Connectionists: CFP: ESANN'2008 special sessions

[esann@dice.ucl.ac.be]

                     ESANN'2008

16th European Symposium on Artificial Neural Networks
  Advances in Computational Intelligence and Learning 

      Bruges (Belgium) - April 23-24-25, 2008

                   Special sessions

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


The following message contains a summary of all special sessions that will
be organized during the ESANN'2008 conference.  Authors are invited to
submit their contributions to one of these sessions or to a regular session,
according to the guidelines found on the web pages of the conference
http://www.dice.ucl.ac.be/esann/.  Deadline for submissions: November 23,
2007.

According to our policy to limit the number of unsolicited e-mails, we
gathered all special session descriptions in a single message, and try to
avoid sending it to overlapping distribution lists.  We apologize if you
receive multiple copies of this e-mail despite our precautions.  


Special sessions that will be organized during the ESANN'2008 conference
=========================================================

1.  Computational Intelligence in Computer Games
     Colin Fyfe (University of Paisley, United Kingdom)

2.  Methodology and standards for data analysis with machine learning tools.
     Damien François (Université catholique de Louvain, Belgium)

3.  Neural Networks for Computational Neuroscience
     David Meunier, Hélène Paugam-Moisy (LIRIS-CNRS, France)

4.  Machine learning methods in cancer research
     Alfredo Vellido (Polytechnic University of Catalonia, Spain), 
     Paulo J.G.Lisboa (Liverpool John Moores University, United Kingdom)

5.  Machine Learning Approches and Pattern Recognition for Spectral Data
     Thomas Villmann (Univ. Leipzig, Germany), 
     Erzsébet Merényi (Rice University, USA), 
     Udo Seiffert (IPK Gatersleben, Germany)


Short descriptions
==================

1.  Computational Intelligence in Computer Games
-----------------------------------------------------------------------
Organized by:
Colin Fyfe (University of Paisley, United Kingdom)

This Special Session of ESANN2008 invites papers on any application of
computational intelligence to computer games. The computational intelligence
techniques may be artificial neural networks, evolutionary algorithms,
artificial immune systems, swarm intelligence or machine learning
techniques. The computer games may be any current kind of games (First
Person Shooter, Real Time Strategy, driving, simulator, board, puzzle,
classic, arcade games...) running on any platform (PC, Mac, Java, Flash,
XBOX 360, Playstation 3, Wii...) or computer simulations of classical
mathematical game theory problems. In all cases, the paper should
demonstrate that the technique used has provided a degree of intelligence in
the computer game.


2. Methodology and standards for data analysis with machine learning tools.
-----------------------------------------------------------------------
Organized by:
Damien François (Université catholique de Louvain, Belgium)

Compared to well established fields, such as linear identification, data
mining might still be considered as an art. There is no standard procedure
leading from data to models; each practitioner develops its own methodology,
depending on the tools he uses and the data he faces. While literature
describing the tools is vast, papers explaining how to use them efficiently
are rare. 

The general methodology for data analysis requires making choices as
experimentation time is often limited; the choice of the feature selection
method, the choice of the model family (e.g. ANN, trees, SVM), the choice of
the algorithm for choosing the model structure (LOO, CV, bootstrap), the
choice of the training algorithm and its parameters (e.g. 3 stage learning
vs orthogonal least squares for RBFN), the choice of the splitting of the
data in training, validation and test sets, are some examples. Hundreds of
methods have been available since decades, but still nobody knows which one
to choose. 

This session is dedicated to methodology-oriented papers describing best
practices, or proposing guidelines, that would help data miners in making
relevant choices thorough the whole process of data mining. Papers should
not propose a new method but rather propose a methodology to choose, or
combine existing models/algorithms/methods that are appropriate to solve a
given data analysis problem. 

A non-exhaustive list of questions that could be addressed is the following:


- What are the adequate early steps in data exploration for modelling? 
- Which models/implementations should I have in my data analysis toolbox? 
- How do I choose a feature selection and/or model selection method? 
- How do I choose proper preprocessing? How do I treat missing data ? How do
I handle non standard data ? 
- What should be done when classes are unbalanced? 
- How to deal with heterogeneous features? How do I standardize/normalise
them? 
- When are trees more appropriate than support vector machines or artificial
neural networks? 
- Which datasets should I use to benchmark my new algorithm/model? How do I
split the data set? 
- Which standards for data and models storage/interchange should I follow? 

Submitted papers will be reviewed according to the ESANN reviewing process
and will be evaluated on their scientific value; originality, correctness,
and writing style.


3. Neural Networks for Computational Neuroscience
-----------------------------------------------------------------------
Organized by:
David Meunier, Hélène Paugam-Moisy (LIRIS-CNRS, France)

Computational Neuroscience aims at explaining the experimental measurements
obtained in electrophysiology (both in animals with intra-cranial recordings
and in humans with techniques such as EEG, MEG) by means of models. The
models can take several forms, one of them being neural networks. The aim of
this session is to contribute to this specific use of neural networks. 

