LEXINGTON, Ky. (Jan. 16, 2026) — As the nation approaches the 250th anniversary of the founding of the United States, the University of Kentucky Department of History is launching a public event series designed to deepen understanding of the American Revolution and its enduring legacy.
Dr. Steve Turley, Brigham Young University
Title: Using Physics in Unusual Places
Abstract: I would sometimes tell students that if they didn’t know what major to choose, they should choose physics because it is the basis of everything else. While this is perhaps a bit overstated, it is valuable for faculty members to keep in mind that most of our students will have careers that look different than our academic pursuits.
I will discuss physics applications I have found outside typical academic settings. As part of an exotic weapons development program, I participated in some of the early development of ultra-cold atoms, the optical Stern-Gerlach Effect and the development of a coherent Lyman-alpha source. While studying efficient ways to compute radar cross sections of stealthy targets, I not only used my background in electromagnetic theory but also some machinery from General Relativity and quantum mechanics.
Work on measuring lifetimes of parts in ion thruster satellite engines used results from astrophysics. After a 25-year academic career, I have been assisting as a volunteer at FamilySearch, an international nonprofit collaborative genealogical platform. To my surprise and delight, I’ve found ways my physics background can be applied to problems in computerizing and indexing genealogical records, preserving privacy, optical character recognition and matching records to family trees.
Title: Optimizing Sampling and Diversity for Differentiable Search and Data Labeling Using Large Language Models for eCommerce
Abstract: While Deep Learning and Generative AI have transformed information retrieval, their reliability and efficiency ultimately depend on foundational statistical principles regarding distribution, diversity and sampling. This talk synthesizes research across search indexing and automated data labeling to demonstrate how classical statistical methodologies can solve critical bottlenecks in modern neural architectures.
By moving beyond the "black box" interpretation of models, we show that enforcing such statistical constraints as maximizing marginal relevance or optimizing sampling distributions can significantly enhance system performance compared to standard deep learning baselines. In this talk, we examine Differentiable Search Indexing (DSI), showing that modifying training objectives with a Maximal Marginal Relevance (MMR)-inspired diversity component forces the model to learn a more representative distribution of information, balancing relevance with diversity.
Also, we address data scarcity in Large Language Models (LLMs) through Active Learning, demonstrating that treating LLMs as probabilistic engines requiring rigorous uncertainty and diversity sampling strategies drastically reduces annotation costs while maintaining high accuracy. These applications illustrate how concepts like loss function optimization and experimental design remain central to advancing state-of-the-art AI systems. The talk illustrates how theoretical statistical ideas translate into real-world industry applications and offers insight into pathways toward industry careers.
Title: Robust trial augmentation using external data
Abstract: Randomized trials often have sample sizes that are too small to produce precise estimates of treatment effects. One approach for improving trial efficiency is to incorporate external data from previously completed trials or observational studies into the estimation process. When the external data are aligned with the trial data and statistical models for nuisance functions are correctly specified, using the external data can yield consistent estimates and enhance efficiency.
Some degree of misalignment or misspecification, however, is usually expected and can threaten trial validity. We develop a class of estimators that exploit randomization to ensure consistency and asymptotic normality, even when the external data are misaligned with the trial. We also propose a procedure that uses members of this class to construct a combined estimator that is consistent and asymptotically normal and can leverage external data even when that data are misaligned with the trial, or when models for nuisance functions are mis-specified or have slow convergence rates.
We show that the efficiency of the combined estimator is no lower than that of each of its component estimators (including the efficient trial-only estimator, if it is used as a component for the combined estimator). Our methods allow investigators to use external data to improve the trial's efficiency without concern for misalignment between the external data and the trial. We examine the finite-sample behavior of the proposed methods in simulation studies and apply them to analyze data from a trial comparing coronary artery bypass grafting surgery plus medical therapy versus medical therapy alone for patients with chronic coronary artery disease.
Title: Statistical ranking with dynamic covariates
Abstract: The Plackett–Luce (PL) model has long been used for rank aggregation in sports analytics and social choice. In this talk, we introduce a covariate-assisted ranking model within the PL framework that incorporates dynamic covariates. This added flexibility enables individualized and dynamic rankings and improves model fit, but it also introduces challenges for analysis. We address these challenges in the context of maximum likelihood estimation (MLE) under a general network topology. Specifically, we establish conditions for model identifiability and the unique existence of the MLE, propose an alternating maximization algorithm for computing the MLE, and prove a uniform consistency result. We illustrate the proposed model through applications to ATP tennis data spanning the past 40+ years and to a horse racing dataset.
