Interplay of noise and recurrence enhances spatial categorization in neural networks

Abstract

We investigate the interplay of recurrence and noise in neural networks trained to categorize spatial patterns of neural activity. We develop a procedure to demonstrate how in the presence of noise, the introduction of recurrence permits to significantly extend and homogenize the operating range of a feed-forward neural network. We also show that the performance of the reconnected network has features reminiscent of non-dynamic stochastic resonance: the addition of noise enables the network to correctly categorize stimuli of subthreshold strength, with optimal noise magnitude significantly exceeding the stimulus strength. We characterize the dynamics leading to this effect, and contrast it to the behavior of a more simple associative memory network in which noise-mediated categorization fails.