This is the authors' abstract. We don't add key points for this paper.
How the human brain performs such a wide variety of possible tasks remains an enigma. Recent work revealed that reduced dimensionality of task representations enables generalization across contexts, yet the mechanisms implementing such compression remain unclear. We hypothesized that network architecture constrains representational compression via low-dimensional region-to-region connectivity mappings. Notably, this contrasts with the common assumption that between-region connections passively carry information. We tested this using fMRI data from participants performing 16 diverse tasks and activity flow modeling, a data-driven neural network modeling approach. As hypothesized, connectivity dimensionality predicted cross-region representational changes: low-dimensional connectivity compressed task representations, while high-dimensional connectivity expanded them. Activity flow modeling demonstrated these connectivity patterns transform rather than passively transfer representations, producing compression-then-expansion from sensory through association to motor regions, enabling cross-task generalization and task-specific implementation. These findings reveal how intrinsic network geometry specifies representational transformations, providing systems-level insights into how network geometry enables flexible cognition.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Chakravarthula et al. (2025) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: