Stability of neural networks and convergence of their sensitivity computation algorithms

✍ Scribed by Kiichi Urahama Member

• Book ID: 104591338
• Publisher: John Wiley and Sons
• Year: 1990
• Tongue: English
• Weight: 500 KB
• Volume: 21
• Category: Article
• ISSN: 0882-1666
• DOI: 10.1002/scj.4690210808

✦ Synopsis

Abstract

This paper considers the continuous‐time model and the discrete‐time model for the neural network, and discusses the stability of the model and the convergence of the sensitivity computation algorithm. First, a sufficient condition is given for the global asymptotic stability of the model. Then the speed with which the solution approaches the equilibrium is evaluated. Based on the results obtained, the local asymptotic stability of the equilibrium solution of the multi‐stable circuit is discussed. The fault‐tolerant property of the model is discussed briefly. Finally, it is shown that under the same condition as the sufficient condition for the asymptotic stability of the model, a certain kind of general sensitivity computation converges in the global sense.