Saccade Target Selection in Naturalistic Visual Search

Rapid and accurate saccadic eye movements of the high-resolution fovea are crucial for performing virtually any visuo-motor task. Indeed, we typically make more than 180,000 saccades during the waking hours of each day, and each one of these movements requires the selection of a saccade target. Considering the vast number of possible saccade targets typically present in natural scenes, saccade target selection is a computationally intensive process, yet it is one that we accomplish rapidly, without conscious awareness, and with great precision. This study focuses on the role of the midbrain superior colliculus (SC) in saccade target selection, as well as its interactions with the cortical frontal eye fields (FEF). When searching for an object, a number of factors influence saccade target selection, including the physical salience of potential targets, their similarity to the target being sought, and whether they have already been fixated and rejected as the target. Most studies have used isolated, point-like stimuli to study target selection, despite the fact that the real world is composed of spatially extended objects, and that objects are the typically relevant perceptual units for directed behavior. This project is one of the few that will use extended object-like stimuli to better understand saccade target selection under more naturalistic conditions. A key advance is that we will use a cutting-edge information theoretic analysis technique known as time-resolved partial information decomposition (PID). PID is particularly suited for research questions aimed at a fine-grained understanding of how neurons dynamically encode a small number of variables over time. This will afford us unprecedented insights into how the SC combines different sources of information to select a saccade target, and how the selectivity of SC neurons changes as new information becomes available. PID will also enable us to rigorously test various models of how the SC integrates salience, target similarity, and inhibitory tagging information during search. Overall, the project will move the field forward significantly both with its cutting-edge analysis techniques and in its focus on target selection in more naturalistic scenes involving spatially-extended objects.