The way to develop models to explain electrophysiology can have two
directions. 

A first one, qualified of "bottom-up" approach, where the modelling consists
in trying to reproduce properties observed at macroscopic level by fitting
neuron models (e.g. Hodgkin and Huxley neuron, spiking neuron,
integrate-and-fire neuron) and parameters to the measured behaviour of
biological neurons. The architecture of the network is also based on
experimental anatomical data (e.g. olfactory bulb architecture, hippocampus
architecture, etc...). This direction includes mean-field approaches, where
the influences of each parameter on the dynamics are studied, and
qualitative approaches, aiming at defining the minimal set of properties
that are necessary to observe a given dynamical behaviour. 

A second one is qualified of "top-down" approach, consisting in using
evolutionary algorithms to let emerge adapted neural networks with regard to
a given task and to study their properties a posteriori. In this case, the
network is studied a posteriori to detect why adapted networks are better
than initial random networks. The network emergent properties can be
studied, by example, with the tools of the theory of complex systems.


4. Machine learning methods in cancer research
-----------------------------------------------------------------------
Organized by:
Alfredo Vellido (Polytechnic University of Catalonia, Spain), 
Paulo J.G.Lisboa (Liverpool John Moores University, United Kingdom)

Neural Networks and Machine Learning methods in general are widely used in
cancer research and published in clinical, as well as methodological
journals. Their acceptance among medical practitioners is steadily
increasing, in part because of demands for advanced data analysis relating
to bioinformatics, but also because of a realization that decision support
will be inherent in the current agenda for personalized medicine. The
application of Machine Learning to medical data may be said to have entered
a period of adolescence, where the early excitement about their potential
has been tempered by the need to assure generality through the use of
principled approaches to complexity control. The excitement that was
communicated during the early phase of development in the late 90’s seems to
have whetted the appetite of clinicians for what these methods can achieve,
initiating close and fruitful collaborations where key clinical questions
are driving new data-based studies, so building clinical relevance, rather
than obsolescence, into study design. 

Machine Learning methods can be applied to a wide range of data types and
problems in cancer research. The range of applications includes exploratory
analysis and predictive inference, with topics ranging from clustering,
through classification, survival analysis, and rule extraction. Hot topics
include knowledge discovery from data, but also the integration of
multimodal data into clinical inference systems, the use of graphical models
for structure finding in large sparse data sets, and methods for robust
performance estimation which include the use of automatic rule extraction
methods to match inference making with clinical expert knowledge. This
special session aims to bring together methodological advances and clinical
relevant case studies of Machine Learning approaches to cancer diagnosis and
prognosis, and oncology-related bioinformatics. ESANN 2008 participants
would benefit from the coming together of a number of internationally
renowned experts in the field, who would provide their expert view on a
broad palette of state-of-the-art theoretical developments and applications.



5. Machine Learning Approches and Pattern Recognition for Spectral Data
-----------------------------------------------------------------------
Organized by:
Thomas Villmann (Univ. Leipzig, Germany), 
Erzsébet Merényi (Rice University, USA), 
Udo Seiffert (IPK Gatersleben, Germany) 

Analysis of spectral data plays an important role in many areas of research
like physics, astronomy and geophysics, chemistry, bioinformatics,
biochemistry engineering, and others. The amount of data may range from
several billion samples in geophysics to only a few in medical applications.
Further, a vectorial representation of spectra typically leads to
huge-dimensional problems. However, spectral vectors are functional, i.e.,
the vector dimensions are not independent. The locations, widths and shapes
of characteristic peaks or valleys (absorptions), as well as their
co-occurences are important for data analyses. These properties should be
used for specific machine learning approaches designed for spectral
analysis. 

This special session seeks contributions which report about new developments
in this field of research: both outstanding applications using specific
techniques and methodologies in machine learning and neural networks for
spectral data, as well as new theoretical developments are solicited. 

The session is intended to cover a broad range of application areas as
outlined in the beginning. A possible (non-exhaustive) list of
applications/problems could be: 
- NMR-based applications in physics, chemistry and biology 
- Remote sensing in astronomy and geophysics 
- Chemometrics 
- Bioinformatics 
- Medical applications 
- Special techniques for utilization of data-intrinsic dependencies 
- High-dimensional data and sparseness 
- Special metrics for data similarities


========================================================
ESANN - European Symposium on Artificial Neural Networks
http://www.dice.ucl.ac.be/esann

* For submissions of papers, reviews,...
Michel Verleysen
Univ. Cath. de Louvain - Machine Learning Group
3, pl. du Levant - B-1348 Louvain-la-Neuve - Belgium
tel: +32 10 47 25 51 - fax: + 32 10 47 25 98 
mailto:esann at dice.ucl.ac.be

* Conference secretariat
d-side conference services
24 av. L. Mommaerts - B-1140 Evere - Belgium
tel: + 32 2 730 06 11 - fax: + 32 2 730 06 00 
mailto:esann at dice.ucl.ac.be
========================================================