TITLE:   Training Asymptotically Stable Recurrent
         Neural Networks

AUTHORS: N.J. Dimopoulos, S. Neville, J. Dorocicz, C. Jubien

IN:      1995 IEEE Conference on Systems, Man, and Cybernetics,
         Vancouver B.C., Canada

ABSTRACT

In this work we present a class of recurrent neural networks 
which are asymptotically stable.  For these networks, we 
discuss their similarity with certain structures in the  
central nervous system, and prove that if an interconnection 
pattern that does not allow excitatory feedback is used,
then the resulting recurrent neural network is stable.  
We introduce a training methodology for networks belonging 
to this class, and use it to train networks that successfully 
identify a number of nonlinear systems.