Network geometry shapes multi-task representational transformations across human cortex
The human brain's ability to perform a wide variety of possible tasks raises fundamental questions about how distributed neural circuits transform and integrate diverse task-related information. Using fMRI data from participants performing 16 diverse tasks and at rest, we investigated how intrinsic connectivity patterns shape task representations across the human cortex. We found that cross-region connectivity dimensionality strongly predicts representational transformations between brain regions: low-dimensional connectivity corresponds to representational compression, while high-dimensional connectivity corresponds to expansion. Modeling activity flow over these connections, we determined that task-evoked activity flows over intrinsic connections generate diverse representational geometries along the cortical hierarchy. Critically, in nearly all brain regions, connectivity-based transformations produced activity patterns that more closely matched their targets than their source representations, demonstrating that connectivity patterns actively transform neural representations. Additionally, we found that regions with lower dimensional connectivity show stronger cross-task similarities, indicative of shared latent task features that generalize neural processing across tasks. In contrast, we found that regions with higher-dimensional connectivity exhibit strong conjunctive coding of task variables, enabling representations for task-specific scenarios. These findings demonstrate how the network geometry of the brain's intrinsic connectivity architecture systematically shapes the transformation of task representations across the cortex to support performance of diverse tasks.