Connectionists: ECML Workshop : Principles and Practice of Multiple Learning Systems

[Gavin Brown]

Apologies for multiple postings....



=============================================================
  Principles and Practice of Multiple Learning Systems
  --- a one-day workshop as part of ECML, Sept 21st 2007

    http://www.cs.man.ac.uk/~gbrown/ppmls/
=============================================================

Multiple Learning Systems, aka multiple classifier systems,
ensembles, information fusion, have emerged as one of the
strongest pattern recognition techniques of the last decade,
influencing every area of Machine Learning and Data Mining.
These techniques, such as Boosting or Bagging, that create
multiple classifiers then combine them by voting or averaging
are becoming a staple part of the ML toolbox.

On a deeper level, the "combining" principle has been extended
in several new directions.  Cluster ensembles, semi-supervised
learning, changing environments, kernel combining methods
are all utilizing this principle. This workshop aims to
highlight these advanced uses of the "combining" principle,
right across the spectrum of machine learning and data mining.

This is the first ECML workshop to address this important
principle. PPMLS aims to cover both the deep theoretical
questions and practical applications of the idea.

=============================================================
Invited Speaker and Programme Committee
=============================================================

Invited Speaker:
  We are honoured to have Dr Ludmila Kuncheva as invited speaker.
  Dr Kuncheva has been a leading influence in the Multiple
  Classifier Systems community, and is the author of the main
  textbook in the field, "Combining Pattern Classifiers".

Organisers:
  Gavin Brown (University of Manchester, UK)
  Elzbieta Pekalska (University of Manchester, UK)

Review Committee:
  Robert Duin (Delft, Netherlands)
  Giorgio Fumera (Cagliari, Italy)
  Tin Kam Ho (Bell Labs, USA)
  Tim Kovacs (Bristol, UK)
  Natalio Krasnogor (Nottingham, UK)
  Ludmila Kuncheva (Bangor, Wales)
  Nikunj Oza (NASA Ames, USA)
  Fabio Roli (Cagliari, Italy)
  Amanda Sharkey (Sheffield, UK)
  Marina Skurichina (Delft, Netherlands)
  Terry Windeatt (Surrey, UK)
  Jeremy Wyatt (Birmingham, UK)


=============================================================
Suggested topics for papers are:
=============================================================

  Learning schemes
     * Classifier selection vs classifier combination
     * Boosting and Bagging methods
     * Feature subspaces / feature projections
     * Learning classifier systems

  Paradigm combination
     * Combining Objective Functions
     * Combining Discriminative and Generative models
     * Combining Kernels
     * Combining Statistical and Structural techniques
     * Combining proximity measures

  Theory
     * Methods of classifier combination
     * Diversity measures
     * Classifier selection vs classifier combination
     * Weak and strong models vs overtraining
     * Handling noise and outliers

  Data flexibility
     * MLS in changing/dynamic environments
     * Cluster ensembles
     * Stream mining
     * Learning from distributed data
     * Learning from multiple data sets
     * Semi-supervised learning

  Applications
     * Image classification/retrieval
     * Data mining
     * Text classification
     * Computer vision
     * Biometric identification

We particularly encourage researchers to share their
both positive *and* negative experiences with the
combining paradigm.


=============================================================
Important dates
=============================================================

  Submission deadline:     June 30th, 2007
  Acceptance notification: July 21st, 2007
  Camera-ready deadline:   July 28th, 2007
  Workshop:                September 21, 2007


=============================================================
Submitting a paper
=============================================================

The workshop proceedings will be formally published with the ECML
proceedings, and in addition distributed on a CD to all ECML 
participants. Selected extended versions may be considered
for a journal special issue.

Papers should be sent by E-MAIL (in PDF format) to:

   ppmls at cs.man.ac.uk

with a limit of 10 pages, in the Springer LNCS style.
Style files and further details can be downloaded from
the workshop website:

   http://www.cs.man.ac.uk/~gbrown/ppmls/

=============================================================