A NEURAL MODEL FOR THE DETECTION OF TEMPORAL STRUCTURE

  • Dražen Domijan

Abstract

According  to  the  temporal  correlation  hypothesis,  synchronization  of  neural  activity  indifferent spatial maps solves the feature binding problem. Here, a new model of visual working memory is proposed which is able to group synchronous temporal events without neural synchronization. Instead, the model is based on the difference in the firing rate. The model integrates discrete inputs over time and compares activity in different integrators. When the amplitude difference in integrators is large enough due to the different rates of evidence accumulation, temporal figure and background are distinguished in the working memory. Computer simulations showed that the model correctly groups events according to their deterministic or stochastic temporal structure. The model is robust with respect to the temporal noise and to the correlation between figure-ground events. Also, the model is able to explain visual prior entry and perceptual asynchrony between colour, motion and orientation.
Section
Full Articles