A neural network model of working memory exhibiting primacy and recency

The authors consider a simple dilute neural network in which the synaptic strengths are bounded, and the probabilities of strengthening and weakening the synapses during learning are different. During the sequential learning of patterns, introvert networks (i.e. those with synapses more easily weakened than strengthened) exhibit recency (i.e. the preferential retention of the latest learned patterns) as in the Hopfield-Parisi model. On the other hand, extrovert networks (i.e. those with synapses more easily strengthened than weakened) exhibit both recency and the novel primacy effect (i.e. the preferential retention of the earliest learned patterns). The occurrence of primacy depends on the initial distribution of the synaptic strengths. The relevance of the model to psychological experiments on working memory is also discussed.