Attention and correlated variability

Understanding how attention reshapes activity across neural populations is central to explaining its effects on perception and behavior. Using fMRI during a visual task with a spatial attention manipulation, we measured BOLD responses to a centrally presented colored dot field while human participants attended either to the stimulus or to the fixation cross as a baseline control. Replicating classical findings, we demonstrated that attention increased overall BOLD activity, with progressively stronger modulation along the visual hierarchy. Importantly, attention also reduced correlated variability (noise correlations) between voxel pairs in visual cortex, a hallmark of attentional modulation previously established in monkey electrophysiology. This shared variability was predominantly low‑dimensional, and attention selectively attenuated its dominant eigenmode, which can reconcile how attention differentially modulates mean activity and noise correlation across cortical areas. Despite these robust changes in population activity, feature decoding in visual cortex remained stable, indicating that reductions in shared variability do not necessarily change encoded feature information. Together, these findings demonstrate that attention reshapes population activity in the human brain in ways that parallel electrophysiological findings in nonhuman animals, highlighting correlated variability as a conserved signature of attentional state.