Statistical Inference and AI Modeling of High-Dimensional Longitudinal Data

In this study, a longitudinal regression model for covariance matrix outcomes is introduced. The proposal considers a multilevel generalized linear model for regressing covariance matrices on (time-varying) predictors. This model simultaneously identifies covariate-associated components from covariance matrices, estimates regression coefficients, and captures the within-subject variation in the covariance matrices. Optimal estimators are proposed for both low-dimensional and high-dimensional cases by maximizing the (approximated) hierarchical-likelihood function. These estimators are proved to be asymptotically consistent, where the proposed covariance matrix estimator is the most efficient under the low-dimensional case and achieves the uniformly minimum quadratic loss among all linear combinations of the identity matrix and the sample covariance matrix under the high-dimensional case. Through extensive simulation studies, the proposed approach achieves good performance in identifying the covariate-related components and estimating the model parameters. Applying to a longitudinal resting-state functional magnetic resonance imaging data set from the Alzheimer's Disease (AD) Neuroimaging Initiative, the proposed approach identifies brain networks that demonstrate the difference between males and females at different disease stages. The findings are in line with existing knowledge of AD and the method improves the statistical power over the analysis of cross-sectional data. 

This talk will first describe the mathematical-statistics framework for representing, modeling, and utilizing invariance and equivariance properties of deep neural networks. By drawing direct parallels between characterizations of invariance and equivariance principles, probabilistic symmetry, and statistical inference, we explore the foundational properties underpinning reliability in deep learning models. We examine the group-theoretic invariance in a number of deep neural networks including, multilayer perceptrons, convolutional networks, transformers, variational autoencoders, and steerable neural networks.
Understanding the theoretical foundation underpinning deep neural network invariance is critical for reliable estimation of prior-predictive distributions, accurate calculations of posterior inference, and consistent AI prediction, classification, and forecasting. Two relevant data studies will be presented: one is on a theoretical physics dataset, the other is on an fMRI music dataset. Some biomedical and imaging applications are discussed at the end. 

With the advance of science and technology, more and more data are collected in the form of functions. A fundamental question for a pair of random functions is to test whether they are independent. This problem becomes quite challenging when the random trajectories are sampled irregularly and sparsely for each subject. In other words, each random function is only sampled at a few time-points, and these time-points vary with subjects. Furthermore, the observed data may contain noise. To the best of our knowledge, there exists no consistent test in the literature to test the independence of sparsely observed functional data. We show in this work that testing pointwise independence simultaneously is feasible. The test statistics are constructed by integrating pointwise distance covariances (Székely et al., 2007) and are shown to converge, at a certain rate, to their corresponding population counterparts, which characterize the simultaneous pointwise independence of two random functions. The performance of the proposed methods is further verified by Monte Carlo simulations and analysis of real data.