Attentional allocation locally warps representational space

Abstract

Attentional allocation is hypothesized to transiently and selectively warp representational space. In the current study, participants viewed brief video clips of five types of animals each performing four actions. In each run, participants performed a 1-back task requiring them to attend to either the animal type or the action performed. Surface-based searchlight SVM classification revealed distinct areas coding for animal category and action category. Action classification was greatest in lateral occipital, superior parietal and postcentral regions, while animal classification was greatest in early visual and ventral temporal cortices. Classification accuracy increased with attention in higher-level cortical areas thought to code for category-level animal and action information, while accuracy in early visual areas decreased with attention. A representational similarity multiple regression analysis implemented with surface-based searchlights revealed that target similarity structures are differentially predictive of the neural similarity structure as a function of attentional task. These results suggest that attention warps distributed representational spaces such that task-relevant representations are more discriminable. Furthermore, cortical areas corresponding to early and late visual processing are differentially impacted by this attentional warping effect.

Meeting abstract presented at VSS 2014