The rapidly increasing volume of data collected for decision support applications in commercial, industrial, medical, and defense domains has made it a challenge to scale up knowledge discovery in databases (KDD), the machine learning and knowledge acquisition component of these applications. Many techniques currently applied to KDD admit enhancement through the wrapper approach, which uses empirical performance of inductive learning algorithms as feedback to optimize parameters of the learning system.

Wrappers include algorithms for performance tuning, especially: optimization of learning system parameters (hyperparameters) such as learning rates and model priors; control of solution size; and change of problem representation (or inductive bias optimization). Strategies for changing the representation of a machine learning problem include decomposition of learning tasks into more tractable subproblems; feature construction, or synthesis of more salient or useful input variables; and feature subset selection, also known as variable elimination (a form of relevance determination).

This workshop will explore current issues concerning wrapper technologies for KDD applications, including:

• problems to which wrappers can be successfully applied
• dimensionality reduction
• feature subset selection with irrelevant variables
• feature extraction
• other change of representation for KDD
• transforms
• ensemble learning (mixture estimation)
• empirical performance tuning and experimentation
• meta-learning
• learning methodologies that admit or support wrappers
• search-based wrappers (selection, partitioning)
• probabilistic wrappers
• metric-based model selection
• statistical validation methods
• evolutionary computation approaches
• genetic algorithms for feature selection and extraction
• feature construction by genetic programming
• "parameterless" GAs for performance tuning
• detection and reduction of GP code bloat
• linkage learning
• multi-objective optimization
• foundational theories
• statistical learning theory
• genetic and evolutionary computation
• intron detection and elimination
• theory of linkage learning
• finding the GA "sweet spot"
• the role of prior knowledge and interactive knowledge elicitation in developing wrappers for KDD

Workshop Audience

This workshop is intended for researchers in the area of machine learning, including practitioners of knowledge discovery in databases (KDD) and statistical and computational learning theorists. Intelligent systems researchers with an interest in high-performance computation and large-scale, real-world applications of data mining (e.g., inference and decision support) will also find this workshop of interest.

Workshop Agenda

(Last updated 05 Dec 2000)

The workshop will consist of a short tutorial, a preliminary discussion period, morning and afternoon paper sessions (each 1.5 hours) followed by open discussion sessions, and two invited talks by KDD practitioners. To foster student participation, we will hold a discussion session on student research after the second paper session. If there is interest, this will be extended to a poster session.

An afternoon parallel track following the invited talk(s) will be between two components of the program committee - the organizing committee and review committee, with the latter group consisting of researchers who will meet with students cf. the IJCAI and AAAI Graduate Student Consortia.

To allow time for informal discussions, we plan to limit most talks to 15-30 minutes, and hold a brainstorming session on special issues of journals and follow-up workshops and symposia at other conferences. To encourage participation but focus discussions on key topics, we invite 2-page research synopses and position papers from attendants who do not submit full papers.

Morning Session:

• 08:30 - 09:00: intro
• 09:00 - 09:30: wrapper tutorial / commentary, William Hsu
• 09:30 - 10:30: invited talk, Foster Provost
• 10:30 - 10:45: coffee break
• 11:00 - 11:30: paper presentation - Nick Street
• 11:30 - 12:00: paper presentation - Edwin Pednault
• 12:00 - 13:00: lunch

[Lunch Break]

Afternoon Session:

• 13:00 - 14:00: invited talk, Charles Elkan
• 14:00 - 14:30: paper presentation - Jesse Reichler
• 14:30 - 15:00: paper presentation - William Hsu
• 15:00 - 15:30: paper presentation - Cecil Schmidt
• 15:30 - 15:45: break
• 15:45 - 16:30: open discussion
• 16:30 - 17:00: planning for future events, special issues
• 17:00 (approx.): ADJOURN

Important Dates

(Last updated 22 Mar 2001)

Full Papers due:	Monday, 02 April 2001 (extended deadline)
Short Papers due:	Friday, 06 April 2001
acceptance notification:	Monday, 09 April 2001
camera-ready copy due:	Friday, 20 April 2001 (extended deadline)
workshop	Saturday, 04 August 2001

Call for Papers and Submission Procedure

(Last updated 09 Mar 2001)

We encourage submissions containing original theoretical and applied concepts in KDD. Experimental results are also encouraged, especially on fielded applications, even if they are only preliminary. We therefore invite two categories of paper submissions:

• research papers
• should not exceed 8 pages including title page
• due 02 April 2001
• short summaries (including position papers)
• should not exceed 2 pages
• due 06 April 2001

We request that authors prepare papers in the standard IJCAI format.

The first page of submitted papers should include the title; a brief abstract; and author names, affiliations, postal addresses, electronic mail addresses, and telephone and fax numbers.

Papers should be submitted in PostScript, PDF, or Microsoft Word 97/2000 file format. Electronic submissions are preferred, but both types of papers may be submitted by one of the following three options:

• E-mail a URL pointer to the paper to workshop-ijcai2001@www.kddresearch.org.
• Attach a PostScript, PDF, or Word file (or .zip, .gz, or .bz2 archive) to workshop-ijcai2001@www.kddresearch.org. Please use the first option (i.e., send a URL) if the file is larger than 5Mb.
• Mail 5 hard copies printed on 8.5" x 11" or A4 paper with at least 1 inch margins on all sides.

If you do not receive e-mail confirmation of the submission, please submit it again and send e-mail to the workshop chair (bhsu@cis.ksu.edu).

Organizing Committee

Dr. William H. Hsu (primary contact)
Department of Computing and Information Sciences
p: Kansas State University, 234 Nichols Hall, Manhattan, KS 66506-2302 * t: (785) 539-7180 * f: (785) 539-7180 * e: bhsu@cis.ksu.edu * w: www.cis.ksu.edu/~bhsu

Dr. Hillol Kargupta
Department of Electrical Engineering and Computer Science
p: University of Maryland Baltimore County, 1000 Hilltop Circle, Baltimore, MD 21250 * t: (410) 455-3972 * f: (410) 455-3969 * e: hillol@cs.umbc.edu * w: www.cs.umbc.edu/~hillol

Dr. Huan Liu
Department of Computer Science and Engineering
p: Arizona State University, Box 875406, Tempe, AZ 85287-5406 * t: (480) 727-7349 * f: (480) 965-2751 * e: hliu@asu.edu * w: www.public.asu.edu/~huanliu

Dr. Nick Street
Department of Management Sciences
p: The University of Iowa, S210 Pappajohn Business Building, Iowa City, IA 52242-1000 * t: (319) 335-1016 * f: (319) 335-0297 * e: nick-street@uiowa.edu * w: dollar.biz.uiowa.edu/~street

Page created: 05 Dec 2000 Last updated: 27 Oct 2002 William H. Hsu