CFP - Neurocomputing: Geometrical Methods in Neural Networks and Learning

[Koning, Esther (ELS)]

CALL FOR PAPERS
 

NEUROCOMPUTING - An International Journal

Editor-in-Chief: Tom Heskes


Journal published by Elsevier Science B.V. - URL:
http://www.elsevier.com/locate/neucom


Special Issue on "Geometrical Methods in Neural Networks and Learning"
 

Understanding the underlying geometric structure of a network's parameter
space is extremely important to designing systems that can effectively
navigate the space while learning. Although modern mathematics is needed in
the research of neural networks, and there are some very powerful results
and techniques in these geometric methods, these are currently scattered in
various sources.

Over the last decade or so, driven greatly by the work on information
geometry, we are seeing the merging of the fields of statistics and geometry
applied to neural network and learning. This requires intense collaboration
and communication.

The interest displayed by the scientific community into these research
topics is also testified by several activities such as the special issue on
"Non-Gradient Learning Techniques" of the International Journal of Neural
Systems (guest editors A. de Carvalho and S.C. Kremer), the Post-NIPS*2000
workshop on "Geometric and Quantum Methods in Learning", organized by S.-i.
Amari, A. Assadi and T. Poggio (Colorado, December 2000), the workshop
"Uncertainty in geometric computations" held in Sheffield, England, in July
2001, organized by J. Winkler and M. Niranjan (University of Sheffield, UK),
the special session of the IJCNN'02 on "Differential & Computational
Geometry in Neural Networks" (session chair: E. Bayro-Corrochano, CINVESTAV,
Guadalajara, Mexico) held in Honolulu, Hawaii (USA), in May 2002, and the
workshop  "Information Geometry and its Applications", held in Pescara
(Italy), in July 2002, organized by P. Giblisco.

For these reasons, the Neurocomputing journal dedicates a Special Issue to
the theory and advanced applications of geometric concepts to neural
learning and optimization, bringing together contributions well founded in
modern mathematics.

The topics of the Special Issue are the theoretical and practical aspects of
geometrical methods for the design of neural networks making emphasis in
geometric learning and  optimization. The readers will have for the first
time a collection of approaches including differential geometrical methods
for learning, the Lie group learning algorithms, the natural (Riemannian)
gradient techniques, learning by weight flows on Stiefel-Grassman manifolds,
the theories for learning on orthogonal group, neurocomputing using Clifford
geometric algebra, the numerical aspects of the solution of the
matrix-equations on Lie groups arising in neural learning/optimization and
related topics.

The Neurocomputing journal invites original contributions for the
forthcoming Special Issue on Geometrical Methods in Neural Networks and
Learning from a broad scope of areas. Some topics relevant to this special
issue include, but are not restricted to:

- Neural principal component/subspace analysis;

-  Neural independent component analysis and blind source separation;

- Natural computing (geometrical algorithms that could take place in neural
circuitry);

-  Selection of subspaces for optimal neural data compression;

-  Neural optimization over the orthogonal group and optimization problems
in tensor algebra;

-  Information geometry;

-  Provably convergent geometric algorithms for real-time learning;

-  Geometry of boosting methods;

- Geometric Clifford algebra for the generalization of neural networks;

-  Geometrical methods of unsupervised learning for blind signal processing;

- Application of Lie operators and use of differential geometry based
learning techniques;

-  Conformal and horosphere models for neurocomputing;

-  Tensorial approach for geometrical neural computation and learning;

-  General graphical model and belief propagation for machine learning;

-  Geometry of statistical-physical methods for learning.
 

Please submit the electronic copy to
http://authors.elsevier.com/journal/neucom including abstract, keywords, a
cover page containing the title and Author(s) name(s), corresponding
Author's complete address including fax and EMail address, and clear
indication to be a submission to the Special Issue on Geometrical Methods in
Neural Networks and Learning.
 

-- Scheduling of the Special Issue:

Deadline for papers submission: March 1, 2004
End of refereeing process and result issuing: July 30, 2004
Submission of the final manuscript: September 30, 2004
 

Guest Editors:
 

Simone Fiori
Faculty of Engineering, University of Perugia
Polo Didattico e Scientifico del Ternano
Loc. Pentima bassa, 21, I-05100 Terni (Italy)
Fax: +39.0744.492925
EMail: fiori at unipg.it
URL: http://www.unipg.it/sfr/


Shun-ichi Amari
RIKEN Brain Science Institute
Laboratory for Mathematical Neuroscience
Wako-shi Hirosawa 2-1, Saitama 351-0198 (Japan)
Fax: +81.48467.9687
EMail: amari at brain.riken.go.jp

URL: http://www.bsis.brain.riken.go.jp/