Ordinary Differential Equations (ODEs) have long provided a robust mathematical framework for modeling the dynamic behavior of brain networks. These models are crucial for understanding the temporal evolution of neural activity, capturing how brain states change over time through the complex interplay of structural connectivity and dynamic functional processes. However, understanding the brain’s dynamic activity goes beyond ODEs alone. The evolving nature of brain networks—whether in healthy cognition, neuroplasticity, or pathological conditions—requires interdisciplinary approaches that integrate dynamic modeling with a variety of techniques to capture the richness of brain activity.
This workshop will bridge the gap between traditional ODE-based models and broader approaches to brain dynamics, including advanced neuroimaging methods, network theory, and data-driven models. We will explore how temporal fluctuations in brain activity—ranging from oscillatory behavior to chaotic dynamics—emerge and interact across different brain regions. Through this lens, we aim to foster discussions on how brain dynamics can be modeled in real-time, how perturbations like external stimuli or pathological changes propagate, and how multimodal data can be leveraged to track these changes.
Key topics of discussion will include:
Brain dynamics and the temporal evolution of neural activity through ODEs and beyond.
The integration of multimodal data (e.g., structural MRI, functional MRI, electrophysiological recordings) for modeling brain dynamics.
Causal inference in neuroimaging and brain networks, examining how dynamic interactions shape cognitive processes and dysfunctions.
Advances in data-driven methods, such as machine learning and generative AI, to model brain dynamics and their relationship to brain structure.
The role of dynamic systems in understanding neuroplasticity, cognition, and neurological disorders.
This workshop aims to bring together experts from both mathematical modeling and neuroscience research to discuss and share innovative methods for modeling the dynamics of neuroimaging and brain networks. Invited speakers from diverse fields, including neuroimaging, systems neuroscience, and computational modeling, will share their insights on the evolving landscape of brain dynamics. We will discuss how these approaches can be applied to a variety of topics, from basic cognition to the pathophysiology of neurological diseases.
Lightning Talks and Poster Session
This workshop will include lightning talks and a poster session for early career researchers (including graduate students). If accepted, you will be asked to do both. In order to propose a lightning session talk and a poster, you must first register for the workshop, and then submit a proposal using the form that will become available on this page after you register. The registration form should not be used to propose a lightning session talk or poster.
The deadline for proposing is September 13, 2026. If your proposal is accepted, you should plan to attend the event in-person.
In-Person Registration
Seats are limited at the venue, which means that in-person registration may be capped prior to the workshop start date. If capacity is reached, a waitlist will be imposed, which the registration form will reflect. Early registration is strongly encouraged.
All in-person registrants must wait to receive an invitation to attend in-person from IMSI before traveling, which generally begin to be sent out 4-6 weeks in advance.
All registrants (online and in-person) will receive zoom links and are welcome to attend online.
Registration Fee
A non-refundable registration fee will be payable by credit card or debit card for any participants invited to attend this workshop in-person. In-person participants agree to pay the non-refundable fee by the deadline given by IMSI. Failure to pay the fee by the deadline may mean that the invitation to attend in-person is revoked.
Guorong Wu
University of North Carolina at Chapel Hill
J
Y
Jingwen Yan
Indiana University Indianapolis
L
Z
Lu Zhang
Indiana University Indianapolis
W
Z
Wenhao Zhang
University of Texas – Southwestern Medical Center
Y
Z
Yu Zhang
Stanford University
Y
Z
Yize Zhao
Yale University
Schedule
Monday, November 9, 2026
8:30-9:00 CST
Registration & Light Refreshments
9:00-10:00 CST
From Connectomes to Foundation Models: Learning Clinically Relevant Brain Representations
Speaker: Yu Zhang (Stanford University)
Mental disorders present marked clinical heterogeneity and frequent comorbidity, challenging symptom-based diagnostic systems and hindering individualized treatment. Large-scale connectome data offer an opportunity to ground this heterogeneity in brain network organization. In this talk, I will introduce recent advances from our lab in contrastive machine learning, a general strategy for isolating clinically relevant connectome variation from shared background variability. Applied to EEG and fMRI connectomes, this framework uncovers symptom-linked connectivity dimensions in autism and transdiagnostic brain-behavior biotypes across ADHD and anxiety disorders that replicate in an independent cohort. Extended to clinical trials, contrasting antidepressant with placebo disentangles drug-specific from placebo-related early connectome changes, revealing that patients with drug-specific changes show greater symptom improvement. I will also present a brain graph foundation model pre-trained on large-scale brain graphs spanning diverse neurological and psychiatric disorders and multiple brain atlases. By combining graph transformer-based pretraining with meta-learning and graph/language prompting, this model enables few-shot and zero-shot adaptation across atlases, disorders, and tasks.
