Proceedings of the Annual Meeting of the Cognitive Science Society

Dynamic functional connectivity (dFC) matrices capture time-varying interactions among large-scale brain networks, but their high spatiotemporal dimensionality complicates the interpretation of how whole-brain functional organization is configured and reconfigured across cognitive tasks. In this work, we introduce a geometric framework that embeds sliding-window dFC from resting-state and task fMRI into a shared low-dimensional manifold. Within this space, resting-state dFC embeddings cluster within a compact intrinsic resting-state region of the manifold that serves as a functional baseline, whereas task engagement yields smooth, looped trajectories that depart from and return to this baseline. Distinct tasks trace separable loop structures that remain stable across parcellation resolutions. Phase-dependent sampling of dFC matrices along these loops reveals structured, time-ordered reconfigurations of functional networks during task performance. Together, these findings depict brain dynamics as continuous, recurrent trajectories rather than discrete state transitions, providing a unified geometric representation of task-evoked dynamic functional reconfiguration.