Title: Designing Intelligent AI: Insights from Human Cognition
Abstract: Modern AI systems are evolving from passive tools toward agent-based systems that can reason, learn and interact with their environments. This talk discusses emerging research directions in generative AI, drawing inspiration from principles of human cognition including continuous learning and adaptation, effective knowledge transfer and multi-objective decision making. The goal is to stimulate new perspectives on building domain-aware and trustworthy AI systems that can operate reliably in complex, real-world settings.
Title: Bayesian Model Criticism: From Holdout Checks to Model Comparison
Abstract: In this talk, I will cover two recent works on Bayesian model criticism. The first is Holdout Predictive Checks (HPCs). HPCs are built on posterior predictive checks (PPCs), which check a model by assessing the posterior predictive distribution on the observed data. PPCs, however, use the data twice: to calculate the posterior predictive and to evaluate it. This situation can lead to uncalibrated p-values. HPCs, in contrast, compare the posterior predictive distribution to a draw from the population distribution, a held-out dataset. This method blends Bayesian modeling with frequentist assessment.
Unlike the PPC, we prove that the HPC is properly calibrated. Empirically, we study HPCs on classical regression, a hierarchical model of text data and factor analysis. In the second work, we introduce the posterior predictive null check (PPN), a method for Bayesian model criticism that helps characterize the relationships between models. The idea behind the PPN is to check whether data from one model's predictive distribution can pass a predictive check designed for another model. This form of criticism complements the classical predictive check by providing a comparative tool.
A collection of PPNs, which we call a PPN study, can help us understand which models are equivalent and which models provide different perspectives on the data. With mixture models, we demonstrate how a PPN study, along with traditional predictive checks, can help select the number of components by the principle of parsimony. With probabilistic factor models, we demonstrate how a PPN study can help understand relationships between different classes of models, such as linear models and models based on neural networks. Finally, we also discuss ongoing work on aggregated posterior checks.
Title: Doubly robust estimation of causal effects for random object outcomes with continuous treatments
Authors: Satarupa Bhattacharjee, Bing Li, Xiao Wu, Lingzhou Xue
Abstract: Causal inference is central to statistics and scientific discovery, enabling researchers to identify cause-and-effect relationships beyond associations. Although traditionally studied within Euclidean spaces, contemporary applications increasingly involve complex, non-Euclidean data structures that reside in abstract metric spaces known as random objects, such as images, shapes, networks, and distributions.
This paper introduces a novel framework for causal inference with continuous treatments applied to non-Euclidean data. To address the challenges posed by the lack of linear structures, we leverage Hilbert space embeddings of the metric spaces to facilitate Frechet mean estimation and causal effect mapping. Motivated by a study on the impact of exposure to fine particulate matter on age-at-death distributions across U.S. counties, we propose a nonparametric, doubly-debiased causal inference approach for outcomes as random objects with continuous treatments.
Our framework accommodates moderately high-dimensional vector-valued confounders and derives efficient influence functions for estimation, ensuring both robustness and interpretability. We establish asymptotic properties of the cross-fitted estimators and employ conformal inference techniques for counterfactual outcome prediction. Validated in both simulation and real-world environmental application, our framework extends causal inference methodologies to complex data structures, broadening its applicability across scientific disciplines.
This event will now be virtual. Click here to join
Join us for a lecture by Pulitzer Prize winner Joseph Ellis, moderated by University of Kentucky professor Amy Murrell Taylor.
About Joseph Ellis
The author of 12 books, Ellis was awarded the Pulitzer Prize for "Founding Brothers: The Revolutionary Generation" and won the National Book Award for "American Sphinx," a biography of Thomas Jefferson. He has taught at Mount Holyoke College, the University of Massachusetts and the U.S. Military Academy at West Point. His commentaries have been featured on CSPAN, CNN, and PBS’s News Hour, and he appears in the major new PBS documentary "The American Revolution."
Ellis’ latest work, "The Great Contradiction," examines how a government that had been justified and founded on the principles articulated in the Declaration of Independence institutionalized slavery and created a tidal wave of western migration by settlers who understood the phrase “pursuit of happiness” to mean the pursuit of Indian lands.
Click here to join virtually
Join us for Annual Graduate Alumni Day, where three UK Mathematics alumni from the University of South Florida, Georgia Tech, and Microsoft return to share the career paths and insights from life in and beyond academia. Come ready to listen, ask and leave with a clearer picture of what's possible.
Schedule:
Visitor Parking
Visitor parking is available nearby. Rates are $2/hour unless noted.
Garage: Cornerstone Garage (PS #5) has entrances on South Limestone and Upper Street. Cost is $2/hour with a $30/exit maximum.
Surface Lots (pay at the meter):
For more visitor parking information, visit transportation.uky.edu/visitoroptions.
