Working memory and temporal patterns: The implications of neural populations.

摘要

Working memory, the ability to temporarily retain information which will be used to guide subsequent behavior, is a central component of our cognitive abilities. Almost 40 years ago, electrophysiological experiments in monkeys established that persistent activity may be the neuronal substrate of working memory. Many computational models have been proposed in order to explain persistent activity, with recurrent connections playing a prominent role in many of these. No model, however, has captured all the important features in working memory networks. This work presents three related models which seek to understand some of these features. In particular, the first model explores the formation of firing rate patterns during the delay period of working memory tasks; the second model explores the dynamics of working memory networks with reduced inhibition, and their possible role in epilepsy; the third model explores the formation of temporal sequences and closed loops of activity. The three models assume the existence of densely connected neural populations (such as minicolumns), which respond similarly to the same stimuli. Another common feature is the role of dynamic synapses: the first two models rely on synaptic facilitation, whereas the third uses temporally asymmetric Hebbian plasticity.