10:00-10:05 CST
Tech Break
10:05-11:05 CST
Connectome Imaging Techniques for Mapping Human Brain Pathways
Speaker: Yonggang Shi (University of Southern California (USC))
With the fast advance of connectome imaging techniques, we have the opportunity of mapping the human brain pathways in vivo at unprecedented resolution. In this talk, I will review the novel computational tools for connectome imaging developed in my group at USC Stevens Neuroimaging and Informatics Institute. Novel algorithms for fiber orientation distribution and compartment modeling from multi-shell diffusion MRI will be first introduced. After that, a suite of topography preserving tractography tools will be presented with their applications in the reconstruction of challenging human brain pathways. I will also discuss our recent work on surface-based fiber tracking algorithms for studying the connectivity of superficial white matter.
11:05-11:35 CST
Coffee Break
11:35-12:35 CST
Modeling Dynamic Brain Connectivity from EEG to Improve Brain–Computer Interfaces
Speaker: Jian Kang (University of Michigan)
Brain–computer interfaces (BCIs) translate brain activity into control signals for communication and assistive technologies. Most current BCI systems focus on well-known signals such as the P300 response, potentially overlooking important information contained in the full EEG signal. Leveraging the complete spatiotemporal structure of EEG data can improve both decoding performance and our understanding of brain function. We propose a Bayesian modeling framework that captures dynamic patterns of brain activity across electrodes and over time. The model summarizes high-dimensional EEG signals through a small number of latent components, enabling efficient analysis while preserving key neural patterns. In addition, it allows relationships between brain regions to evolve over time and differ across stimulus types, providing a flexible representation of functional connectivity. Applied to real BCI data, the proposed approach improves signal characterization by recovering known responses, identifying new patterns, and revealing dynamic changes in brain connectivity associated with user intent. These results demonstrate the potential of our method to enhance both BCI performance and neuroscientific insight.
12:35-13:35 CST
Lunch Break
13:35-14:35 CST
Alternative Diffusion Map for Brain Network Analysis
Speaker: Yuping Wang (Tulane University)
We propose a novel framework for multi-task fMRI based brain network analysis with alternating diffusion map (ADM). ADM is a manifold representation to extract geometry-preserving low-dimensional embeddings that can parameterize the intrinsic variables driving the phenomenon of interest. We first separately present resting-state and task-based functional connectivity networks from multi-task fMRIs, and then utilize the ADM to fuse the information to extract significant low-dimensionalembeddings. Then we use the embeddings as fingerprints to make prediction about an individual's cognitive behaviors. Finally, we validate the fusion model based on resting-state and memory task fMRI for the prediction of intelligence quotient (IQ), offering an insight on brain-cognitive behavior interactions. This is a joint work with Dr. Li Xiao.
14:35-15:35 CST
Social Hour/Poster Session
Tuesday, November 10, 2026
8:30-9:00 CST
Sign-in & Light Refreshments
9:00-10:00 CST
Connectomic constraints on intrinsic neural timescales
Speaker: Linden Parkes (Rutgers University)
TBA
10:00-10:05 CST
Tech Break
10:05-11:05 CST
Unlocking the Translational Potential of Connectomics Data
Speaker: Archana Venkataraman (Boston University)
Connectomics data has become ubiquitous in computational neuroscience by allowing us to
parse the intrinsic organization of the brain, identify biomarkers of different neural processes,
and discriminate between patient and control groups. While such methods have yielded novel
and valuable insights into the brain, there remains a notable gap between the data and models
we are analyzing and the patient care pathway. This talk will showcase recent work in my lab
that attempts to bridge this translational gap. First, I will snapshot our work on seizure detection
from resting-state EEG. We use a simple transformer architecture to combine spatial and
temporal information in the continuous EEG recordings. Our model accurately pinpoints not
only the time of seizure onset but also the involved areas of the scalp across a large clinical
dataset. We then turn our attention to resting-state fMRI. In the second project, I will describe
our multimodal deep learning framework that combines structural and functional connectivity.
