Abstract
In a recent line of psychophysical experiments, we found that working memory for a surface feature (color) interacts dynamically with saccadic motor planning, even if subjects are instructed to make saccades based only on spatial cues. A match between the remembered color and the color of either the designated target or a distractor influences saccade target selection, metrics of averaging saccades, and saccade latency in a systematic fashion. We give a theoretical account for these effects using the framework of dynamic neural fields, in which neural processes are modeled through the evolution of continuous activity distributions over metric feature spaces. In an architecture that is consistent with visual processing pathways in the primate cortex, we use separate multi-layer representations for spatial and surface feature information, which are both coupled bidirectionally to a combined perceptual representation of visual input. Peaks of activity in the top layer of the spatial representation indicate the metrics of saccadic motor plans. In the feature representation, the contents of working memory are represented by activity peaks that are self-sustained by means of lateral interactions. Although these memory peaks do not evoke any overt activity in the earlier perceptual representations by themselves, they influence the evolution of activity in response to a visual stimulus. They can thereby exert a biasing effect on the formation of a motor plan. With this model, we simulated the complete experimental time course, including formation of working memory from a visual cue, planning and execution of saccades under different stimulus conditions, and subsequent test of the memory performance. We were able to replicate the key experimental observations regarding saccade target selection, metrics, and latency. Our work shows how neural processes supporting perception, memory, and motor planning can interact even if they are localized in distinct representational structures.