Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!oitnews.harvard.edu!purdue!lerc.nasa.gov!magnus.acs.ohio-state.edu!math.ohio-state.edu!howland.reston.ans.net!EU.net!peer-news.britain.eu.net!bcc.ac.uk!pc6-033.ion.bpmf.ac.uk!W.ElDeredy
From: W.ElDeredy@ucl.ac.uk (Wael El-Deredy)
Subject: Feature Selection Workshop (Final CFP))
Sender: news@ucl.ac.uk (Usenet News System)
Message-ID: <W.ElDeredy.57.000AE7F3@ucl.ac.uk>
Date: Fri, 12 Jan 1996 10:54:22 GMT
Organization: Institue of Neurology
X-Newsreader: Trumpet for Windows [Version 1.0 Rev A]
Lines: 129


                         Final Call for Papers/Participation
                               One-day Workshop
Intelligent Feature Selection: Statistical and Neural Approaches
                     University of Sussex, Brighton, UK
                                 1 April, 1996
                     (Part of AISB'96 workshop series)
                 http://www.cogs.susx.ac.uk/aisb/aisb96

Selecting a subset of representative features is an important part
in the design of classification and prediction systems and often a
key factor in determining their performance. This Workshop is
concerned with the selection of a particular subset of features or
variables from a potentially larger set of initial measurements.
This is a long-standing problem which has seen significant
contributions from statisticians and practitioners in traditional
AI. More recently, these methods of variable selection are being
extended to non-linear neural networks. Relevant applications
cover a wide range from data poor static problems to data rich
problems with time dynamics. One example is spectroscopy, where
the dimension of pattern vectors is very large compared the
limited number of observations. Here, there is an interest in
reducing the number of variables using the data, without making
prior assumptions about the chemistry of the process either
explicitly through an a priori choice of explanatory variables, or
implicitly by taking combinations of the input variables. In time-
series analysis, there is also a need to identify a core of
relevant variables, which may be irregularly spaced in time. In
these, as in many other areas, there is a demand for fast,
consistent and efficient algorithms from feature selection
particularly when the number of variables is large and the number
of observation is limited.

The Workshop will discuss the theoretical foundations and
practical considerations for the design of algorithms that choose
a smaller number features or input variables which best represent
a process, including: ways of quantifying what is meant by 'best',
how this goal can be achieved in different application areas, the
implications of the selected subset on the meaning and the
interpretation of the predictor and the consequences on its future performance.

Workshop aims:
To bring together theoreticians and practitioners interested in feature 
selection.
To bridge across statistical and neural network methods.
To review criteria for selection and discuss guidelines for best practice.

Focal Points
* Tools for feature selection from statistics and neural networks
* Selection criteria and consistency of selection
* Implications on model transparency and model performance
* Assessment of fitness and testing the hypothesis of 'best' feature subset
* Practical considerations and case studies

**Plenty of time will be dedicated in both session for open discussions

Target Participants:
Theoreticians: Statistics, signal processing and neural networks
specialists. Practitioners: spectral estimation, time series
analysis, medical diagnosis, financial modelling, consumer
modelling, astronomy, etc.

Invited speakers:
David Mackay, University of Cambridge: Bayesian methods of feature
selection for neural networks.
and
Phil J. Brown, University of Kent: Statistical methods for
variable selection and calibration.

Call for papers
Contributions are invited both from theoreticians willing to
present and discuss rigorous generic approaches to feature or
variable selection to a wide audience and practitioners describing
relevant case studies and applications. Although submissions are
strongly encouraged, those wishing to participate without papers
are asked to submit a summary of interests.
Four (4) copies of the proposed contribution MUST arrive by Friday
16 February, 1996. Preferred size is two A4 pages. Faxed or E-
mailed submissions will be allowed provided that originals are
also sent by post and are received before the notification date
(below) in order to be included in the workshop proceedings. Send
contributions to Wael El-Deredy (address below). Submissions will
be reviewed by at least two referees and notification of
acceptance will be sent by 26 February, 1996. Publication other
than in the Workshop Proceedings will depend on the number and
quality of submitted contributions.

Workshop principal organiser:
Dr. Paulo Lisboa				
Dept. of Electrical Engineering and
Electronics
University of Liverpool
lisboa@liv.ac.uk	

Program enquiries and submissions:
Wael El-Deredy
Institute of Neurology
London WC1N 3BG
tel. 0171 837 3611 x4169
fax 0171 278 7894
W.Elderedy@ion.bpmf.ac.uk

Registration and administrative enquiries
Alison White
AISB96 local orginsation chair
University of Sussex
Brighton BN1 9QH, U.K.
Tel: +44 1273 678448
Fax: +44 1273 671320
alisonw@cogs.susx.ac.uk

Important Dates:
Submission Deadline				16 February, 1996
Notification of acceptance (Faxed or E-mailed) 		26  February, 1996
Early registration deadline 				1 March, 1996
Workshop					1 April, 1996

Workshop Fees (in pounds Sterling)
				AISB members	 Non members	Student
Before 1 March			65		        80		45
After   1 March			85	                      100		60


Wael El-Deredy                               W.ElDeredy@ucl.ac.uk
Dept. of Neurological Surgery                Tel +44 (0)171 837 3611 x 4169
Institute of Neurology - Queen Square        Fax +44 (0)171 278 5069
London WC1N 3BG, UK


