Representational Geometries of Perception and Working Memory

This study compared the representational geometry of visual perception and visual working memory using human 7T fMRI. Observers viewed a face-scene blended sample, attended to either the face or scene, and judged whether the attended aspect matched a subsequent test image. We found that the stimulus coding dimension distinguishing faces and scenes was rotated when perceptual content entered working memory, in both visual and association areas. In these regions, exclusive combinations (i.e., perceived face and memorized scene vs. perceived scene and memorized face) were linearly decodable, indicating that perception and working memory occupy distinct subspaces. Such high-dimensional, flexible coding may help prevent confusion between perception and memory. In contrast, robust cross-decoding across perception and working memory revealed factorized representational formats, consistent with generalizability of their contents. The balance between flexibility and abstraction varied along the cortical hierarchy: early sensory regions emphasized flexibility, whereas transmodal regions showed greater abstraction.