Statistics and Data Science Seminar
This seminar typically meets on day(s): W at 4 p.m., though exceptions may exist.
Upcoming Seminars
The following seminars are upcoming, you may instead view past events in this series.
Oct. 21, 2026
Kai Zhang of
UNC :
5 p.m. in 636 SEO
Abstract
Interpretability and computational efficiency are central challenges in AI. This talk explores how data bits can address these challenges while providing a common foundation for statistical inference and modeling. We begin with the binary expansion approximation of uniformity (BEAUTY) equation, which approximates the characteristic function of any copula distribution, in any dimension, using expectations of interactions among data bits. Building on this representation, we develop distribution-free tests of independence through symmetry statistics of data bits. These tests are powerful, interpretable, and computational efficient. We then use the atomic linearity of data bits to express the conditional mean of a binary response as a linear combination of interactions among predictor bits. This representation provides new insights into generalized linear models and neural networks. Finally, we present the binary expansion group intersection network (BEGIN), which characterizes conditional independence among binary vectors in terms of covariance matrices and graphical models. Together, these developments provide a foundation for AI methods that combine flexible modeling, transparent statistical explanations, and efficient computation.
Hans-Georg Müller of
UC Davis :
4 p.m. in 636 SEO
Abstract
Adjusting for confounding and imbalance when establishing statistical relationships is an important task. Causal inference methods have emerged as a popular tool to achieve this for regression relationships with scalar responses and also distributional responses. We introduce a general framework for causal inference when responses reside in geodesic metric spaces, drawing on a novel geodesic calculus, which facilitates the quantification of treatment effects through the concept of geodesic average treatment effect (GATE). Employing Fréchet regression, we obtain doubly robust estimation of GATE with rates of convergence. Theoretical results are complemented with various data illustrations. Based on joint work with Daisuke Kuriso, Yidong Zhou and Taisuke Otsu.
Oct. 28, 2026
Dr. Jiehuan Sun of
UIC Biostat :
4:30 p.m. in SEO 636
Nov. 4, 2026
Xinran Li of
University of Chicago :
4:30 p.m. in 636 SEO
Abstract
TBA
Nov. 18, 2026
Soumendra Lahiri of
Washington University in St. Louis :
4:30 p.m. in 636 SEO
Abstract
TBA
Dec. 2, 2026
Jane-Ling Wang of
University of California, Davis :
4:30 p.m. in 636 SEO
Abstract
TBA