TR: Dynamics of recurrent networks with and without hidden layer

[Yves Moreau]

The following technical report is available via ftp or the World Wide
Web:


WHEN DO DYNAMICAL NEURAL NETWORKS WITH AND WITHOUT 
HIDDEN LAYER HAVE IDENTICAL BEHAVIOR?

Yves Moreau and Joos Vandewalle
K.U.Leuven ESAT-SISTA


K.U.Leuven, Elektrotechniek-ESAT, Technical report ESAT-SISTA TR97-51

To get it from the World Wide Web, point your browser to:
ftp://ftp.esat.kuleuven.ac.be/pub/SISTA/moreau/reports/rnn_equivalence_tr97-51.ps

To get it via FTP: 
ftp ftp.esat.kuleuven.ac.be
cd pub/SISTA/moreau/reports
get rnn_equivalence_tr97-51.ps


ABSTRACT
========

We present conditions for the input-output equivalence between
dynamical neural networks with a hidden layer and dynamical neural
networks without hidden layer. We achieve this result by proving a
general result on transdimensional changes of coordinates for
dynamical neural networks with a hidden layer. This result is of
interest because dynamical neural networks without hidden layer are
more amenable to analytical studies.

Keywords: dynamical neural network, input-output equivalence,
transdimensional change of coordinates


--------------------

Yves Moreau

Department of Electrical Engineering
Katholieke Universiteit Leuven
Leuven, Belgium

email: moreau at esat.kuleuven.ac.be

homepage: http://www.esat.kuleuven.ac.be/~moreau

publications:
http://www.esat.kuleuven.ac.be/~moreau/publication_list.html