ART-MED LAB · RESEARCH · OCTOBER 2026
Digital Twin for Healthcare and Development
Generative models, neurophysiological observations, and uncertainty quantification support research on individualized, changing human states.
Human digital twins of neurophysiology
Our NSF FDT-BioTech collaboration (#2436216) investigates mathematical and statistical foundations for human digital twins of neurophysiological modeling and uncertainty quantification. I serve as Co-PI.
We investigate how neuroimaging and behavioral observations can inform individualized, time-evolving models. The research connects computational representations of neural dynamics with methods that describe uncertainty across heterogeneous participants and changing observations.

From heterogeneous observations to useful individual models
People differ in their neural activity, behavior, developmental histories, and responses to their environment. A useful computational model must account for those differences and remain interpretable when measurements are incomplete or change over time.
Our research direction examines reconstruction fidelity, stability under missing or changing inputs, and uncertainty across participants. Neuroimaging, physiological measures, behavioral observations, and contextual information provide complementary perspectives; their integration remains an active research objective.
- How do representation choices preserve individual differences in neural dynamics?
- How should uncertainty change when a measurement is missing or recording conditions differ?
- Which model outputs can support a testable hypothesis about longitudinal change?
- How can computational models inform, and be informed by, an embodied system’s observations?
Generative modeling of functional connectivity
Our comparative study of variational and conditional variational autoencoders examined functional connectivity in autism fMRI data, including the effects of architecture and phenotypic conditioning. This provides a computational foundation for studying heterogeneity in neural dynamics.
The work motivates a broader research program linking generative representations with individualized modeling. A model’s reconstruction or predictive performance is evaluated separately from the usefulness of an intended healthcare application.
Conditional Variational Autoencoder for Functional Connectivity Analysis, Bioengineering 2023 ↗
Development, longitudinal observation, and personalized assistance
Prospective applications include progression analysis and hypothesis-driven intervention planning. We aim to connect model-based observations with clinician-guided robotic activities that can gather contextual information and adapt over time.
Each application requires its own validation. Our studies will examine whether the models remain stable and informative across participants and sessions before interpreting their outputs as guidance for care.
Personalized assistive interaction → Medical AI and telemedical robotics →
Research gallery · Projects and support · Publications and CV