DOI: 10.1051/jp1:1995147
J. Phys. I France 5 (1995) 573-580

Dynamics of a Neural Network Composed by two Hopfield Subnetworks Interconnected Unidirectionally

L. Viana¹ and C. Martínez<sup>2

1</sup>  Laboratorio de Ensenada, Instituto de Física, UNAM, A. Postal 2681, 22800 Ensenada, B.C., México
²  Centro de Investigación Científica y de Educación Superior de Ensenada, Programa de Posgrado en Física de Materiales, A. Postal 2681, 22800 Ensenada, B.C., México

(Received 9 November 1994, accepted 30 January 1995)

Abstract
By means of numerical simulations, the dynamical behaviour of a Neural Network composed of two Hopfield Subnetworks interconnected unidirectionally and updated synchronically (at zero temperature T=0) is studied. These connections are such that each of the N neurons on the first subnet sends information to exclusively one of the N neurons on the second. A set of p patterns, composed by a subpattern for each of the subnets, is stored according to a Hebb-like rule. The recoverability of one particular subpattern by the second subnet is studied as a function of N and the load parameter in the case when the first subnetwork has already recovered its corresponding subpattern.

© Les Editions de Physique 1995