A neuronal model for the shaping of feature selectivity in IT by visual categorization

Neurophysiology results have shown that learning a visual categorization task shapes the selectivity of inferiotemporal cortex neurons to task-relevant features of the stimuli. In this work, we propose a biologically realistic mean-field neuronal model of a two-layer network to explain these experimental results. We show that the enhancement of feature selectivity in one layer of the model can emerge due to input coming from another layer, corresponding to a region encoding stimulus category, possibly in prefrontal cortex. Further, we explore the behavior of the network in function of the weights of the connections between its two layers.