Abstract

A central goal in cognitive neuroscience is to map mental processes onto their physical substrates. Traditionally, this goal is pursued by associating different mental processes with spatially-based anatomic localization. On the contrary, much less is known about the temporal dynamics of neural activity, particularly at the scale of the whole brain. Research into whole-brain dynamics has been challenging primarily due to the lack of an analytic framework that can summarize such dynamics, expose the underlying mechanisms, and associate the mechanisms with cognitive processes. In this dissertation, I describe my doctorate research aimed at building such a framework and deploying it to study cognitive individual differences. First, I modeled the resting state brain dynamics recorded by functional magnetic resonance imaging (fMRI) from hundreds of individuals in the Human Connectome Project (HCP), using the Mesoscale Individualized NeuroDynamics (MINDy) framework. Analysis revealed that the resting brain was not in a monolithic state, but instead exhibited multiple stable states (i.e., attractors) with distinct functional network activation patterns, suggesting that large-scale brain dynamics indeed possess a rich temporal structure. I then extended the MINDy framework to simultaneously estimate the influence of cognitive task-related variables on the neural dynamics, entitled MINDy-X (MINDy with eXogenous inputs). I applied MINDy-X to model and compare resting state and N-back working memory task state brain dynamics in the HCP fMRI dataset. Both resting and task state dynamics can be explained by a single model with a task-related modulatory input, suggesting a unifying mechanism across the two cognitive states. Interestingly, various topological and geometrical features of the modeled dynamics were predictive of individual differences in task accuracy and reaction time, indicating the utility of studying cognitive function through the lens of large-scale brain dynamics. Finally, to unearth the mechanism underlying high-dimensional nonlinear dynamics (such as those captured by MINDy) and to compare such mechanisms across individuals and cognitive states, I developed a mathematical-computational method called DFORM (Diffeomorphic vector field alignment FOR learned Models). In DFORM, an artificial neural network is used to learn a nonlinear coordinate transformation that maximally aligns the trajectories of the transformed system to those of a target system, enabling geometry-agnostic comparison between two dynamics. Additionally, DFORM can be used to locate important low-dimensional invariant sets within a high-dimensional model, including unstable limit sets that are very difficult to find using simulation-based methods. Overall, through this work, I have established a vocabulary to describe large-scale brain dynamics across various cognitive states and demonstrated its value in studying cognitive individual differences. I also developed an accessible tool to analyze and compare such dynamics despite high-dimensionality and nonlinearity. My work overcomes the methodological limitation in studying large-scale brain dynamics and provides many new opportunities for cognitive neuroscience research.

Committee Chair

ShiNung Ching

Committee Members

Todd Braver, Geoffrey Goodhill; Janine Bijsterbosch; Jeffrey Zacks

Degree

Doctor of Philosophy (PhD)

Author's Department

Biology & Biomedical Sciences (Neurosciences)

Author's School

Graduate School of Arts and Sciences

Document Type

Dissertation

Date of Award

6-12-2026

Language

English (en)

Download

Included in

Neurosciences Commons

Share

COinS