Connectionists: Neurocomputing Special Issue: Advances in Learning with Label Noise

[Benoît Frénay]

[Apologies if you receive multiple copies of this CFP]

*Call for Papers: Neurocomputing Special Issue on "Advances in Learning 
with Label Noise**"

* *_AIMS AND SCOPE_*

Label noise is an important issue in classification. It is both 
expensive and difficult to obtain completely reliable labels, yet 
traditional classifiers expect a perfectly labelled training set. In 
real-world data sets, however, the labels available often contain 
mistakes.  Mislabelling may occur for several reasons, including lack of 
information, speedy labelling by non-experts, the subjective nature of 
class memberships, expert errors, and communication problems. 
Furthermore, label noise may take several different forms -- for 
instance, labelling errors may occur at random, or may depend on 
particular values of the data features, or they may be adversarial. 
Errors may affect all data classes equally or asymmetrically. A large 
body of literature exists on the effects of label noise, which shows 
that mislabelling may detrimentally affect the classification 
performance, the complexity of the learned models, and it may impair 
pre-processing tasks such as feature selection.

Many methods have been proposed to deal with label noise. Filter 
approaches aim at identifying and removing any mislabelled instances.  
Label noise sensitive algorithms aim at dealing with label noise during 
learning, by modelling the process of label corruption as part of 
modelling the data. Some methods have been modified to take label noise 
into account in an embedded fashion. The current literature on learning 
with label noise is a lively mixture of theoretical and experimental 
studies which clearly demonstrate both the complexity and the importance 
of the problem. Dealing with mislabelled instances needs to be flexible 
enough to accommodate label uncertainty, yet constrained enough to guide 
the learning process in its decisions regarding when to trust the label 
and when to trust the classifier.

This special issue aims to stimulate new research in the area of 
learning with label noise by providing a forum for authors to report on 
new advances and findings in this problem area.  Topics of interest 
include, but are not limited to:

  * new methods to deal with label noise;
  * new applications where label noise must be taken into account;
  * theoretical results about learning in the presence of label noise;
  * experimental results which provide insight about existing methods;
  * dealing with different types of label noise (random, non-random,
    malicious, or adversarial);
  * conditions for the consistency of classification in the presence of
    label noise;
  * label noise in high dimensional small sample settings;
  * the issue of model meta-parameters/order selection in the presence
    of label noise;
  * feature selection and dimensionality reduction in the presence of
    label noise;
  * label-noise aware classification algorithms in static and dynamic
    scenarios;
  * on-line learning with label noise
  * learning with side information to counter label noise;
  * model assessment in the presence of label noise in test data.


_*SUBMISSION OF MANUSCRIPTS*__*
*_
If you intend to contribute to this special issue, please send a title 
and abstract of your contribution to the guest editors.

Authors should prepare their manuscript according to the Guide for 
Authors available at http://www.journals.elsevier.com/neurocomputing. 
All the papers will be peer-reviewed following the Neurocomputing 
reviewing procedures.  Authors must submit their papers electronically 
by using online manuscript submission at http://ees.elsevier.com/neucom. 
To ensure that all manuscripts are correctly included into the special 
issue, it is important that authors select "SI: Learning with label 
noise" when they reach the "Article Type" step in the submission process.

For technical questions regarding the submission website, please contact 
the support office at Elsevier or the guest editors.

_*IMPORTANT DATES*_

Deadline of paper submission: 15 February 2014
Notification of acceptance: 15 July 2014

_*GUEST EDITORS*_

Benoît Frénay (Managing Guest Editor)
Université catholique de Louvain, Belgium
E-mail: benoit.frenay at uclouvain.be <mailto:benoit.frenay at uclouvain.be>
Website:http://bfrenay.wordpress.com <http://bfrenay.wordpress.com/>
Phone: +32 10 478133 <tel:%2B32%2010%20478133>

Ata Kaban (Special Issue Guest Editor)
University of Birmingham, United Kingdom
E-mail: A.Kaban at cs.bham.ac.uk <mailto:A.Kaban at cs.bham.ac.uk>
Website:http://www.cs.bham.ac.uk/~axk <http://www.cs.bham.ac.uk/%7Eaxk>
Phone: +44 121 41 42792 <tel:%2B44%20121%2041%2042792>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20130918/966232a4/attachment.html>