solely on the tutorial presenter's own personal
research.
All proposal must be received by Friday, February 7, 1997 (or set
via airmail and postmarked by January 29, or sent via overnight
carrier by February 6.) More information about the tutorials as well
as information about submitting a tutorial proposal can be found at
URL http://www.cs.wustl.edu/~sg/tutorial96.html.
If you have any questions contact Sally Goldman (sg@cs.wustl.edu).
KDD Nuggets Index