Conference on Cognitive Computational Neuroscience

Saccadic target-selection is often described by winner-takes-all (WTA) dynamics. However, traditional models focus on only a handful of competing targets or on a single target. How the brain generates sequences of eye movements through repeated selection among many targets remains underexplored. To tackle this question, we massively scale up a biophysical dynamic mean-field WTA model consisting of a full cortical sheet of about 1000 competing populations with overlapping receptive fields. This decouples neuronal populations from circumscribed target locations and allows for target selection in response to naturalistic saliency distributions. We model visual and movement neuron populations in the frontal eye-fields. Movement neurons receive local salience as input, relayed by visual neurons, and compete to choose the next target. A corollary discharge feedback signal aids the selection process by enhancing the input to the visual neurons that feeds into the winning movement neuron population. A noteworthy feature of our model is that this corollary discharge leads to the emergence of receptive-field shifts of visual neurons, occurring just before saccade onset. To ensure sequential selection, a suppressive feedback signal - inhibition-of-return - relayed by visual neurons to the movement neuron population that won the competition. Together, these mechanisms produce realistic spatiotemporal saccadic eye movements.