A hierarchical model of visual processing simulates neural mechanisms underlying reflexive attention

The visual system deploys attention reflexively in response to important stimuli to facilitate rapid selection. To better understand how a system that is distributed across a hierarchy of brain areas coordinates this rapid response, we have developed a computational model that simulates reflexive attention as a recurrent lock-on state between topographically aligned cortical areas. This model provokes key questions about reflexive attention that we study through the N2pc, an EEG component that indexes the timing and lateralization of stimulus-driven attention. A key finding from such research is that for two sequential targets sharing a location, the second elicits no N2pc, despite being easily perceivable. This suggests that attention locks-on to a target's location and carries forward in time to a second target without being redeployed. Here, we assess key properties of reflexive attention in four EEG experiments. First, the lock-on N2pc effect generalizes to feature target types. Second, the lock-on state is specific to a location, not just a visual hemifield. Third, reflexive attention decays as the interval between two targets increases, such that a second target will once-again elicit an N2pc at longer intervals. Finally, the lock-on state is not specific to a particular target type, which implies that reflexive attention is mediated using representations that are not stimulus-specific. These results provide important constraints on our understanding of visual attention. We incorporate those constraints in a formalized model of attention that elucidates the link between neural mechanisms and a key neural correlate of attention.