IJPAM: Volume 39, No. 4 (2007)

DYNAMIC PROPERTIES OF RECURRENT NEURAL
NETWORKS AND ITS APPLICATIONS

Leong-Kwan Li$^{1}$, S. Shao$^{2}$
$^{1}$Department of Applied Mathematics
The Hong Kong Polytechnic University
Hung Ham, Kowloon, HONG KONG, P.R. CHINA
e-mail: [email protected]
$^{2}$Department of Mathematics
Cleveland State University
2121 Euclid Avenue, RT 1527, Cleveland, OH 44115, USA
e-mail: [email protected]


Abstract.We study the dynamics of the leaky integrator recurrent neural network. Our results show that there exists at least one equilibrium point of the system, and the set of solutions of the leaky integrator recurrent neural dynamics is positive invariant and attractive. The globally exponential stability property of the system has been discussed. Our examples show that the leaky integrator recurrent neural network together with the state space search algorithm can be an effectively tool for many applications including data compression and learning the short-term foreign exchange rates.

Received: August 1, 2007

AMS Subject Classification: 37C70, 37C75, 68T05, 68T20

Key Words and Phrases: leaky integrator, recurrent neural networks, positive invariant, attractive set, global exponentially stability

Source: International Journal of Pure and Applied Mathematics
ISSN: 1311-8080
Year: 2007
Volume: 39
Issue: 4