The role of neural networks in the study of the posterior parietal cortex

The use of a neural network model of a cerebral cortical area as an aid to understanding this area's function is reviewed. The basic model is a feedforward multilayer network that learns to transform the coordinates of a visual stimulus from a retinocentric to a craniocentric reference frame using backpropagation. An extension of the model to one that transforms retinal coordinates into body-centered ones predicts response properties that are confirmed by neurophysiological experiments. The simulation of electrical stimulation of the model predicts a pattern of effects similar to the one obtained by stimulation of a specific region of the parietal cortex. The study of the response properties of the model's units provides a simple explanation of how the parietal cortex might compute coordinate transformations and of why certain manipulations such as stimulation should produce the effects observed. >