[Apologies if you receive multiple copies of this announcement.]
--------------------------------------------------------------------------
First International Workshop on
Data Mining Lessons Learned
(DMLL-2002)
http://www.hpl.hp.com/personal/Tom_Fawcett/DMLL-workshop.html
An ICML-2002 Workshop, Sydney, 8-12 July 2002
http://www.cse.unsw.edu.au/~icml2002/
FIRST CALL FOR PAPERS
Description
Data mining is concerned with finding interesting/valuable patterns in
data. Many new techniques are emerging that enable analysis and
visualization of large volumes of data of various types. What we see
in technical papers are mostly success stories of these new
techniques. However, a typical paper accepted to a ML or DM conference
shows the advantages of the new method compared to other methods by
reporting only marginal accuracy improvements on standard benchmark
data, mostly from the UCI repository. In rude terms, one can claim
that many of the recent ML and DM developments in classifier induction
are simply overfitting UCI datasets.
Journal, conference and workshop papers seldom report on failures, yet
reported failures are crucial for increasing awareness of which
approach to use in which situations. At conferences, we often hear of
successful applications; but we rarely hear of steps leading to
success, steps that failed, and representation choices that were
critical to success. In addition, papers rarely include expert
evaluations of results. Fortunately, recent years have seen an
increase in popularity of data mining challenge problems, such as the
KDD Cup, the COIL challenge and the PTE challenge. Since they pose
specific problems to data mining researchers and provide expert
evaluation of solutions, the can serve as a source of such experience.
The purpose of this workshop is to assemble data mining practitioners
and to gather experience from successful and unsuccessful data mining
endeavors. We seek to learn from successes (evaluated by domain
experts) and, most importantly, from failed data mining endeavors (why
was an approach not successful, what can dataminers learn from this
experience). The main aim is to start gathering the lessons learned.
Goals
The aim of this workshop is to collect the experience gained through
different challenge competitions as well as the lessons learned from
the failed and successful approaches to solving particular data mining
applications. Authors are invited to report on lessons learned from
submissions to challenge problems, experiences in engineering the
representations to be particularly useful for the problem at hand,
what kind of output the expert evaluating the solutions liked (e.g.,
should it be natural language) and how intelligible the output was to
the expert, why some particular solutions - despite their good
predictive accuracy - never made it to practice, obtained results
which were not originally expected before solving the problem.
An ideal contributions to this workshop would describe one problem in
sufficient detail, either an application or a challenge problem.
- For applications, we expect authors to describe approaches which
failed, why they failed, what has to be done to succeed if at all,
what was the expert's reactions/opinion/suggestions, what the expert
liked/disliked strongly, what are the conditions to make the system
practically useful, what were the reasons for success and what were
the reasons for failures.
- For challenge problems, lessons learned from individual
contributions and from the organization of such a challenge can be
reported as well as the characteristics of successful approaches that
made them appealing to experts evaluating the results.
Contributions not desired for this workshop would report on marginal
improvement over existing methods using artificial synthetic data or
UCI data involving no expert evaluation.
Workshop format
The workshop will consist of presentations of submitted papers, and a
panel discussion.
If there is sufficient interest and the workshop contributions are of
sufficient quality, we may publish a journal special issue, edit a
book, or do both.
Paper submission
We invite submission of long and short papers. Long papers should no
longer than eight pages and short papers should be no longer than four
pages, in the ICML-2002 format. Papers will be published
electronically (and most probably also as ICML workshop notes). Please
use the same instructions as for ICML-2002 papers, available at
http://www.cse.unsw.edu.au/~icml2002/format.html.
Electronic submissions should be either in PDF (preferred) or
gzip-compressed PostScript. In addition, please send the title page in
plain ASCII format.
Submissions should be emailed both to Nada Lavrac (Nada.Lavrac@ijs.si)
and to Tom Fawcett (tom_fawcett@hp.com)
Important Dates
22 April Paper submission deadline
10 May Notification to participants
31 May Camera ready copies
5 June Working notes due
9 July Workshop held
Organizers
Nada Lavrac
J. Stefan Institute
Jamova 39
1000 Ljubljana
Slovenia
E-mail: nada.lavrac@ijs.si
URL: http://www-ai.ijs.si/NadaLavrac/
Hiroshi Motoda
Division of Intelligent Systems Science,
The Institute of Scientific and Industrial Research,
Osaka University
8-1 Mihogaoka, Ibaraki, Osaka 567-0047
Japan
E-mail: motoda@sanken.osaka-u.ac.jp
URL: http://www.ar.sanken.osaka-u.ac.jp/motodaeg.html
Tom Fawcett
MS 1143
Hewlett-Packard Laboratories
1501 Page Mill Rd.
Palo Alto, CA 94304
USA
E-mail: tom_fawcett@hp.com
URL: http://www.hpl.hp.com/personal/Tom_Fawcett/
Program committee
Marko Bohanec marko.bohanec@ijs.si
Andrea Danyluk andrea@cs.williams.edu
Charles Elkan elkan@cs.ucsd.edu
Dragan Gamberger dragan.gamberger@irb.hr
Christophe Giraud-Carrier cgc@elca.ch
Ross King rdk@aber.ac.uk
Huan Liu hliu@asu.edu
Dunja Mladenic dunja.mladenic@ijs.si
Foster Provost fprovost@stern.nyu.edu
Patricia Jean Riddle pat@cs.auckland.ac.nz
Maarten van Someren maarten@swi.psy.uva.nl
Ashwin Srinivasan Ashwin.Srinivasan@comlab.ox.ac.uk
Einoshin Suzuki suzuki@dnj.ynu.ac.jp
Shusaku Tsumoto tsumoto@computer.org
This archive was generated by hypermail 2b29 : Sun Mar 17 2002 - 15:48:22 PST