Abstract
Humans can effortlessly identify objects in natural scenes, and they can shift attention rapidly between objects in a scene. We recently showed showed that much of neocortex is selective for object categories (Huth et al., SFN 2010). How are these categorical representations modulated when attention is directed toward specific categories? Neurophysiological studies in area V4 have shown that feature-based attention causes shifts in tuning toward the attended feature (Mazer et al., 2003; David et al., 2008). Therefore, here we sought to determine whether category-based attention shifts object category tuning toward the attended category. Human subjects viewed natural movies for 60 minutes while fixating steadily. Blood-oxygen level-dependent (BOLD) responses were recorded continuously using whole-brain fMRI. Two category-based attention tasks alternated in ten minute blocks: search for "humans" and search for "vehicles". Subjects depressed a button whenever an exemplar of the target category appeared in the movie. Categorical tuning was estimated for each voxel using the modeling approach developed in our earlier studies. First, the objects in each one-second movie clip were labeled using 900 categories drawn from WordNet. Then regularized linear regression was used to estimate categorical tuning for each voxel under each condition of category-based attention. We find that category-based attention causes many voxels to shift their tuning toward the target object. These shifts cannot be described by simple additive (baseline) or multiplicative (gain) changes at the level of single voxels. Large tuning shifts were detected in most of occipito-temporal cortex, and in fronto-parietal brain areas commonly assumed to be a control network of attention. These data demonstrate that category-based attention shifts tuning dynamically to more closely match the attended object category. This pattern suggests that the visual system implements a matched filter scheme to optimize processing of behaviorally-relevant objects during natural vision.
Meeting abstract presented at VSS 2012