Speaker: Rajarshi Guhaniyogi, Associate Professor of Statistics, Texas A&M University

Title: Bridging Statistical, Scientific and Artificial Intelligence: Interpretable Deep Learning for Complex Functional and Imaging Data

Abstract: The rapid growth of large structured datasets presents both exciting opportunities and significant challenges for modern statistical inference. In this talk, I will focus on two motivating problems: (1) building scalable functional surrogates for computer simulation studies in Sea, Lake and Overland Surge Heights (SLOSH) simulator, and (2) predicting amplitude of spatially indexed low-frequency fluctuations (ALFF) in resting state functional magnetic resonance imaging (fMRI) as a function of cortical structural features and a multi-task co-activation network capturing coordinated patterns of brain activation in a large neuroimaging study of adolescents. While carefully designed hierarchical Bayesian methods provide a principled framework for inference with rigorous uncertainty quantification in such datasets, their scalability is often limited in high-dimensional and large-scale settings. To address these limitations, we develop deep neural network (DNN)-based generative models specifically designed for functional outputs with vector, functional, and network-valued inputs. The proposed framework provides interpretable inference through calibrated uncertainty measures and allows rapid computation in large-sample, high-resolution regimes. They also establish connections between Bayesian nonparametrics and modern deep learning, showing how deep Gaussian process priors and functional and object-valued regression can be unified within a scalable generative modeling framework. This is a joint work with the postdoctoral scholar Yeseul Jeon, and scientists from UC San Francisco Medical school and Los Alamos National Laboratories.