Copy Link
Add to Bookmark
Report
Machine Learning List Vol. 2 No. 03
Machine Learning List: Vol. 2 No. 3
Saturday, Jan 20, 1990
Contents:
Over/under generalization in ID3
Machine learning and neural networks
ASIS '90 Workshop on Classification Research
Journal of Intelligent systems: Special Issue on Knowledge Acquisition
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 /usr2/spool/ftp/pub/ml-list/V<X>/<N> or N.Z
where X and N are the volume and number of the issue; ID & password: anonymous
- ----------------------------------------------------------------------
Subject: over/under generalization in ID3
Date: Thu, 11 Jan 90 09:56:14 PST
From: Don Cohen <donc@vaxa.isi.EDU>
I was surprised to see this question. If the problem is to
classify instances into two classes, and one occurs much more
often, I'd EXPECT that a reasonable classifier would more
often guess the more probably class when the evidence was
weak, and therefore one should expect to get more errors of
that type. This is the right thing to do, in terms of
minimizing the expected number of misclassifications. It
makes sense to "correct" this only if the goal is not just
to minimize the number of misclassifications (for instance,
if one type of error is more costly than the other).
- ----------------------------------------------------------------------
Subject: Machine learning and neural networks
Date: Sat, 20 Jan 90 18:25:29 -0800
From: Michael Pazzani <pazzani@ICS.UCI.EDU>
Nearly every week, I hear of a new neural net application in diverse
areas such as backgammon, predicting solar activity, classifying whale
songs as well as areas some bordering on the absurd (e.g., suggesting
to see if backpropagation can find a stealth bomber).
Several experiments have shown that backpropagation is not more
powerful and are more time consuming than ID3 and other "standard" ML
techniques. Are there also examples of successful machine learning
applications (soybean diseases and thyroid diseases are growing old)
or are the researchers using backpropagation and other neural network
techniques more clever in finding applications and engineering input
representations?
- -Mike
- ----------------------------------------------------------------------
Date: Wed, 17 Jan 90 11:26:02 EST
From: Susanne M Humphrey <humphrey@mcs.nlm.nih.GOV>
Subject: ASIS '90 Workshop on Classification Research
ASIS '90 Workshop on Classification Research
Organized by ASIS Special Interest Group on Classification Research (SIG/CR)
Call for Papers
The American Society for Information Science Special Interest Group on
Classification Research (ASIS SIG/CR) invites submissions for the ASIS '90
Classification Research (CR) Workshop, to be held at the 53d Annual Meeting of
ASIS in Toronto, Canada. The Workshop will take place Sunday, November 4,
1990, 9am - 5pm. ASIS '90 continues through Thursday, November 8.
The Workshop is designed to be an exchange of research ideas by participants,
addressing creation, development, management, representation, display,
comparison, compatibility, theory, and application of classification schemes.
Emphasis will be semantic classification, in contrast to statistically-based
schemes. However, a topic like statistical techniques used for developing
explicit semantic classes, which in turn might be applied to databases, would
be in scope. Topics include, but are not limited to:
- - Warrant for concepts in classification schemes.
- - Concept acquisition.
- - Basis for semantic classes.
- - Automated techniques to assist in creating classification schemes.
- - Knowledge representation systems.
- - Relations and their properties.
- - Inheritance and subsumption.
- - Classification algorithms.
- - Procedural knowledge in classification schemes.
- - Reasoning with classification schemes.
- - Software for managing classification schemes.
- - Interfaces for displaying classification schemes.
- - Data structures and programming languages for classification schemes.
- - Comparison and compatibility between classification schemes.
- - Applications such as subject analysis, natural language understanding,
information retrieval, expert systems
The CR Workshop welcomes submissions from various disciplines. Attendance
will be limited to authors of papers. Those interested in participating are
invited to submit short (2-3 single-space page) position papers, reflecting
substantive work that has been performed in the above areas or other areas
related to semantic classification schemes. Submissions may include
background papers as attachments. Position papers will be published in
proceedings to be distributed prior to the Workshop. Participants are
encouraged to distribute background papers at or prior to the Workshop. Lunch
will not be served; however, refreshments will be available during the day.
Workshop registration fee is $30.
Order of preference for mode of transmitting submissions: [1] Electronic mail
[2] Diskette accompanied by paper copy [3] Paper copy only (fax or postal).
Electronic submissions should be ASCII text; paper-only submissions should be
keyable as ASCII. Submissions should be sent to arrive by May 1, 1990, to:
Susanne Humphrey
Lister Hill National Center for Biomedical Communications
National Library of Medicine
Bldg 38A, Rm 9N903
Bethesda, MD 20894
Internet: humphrey@mcs.nlm.nih.gov
Fax: 301-496-0673
Phone: 301-496-9300
For additional information, contact the CR Workshop Co-Chairs, Susanne
Humphrey, as above, or Barbara Kwasnik, School of Information Studies, 4-206
Center for Science and Technology, Syracuse University, Syracuse, NY
13244-4100, e-mail bkwasnik@suvm.bitnet, fax 315-443-1954, telephone
351-443-4547 (direct office) or 351-443-2911 (department office).
- ----------------------------------------------------------------------
Subject: Journal of Intelligent systems: Special Issue on Knowledge Acquisition
From: westphal@ida.org (Christopher Westphal)
Date: 18 Jan 90 01:00:27 GMT
CALL FOR PAPERS
International Journal of Intelligent Systems
Special Issue on Knowledge Acquisition
The Journal of Intelligent Systems is planning a special issue
on knowledge acquisition - edited by Ken Ford and Jeff Bradshaw.
Preference will be given to formal/theoretical and cognitive
approaches to knowledge acquisition rather than expositions of
any particular implementation or application (unless clearly
supported by a theoretical foundation).
Authors must submit three copies of the complete manuscript to
Ken Ford (at the address below) by March 30, 1990. Manuscripts
should consist of between 20 & 40 double-spaced pages. All papers
will be subjected to a timely but thorough review.
For further information contact:
Kenneth M. Ford
Institute for Human & Machine Cognition
Division of Computer Science
The University of West Florida
11000 University Parkway
Pensacola, Florida 32514
kford@uwf.bitnet
(904) 474-2551
- ----------------------------------------------------------------------
END of ML-LIST 2.3
