Institute for Mathematics and Its Applications
Xiao-Jing Wang, Brandeis University
Working memory is a memory function by which humans and animals hold ``on-line" and manipulate information for a short time span of a few seconds. In this talk, I shall first review the experimental data about the neuronal correlates of working memory, namely persistent activity and dynamical tuning in the prefrontal cortex (PFC). Then a network model is presented for the PFC neural network behaviors during a visuospatial delayed-response task. It is assumed that the spatial cue stimulus is encoded in working memory storage by a continuum of network activity profiles, which coexist with a spontaneous rest state. Such a bistability can be realized in the model by a neocortical network architecture with recurrent excitation and inhibition. We contrast the bistable PFC-like network to the V1-like network for orientation tuning. Moreover, because of the spatial translational invariance of the network activity profiles, the robustness of memory storage against noise and distractors raises serious problems. Possible cellular/synaptic mechanisms for reliable visuospatial working memory will be discussed.