Modeling the functional development of human visual motion area MT+

Abstract

Area MT+ is a patch of middle temporal cortex that plays a critical role in our ability to perceive motion in the visual modality. Recent neuroimaging studies of congenitally blind adults suggest that this brain area can “learn” to represent auditory motion, but only when individuals are deprived of visual input from birth. Here I present a parallel distributed processing network that behaves similarly to area MT+. Its internal connection weights are such that it is able to compute the direction of motion by comparing the locations of two sequentially-presented visual inputs. Trained on visual + auditory input, it continues to respond only to visual motion. In the absence of visual inputs, it learns to detect motion in auditory inputs. My network is characterized by innate processing biases, coupled with a capacity for flexibility. I argue that this implementation is a plausible model of the neural network that constitutes area MT+.