Abstract
Preparatory attention is often studied using the cueing paradigm. According to the prevailing theory, following an attention-directing cue, top-down signals from the frontoparietal attention control regions propagate to visual cortex to bias sensory neurons to enable stimulus selection. Despite years of research, the underlying neural mechanisms remain to be better elucidated. We recorded fMRI data from participants performing a cued visual spatial attention task. At the beginning of each trial, participants were asked to covertly deploy attention to one of the two visual fields. Following a random cue-target period, a stimulus appeared either at the attended location or at the unattended location, and participants discriminated the stimulus appearing at the attended location and ignored the stimulus appearing at the unattended location. Applying MVPA to fMRI data, we reported the following findings: (1) attend-left vs attend-right can be decoded from the cue-evoked neural activity in all visual areas, (2) stimulus-left vs stimulus-right can be decoded from the target-evoked activity in all visual areas, (3) classifiers built on the cue-evoked data can decode stimulus-evoked activity in all visual areas and vice versa, and (4) the higher the cross-decoding accuracy, the better the behavioral performance. These results suggest that top-down control signals form neural patterns in the cue-target period that resemble the neural patterns evoked by the stimulus and these “attentional templates” enable stimulus selection and improve behavior.