Stochastic dynamics of neural networks after learning

Abstract

We study probabilistic synchronous dynamics of Little-Hopfield neural networks with asymmetric interneuronal synaptic connections adjusted in accordance with a learning rule given in [3]. Types of behaviour of such systems are analysed in dependence of a vector-parameter characterizing degrees of freedom in determining synaptic couplings. We have found that in the case of small level of noise networks can store memorized patterns but as an amount of noise is increased behaviour becomes more complex and, beginning with some critical value the motion seems to be chaotic.