When: Wednesday, February 11, 10:00 AM
Where: Tyler 053

Abstract
Neuroimaging data present fundamental statistical challenges: they are high-dimensional and exhibit complex structures. In this talk, I present Bayesian methods developed for functional data and matrix-valued data motivated by neuroimaging applications, emphasizing interpretability, scalability, and uncertainty quantification.

I first introduce Bayesian methods developed for two types of functional data derived from evoked electroencephalogram (EEG) signals in multi-condition settings, including (1) time-frequency representations and (2) dynamic functional connectivity. The proposed models jointly account for the data’s multilevel structure, functional nature, and subject-level covariates by incorporating covariate-dependent fixed effects and multilevel random effects. These methods are evaluated through simulations and applied to EEG data examining the effects of alcoholism on cognitive processing in response to visual stimuli. I then introduce Bayesian methods developed for matrix-valued data, with applications to predicting scalar outcomes using resting-state functional magnetic resonance imaging (fMRI) functional connectivity. Lastly, I conclude by discussing future directions for developing structure-aware statistical models for neuroimaging data and beyond.