ANALYSIS OF THE EFFECTS OF NOISE ON A MODEL FOR THE NEURAL MECHANISM OF SHORT-TERM ACTIVE MEMORY

Zipser (1991) showed that the hidden unit activity of-fully recurrent neural network model, trained on a simple memory task, matched the tem poral activity patterns of memory-associated neurons in monkeys perfor ming delayed saccade or delayed match-to-sample tasks. When noise, sim ulating random fluctuations in neural firing rate, is added to the uni t activations of this model, the effect on the memory dynamics is to s low the rate of information loss. In this paper, we show that the dyna mics of the iterated sigmoid function, with gain and bias parameters, is qualitatively very similar to the tonic response properties of Zips er's multiunit model. Analysis of the simpler system provides an expla nation for the effect of noise that is missing from the description of the multiunit model.