Network geometry shapes multi-task representational transformations across human cortex

Abstract

Flexible cognition requires the brain to generalize across tasks while preserving task-specific information. Previous work showed that task representations are compressed in association cortex and expanded in sensory and motor systems, but the mechanisms underlying these transformations are unclear. Here we test whether the geometry of intrinsic brain connectivity constrains how representations change between cortical regions. We analyzed fMRI acquired during 16 diverse cognitive tasks together with activity flow modelling, finding that connectivity dimensionality predicts representational changes across cortex. Low-dimensional connectivity compressed task representations and increased similarities across tasks, whereas high-dimensional connectivity expanded representations and supported conjunctive coding of task features. Activity flow modelling reproduced the observed compression to expansion pattern along the sensory-association-motor hierarchy and generated representational geometries that more closely matched targets than sources, consistent with transformation rather than transfer of information. These findings identify network geometry as a systems-level principle that shapes cortical representations and supports flexible cognition.