Visual statistical learning is associated with changes in low-dimensional cortical architecture

Abstract

Our ability to automatically learn environmental regularities over time to make predictions about the world is central to human cognition and behaviour. Despite its importance, the precise roles of cortical and subcortical regions in this statistical learning process remain debated, with limited understanding of how these different brain systems reorganize their activity during learning. To address this, we analyzed human functional MRI brain activity during a visual statistical learning task, whereby individuals implicitly learned to associate pairs of images embedded within a larger sequence. By projecting individuals’ patterns of cortical and subcortical functional connectivity onto a low-dimensional manifold space, we found that statistical learning was associated with changes along a single neural dimension describing connectivity across the visual-parietal and perirhinal cortex. During learning, we found that regions within the visual cortex expanded along this dimension, reflecting their increased segregation from other networks, whereas regions within the dorsal attention network contracted, reflecting their integration with higher order transmodal cortex. Notably, when learning was interrupted, we found that the perirhinal cortex and entorhinal cortex, which did not initially show learning-related effects, now contracted along this same neural dimension, reflecting their increased integration with the default mode and dorsal attention network, and decreased covariance with visual cortex. While prior work has associated statistical learning with either broader cortical, or medial temporal lobe activity changes, our findings suggest a more integrative view, whereby cortical regions reorganize during the formation of learned associations and the medial temporal lobe responds to their violation.