The graph-based interactions in this model can accurately localize the seizure onset zone across
a heterogeneous cohort of temporal and extratemporal lobe epilepsy patients. Finally, I will
describe our seminal work that uses dynamic functional connectivity to simultaneously localize
the language and motor areas of the eloquent cortex for preoperative mapping. Our model
leverages specialized convolutional filters that extract graph-based features from the dynamic
connectivity matrices, an LSTM attention network to weigh the relevant time points and
multitask classification to simultaneously localize different eloquent subsystems. Taken together,
these projects underscore an important future for connectomics data in the realm of patient
care.
11:05-11:35 CST
Coffee Break
11:35-12:35 CST
Message in a bottleneck: architected interpretability for brain
Speaker: Sergey Plis (Georgia State University)
Deep learning models can predict diagnostic status from neuroimaging alone, linking brain measurements to clinical categories often defined by symptoms and behavior. Yet accurate prediction does not, by itself, explain the brain processes underlying those categories. How can we design models whose predictive success also helps us learn about brain dynamics?
Methods ranging from saliency analysis to mechanistic interpretability seek to explain how trained models make predictions. In this talk, I will present architected interpretability (ArchInterp), a complementary approach that makes an interpretable representation an explicit requirement of the architecture. A structured bottleneck between encoder and decoder requires information used for prediction to pass through a representation that we can inspect and constrain. The aim is to retain the predictive power of flexible neural networks while making what they learn accessible to scientific inquiry. In our work on functional neuroimaging, this bottleneck takes the form of a graph representing interactions among brain regions, making those interactions an explicit part of the model's computation.
Requiring the model to use this graph ties its predictive performance to the learned interactions, but does not by itself establish their biological meaning. Moving toward an explanation of brain dynamics therefore requires determining what the graph's edges represent and under what assumptions they support a causal interpretation. I will discuss recent advances in ArchInterp through this lens, showing how they inform our next steps toward causally interpretable models of functional neuroimaging. The broader goal is to use predictive models to develop and test explanations of how brain dynamics relate to behavior and disorder.
12:35-13:35 CST
Lunch Break
13:35-14:35 CST
A Bayesian deep learning framework for brain dynamic functional connectivity linked with cognition
Speaker: Yize Zhao (Yale University)
Recent neuroimaging studies have revealed that brain functional networks exhibit temporal fluctuations during cognitive tasks, sparking growing interest in dynamic functional connectivity. However, few existing methods effectively incorporate temporal dynamics when examining the relationship between brain connectivity and cognition. In this work, we propose a Bayesian deep learning framework that decomposes functional connectivity into dynamic and stationary components, thereby disentangling transient fluctuations from stable network structures. By leveraging deep kernel learning, the model flexibly represents transient functions as continuous processes, allowing the data to guide the discovery of meaningful temporal patterns beyond conventional assumptions. Building on this representation, the dynamic component is modeled via a thresholded Gaussian process, which captures smooth temporal variations while inducing sparsity to highlight significant temporal signals. Simultaneously, the stationary component is modeled using a deep neural network with an integrated thresholding layer, enabling the learning of complex nonlinear relationships and enhancing interpretability by emphasizing key connections. This integrative approach balances model flexibility, temporal resolution, and interpretability, providing a powerful tool for predicting cognitive outcomes from functional neuroimaging data. We demonstrate our method outperforms existing approaches in both prediction accuracy and feature selection through extensive simulations. Applied to data from the Adolescent Brain Cognitive Development (ABCD) study, our framework uncovers dynamic brain signals during cognitive tasks that significantly contribute to variations in cognitive performance.
