Stability and adaptation in artificial neural systems

1 September 1989

journal article
research article
Published by American Physical Society (APS) in Physical Review A

Vol. 40 (5) , 2681-2688
https://doi.org/10.1103/physreva.40.2681

Abstract

A model of an artificial neural network for pattern association is discussed, which is dynamically stable and employs a weight matrix with asymmetric connections of a specific kind. The relationship to earlier stable models making use of symmetric connection matrices is established. A Lyapunov function for the system is explicitly specified. It is argued that its behavior as a function of time during the learning phase indicates the quality of the network’s performance regarding pattern association. This view is supported by a numerical example, which, in addition, serves to demonstrate the virtues of this model in comparison to others. The crucial role of the constants entering artificial neural systems is emphasized.

Keywords

This publication has 10 references indexed in Scilit:

Nonlinear neural networks: Principles, mechanisms, and architectures
Neural Networks, 1988
Bidirectional associative memories
IEEE Transactions on Systems, Man, and Cybernetics, 1988
Adaptive bidirectional associative memories
Applied Optics, 1987
Parallel Distributed Processing
Published by MIT Press ,1986
Neurons with graded response have collective computational properties like those of two-state neurons.
Proceedings of the National Academy of Sciences, 1984
Absolute stability of global pattern formation and parallel memory storage by competitive neural networks
IEEE Transactions on Systems, Man, and Cybernetics, 1983
Neural networks and physical systems with emergent collective computational abilities.
Proceedings of the National Academy of Sciences, 1982
Non-equilibrium processes in heavy-ion collisions at relativistic energies
Nuclear Physics A, 1979
Synergetics
Published by Springer Nature ,1978
Neural theory of association and concept-formation
Biological Cybernetics, 1977