Abstract
Efficient visual search involves exploring novel regions of a scene and ignoring previously analysed ones. Computational models of search incorporate this through an activity map in which regions of space that have been previously foveated are tagged with an inhibitory signal. We trained a rhesus monkey to search for green targets among red distractors in a grid-like array while we recorded activity from the superior colliculus (SC), a midbrain structure causally involved in saccade target selection. In a dynamic search condition, each target changed color to the distractor color upon being foveated, such that the monkey had to ‘extinguish’ all the targets by fixating each one. In contrast, in the memory search condition, targets did not change color upon fixation. This required the monkey to remember the locations that had already been foveated in order to search efficiently. Overall, search was more efficient in the dynamic than the memory search condition, as measured by the number of saccades per trial and the proportion of saccades to targets. SC visual activity for a given stimulus was reduced during a 100–150 ms period following fixation onset when the stimulus had been previously foveated, even after controlling for the target-distractor identity of the RF stimulus. At the individual cell level, visual cells showed more suppression than visuomotor and motor cells. We also observed a sinusoidal modulation of the cells’ mean activity in the suppression period based on the time since the distractor was last foveated. These findings point to a role for the SC in mediating inhibitory tagging during visual search, similar to what has been observed in area LIP.
Acknowledgement: NIH EY014885