A mathematical model of working memory system in higher order visual processing

Abstract

The purpose of this study is to understand the function of working memories while seeking a target during visual search tasks by using a neuron network model. A previous study indicated that the model which can reproduce gaze shifts while searching a target must contain a mechanism with which allocating attention to the visual attributes dynamically shifts (Sugimoto & Kohama, 2011) and maintaining inhibition to a past gazed position for several steps. However, the target information must set as a parameter before initiation of searching tasks, and also implementation of the inhibition of return in this model was not based on the physiological evidence. In this study, we proposed a modified version of Sugimoto & Kohama's model with considering the working memories for the visual attributes derived by PF and IT network. Furthermore, we added a network which assumed TPJ between V4 and LIP to memorize the locus of past gazed regions for the use of the inhibition of return. The visual search simulation results suggest that the target detection rate of our model, without giving any target information, is significantly higher than that of the previous model.