14:35-15:00 CST
Coffee Break
15:00-16:00 CST
Lightning Talks
Wednesday, November 11, 2026
8:30-9:00 CST
Sign-in & Light Refreshments
9:00-10:00 CST
TBA
Speaker: Sharmistha Guha (Texas A&M University, College Station)
TBA
10:00-10:05 CST
Tech Break
10:05-11:05 CST
Being You, A Basic Algorithm for Machine Intelligence
Speaker: Guorong Wu (University of North Carolina at Chapel Hill)
Human cognition arises from coordinated spiking dynamics in distributed cortical columns, where perceptual representations are shaped by experience-dependent plasticity in the vertical connections across cortical layers. Inspired by this notion, our central premise is a two-layer computational model of sensorimotor learning: Upper-layer neurons encode sensory input while indexing a spatial reference frame anchored in the lower layer. Upon movement, the lower layer transitions to a new location, generating a prediction of the next sensory input in the upper layer. To that end, we develop a brain-inspired learning primitive in which cognition-level neural synchrony emerges through iterative bottom-up and top-down interactions between micro-scale dynamics of spiking neurons and a macro-scale mechanism of oscillatory synchronization. This research opens a new window for advanced learning-enabled intelligence by addressing fundamental questions of neural information representation and mechanisms of cognition.
11:05-11:35 CST
Coffee Break
11:35-12:35 CST
Dynamic resting state functional connectivity: A time-varying dynamic network model
Speaker: Fei Jiang (University of California, San Francisco (UCSF))
This talk will present two statistical methods for analyzing brain signal data. In the first application, we introduce a technique for extracting dynamic functional connectivity from MEG data, aimed at enhancing early screening for Alzheimer's disease. In the second application, we describe a deep learning framework designed to detect waveform abnormalities in EEG and MEG data to support epilepsy diagnosis.
12:35-13:35 CST
Lunch Break
13:35-14:35 CST
Lagrangian Causal Dynamics of Brain Networks
Speaker: Moo K. Chung (University of Wisconsin, Madison)
Causal interactions in brain networks are inherently dynamic and contain feedback loops that cannot be represented naturally by acyclic graphical models. We introduce a variational framework in which directed interactions are modeled as flows evolving according to a Lagrangian principle. The resulting dynamics, governed by the Hodge Laplacian, provide a continuous-time description of how causal interactions evolve, dissipate, and organize into feedback loops. Their long-time limit isolates cyclic interactions, establishing a direct connection between differential equations, topology, and causal structure: feedback loops emerge as persistent modes of an underlying dynamical system rather than as violations of an acyclic causal model. This framework provides a principled approach for studying cyclic information flow and a mathematical foundation for identifying recurrent interactions in time-varying brain networks. The framework shifts causal modeling from static directed graphs toward dynamical systems in which persistent cyclic structure naturally emerges from the geometry and long-term behavior of evolving causal interactions. The talk is based on arXiv:2606.08202.
14:35-15:00 CST
Coffee Break
15:00-16:00 CST
Lightning Talks
Thursday, November 12, 2026
8:30-9:00 CST
Sign-in & Light Refreshments
9:00-10:00 CST
From Low-Rank Connectivity to Task Subspaces: Disentangling the Sub-Circuits That Drive Neural Dynamics
Speaker: Anqi Wu (Georgia Institute of Technology)
Neural population activity simultaneously reflects sensory inputs, actions, choices, memory, and internal state. Linking circuits to computation requires answering two coupled questions: which recurrent interactions generate each of these signals, and when each signal is expressed during behavior. I will present two complementary latent dynamical models that address these questions directly from population recordings. First, I will introduce a low-rank recurrent network model with partial disentanglement. Because the recurrent connectivity is low-rank, the network's dynamics decompose into a small set of latent subspaces, and each subspace can be traced back to the sub-circuit that drives it. A partial-correlation penalty renders these subspaces statistically independent rather than merely orthogonal, which is what makes the recovered sub-connectivity interpretable. Validated on coupled ODE systems (Lorenz and Thomas attractors) and applied to monkey motor cortex and cortex-wide voltage imaging, the model recovers sub-circuits that are invisible in the full connectivity — including an air-puff-evoked oscillation confined to a single S2→M2→RSC pathway. Because each sub-circuit is isolated, it can be perturbed independently in silico. Second, I will describe a task-conditioned attention model for the regime where these assumptions break down: task variables are correlated rather than independent, and their subspaces are expressed at different, trial-varying times. Scalar task labels provide weak supervision while temporal attention learns when stimulus, choice, or memory information is encoded, without requiring trial alignment. Gradient saliency then maps each subspace back to its contributing neurons moment by moment, exposing how information flows between subspaces within a trial. Together, these approaches connect recurrent connectivity structure to the latent dynamics it generates, and extend that picture to non-stationary regimes where a single autonomous dynamical system is not sufficient.
