| |
Abstract:
Human subjects tend to perform tasks faster and more
accurately with practice. Cognitive-level explanations of these
behavioral changes often involve some form of threshold
modification to node activity or changes in connection strengths
such that representations become active faster, at higher levels,
and/or with greater precision. However, the changes in neural
activity associated with stimulus repetition and improved
performance are most often decreases rather than increases - a
phenomenon known as "repetition suppression" (Desimone, 1996). We
propose that the behavioral improvement following stimulus
repetition involves greater neural synchronization and more
efficient neural processing that arises from a reduction of
activity. We show that artificial networks of spiking neurons with
"synaptic depression" (an automatic reduction in synaptic efficacy
following pre-synaptic activity), can account for many of the
empirical findings associated with repetition suppression in humans
and monkeys. Synaptic depression leads to reductions in both the
mean and variance of neural firing rates which dynamically enhances
neural synchronization. As neurons synchronize, processing
efficiency increases because fewer spikes are required to fire
post-synaptic neurons. This can improve the rate of information
transmission, allowing earlier propagation of individual spikes
throughout an entire processing pathway.
|
