Working memory capacity of spiking neural network model

Abstract

Working memory (WM) is a system for shortterm storage and manipulation of information. Neural circuits of the prefrontal cortex (PFC) of the brain are assumed to be responsible for WM implementation. WM capacity is limited by just a few elements, and memorization accuracy decay when a set of items to be memorized is too large. We studied a computational model of working memory formation based on spiking neural network. The model imitates working memory formation within synaptic theory: memorized items are stored in form of short-term potentiated connections in selective population but not in form of persistent activity. Working memory capacity was studied in dependence of time scales of synaptic facilitation and depression and background excitation of the network. Estimated WM capacity is shown to be possibly larger than classical experimental estimations of four items. But capacity strongly depends on intrinsic parameters of neural networks.