10:00-10:05 CST
Tech Break
10:05-11:05 CST
Generative Models as Differential Equations: Structure, Regularity, and Computation
Speaker: Lars Ruthotto (Emory University)
Diffusion models and flow matching have achieved remarkable progress in image generation and are increasingly used to model scientific data, including brain imaging. This talk emphasizes their interpretation as differential equations that transport the data distribution to a known prior via a learned velocity field. That viewpoint exposes their theoretical underpinnings and grounds our discussion of three issues that arise for scientific data. (1) Many data types are confined to a lower-dimensional constraint set, where no ambient density exists and the score blows up; I present a constraint-aware perturbation that restores stable training. (2) When training samples are scarce, as in many neuroimaging studies, the exact minimizer of the training objective reproduces the training data; penalizing derivatives of the velocity gives a well-posed problem whose solution does not memorize. (3) Training continuous-time models is costly in memory and time. Low-precision arithmetic reduces both, provided the ODE solver and its adjoint are designed for it; our open-source package rampde does this and serves as a drop-in replacement for torchdiffeq. The talk will illustrate these results and some open issues with numerical examples.
11:05-11:35 CST
Coffee Break
11:35-12:35 CST
TBA
Speaker: Lu Zhang (Indiana University Indianapolis)
TBA
12:35-13:35 CST
Lunch Break
13:35-14:35 CST
Continuous biomarker dynamics refine event-based staging of Alzheimer’s disease progression
Speaker: Jingwen Yan (Indiana University Indianapolis)
Alzheimer’s disease (AD) evolves continuously over decades, motivating disease models that can resolve subtle progression and provide individualized biological staging. We developed a generative model that jointly estimates continuous biomarker trajectories and individual disease stage. In longitudinal amyloid PET, the model showed greater temporal consistency than DEBM and recovered interpretable regional progression patterns. Simulations showed that event-based staging provided limited resolution before and between abnormality events, whereas the generative model continuously tracked the underlying disease process. These findings support continuous biomarker modeling for higher-resolution and biologically interpretable staging of AD progression.
14:35-15:00 CST
Coffee Break
15:00-16:00 CST
Open Discussion
Friday, November 13, 2026
8:30-9:00 CST
Sign-in & Light Refreshments
9:00-10:00 CST
TBA
Speaker: Wenhao Zhang (University of Texas Southwestern Medical Center)
TBA
10:00-10:30 CST
Coffee Break
10:30-11:30 CST
Mapping the Developing Brain Before Birth: Fetal MRI Methods and Longitudinal Perspectives
Speaker: Ruiying Liu (Emory University)
Abstract: Fetal development lays the foundations for brain structure and function across the lifespan. MRI offers a unique window into this process, but fetal motion, rapidly changing anatomy, and the difficulty of tracking development across gestation present substantial challenges. This talk will present our work on fetal structural and functional MRI, focusing on motion correction and segmentation as foundations for studying early brain development. We will discuss how these methods enable the characterization of brain growth, emerging anatomical organization, and functional connectivity before birth. We will also explore how longitudinal imaging can reveal developmental trajectories and help relate structural maturation to changes in functional organization. Finally, we will place fetal imaging within a lifespan perspective, highlighting opportunities to connect prenatal development with later brain organization and dynamics.
11:30-11:45 CST
Workshop Survey and Close
Registration
IMSI is committed to making all of our programs and events inclusive and accessible.
Contact [email protected] to request
disability-related accommodations.
In order to register for this workshop, you must have an IMSI account and be logged in.