Abstract
Past work has shown that observers can exert voluntary control over which locations are prioritized during scene processing (spatial attention) as well as which items from a scene are encoded into working memory (working memory gating). One fundamental question that remains, however, is whether these two phenomena reflect the operation of a single or multiple forms of attentional control. Recent work attempting to separate these forms of attention has been limited either due to a confound between the number of relevant objects and spatial locations, or by the sensitivity of the measures used. To combat the the object-location confound, we designed a novel change detection task using “dot clouds” which can change in size and overlap in space. To improve sensitivity, we designed the task to enable simultaneous decoding of spatial attention and WM load using modern machine learning approaches. We fit inverted encoding models to alpha power to decode both the location and precision of spatial attention, and fit a logistic regression model to raw EEG signals to decode load. We found that spatial attention during the delay period spread with the size of dot clouds, but did not vary with the number of dot clouds when they took up the same amount of area. In contrast, a load decoding model trained on these matched area conditions could successfully separate set sizes regardless of spatial extent in held-out conditions. Together, these results suggest the existence of 2 distinct forms of attentional control: directing spatial attention, and object-based encoding into WM.