2025 Seminars

Seminar Announcement

Kexin Xie
Ph.D. Candidate, Department of Statistics

Virginia Tech

"Beyond A×B: Estimation and Variable Selection for
Interpretable Models with Conditional Main Effects"

November 19, 2025, 1:30pm-2:20pm (CT)
E134 Howe Russell

Understanding interaction effects among variables is important for statistical and machine learning models in various applications. The conventional approach of quantifying interactions as the product of variables often lacks clear interpretability, especially in complex systems. The concept of conditional main effects (CME) provides a more interpretable framework for capturing interaction effects by quantifying the effect of one variable conditional on the level of another. In this talk, we investigate the limitation of a bi-level variable selection of significant CMEs due to the lack of considering the coupling problem between variables. To address these limitations, we propose an adaptive CMEnet method for CME selection under the generalized linear model (GLM) framework. The proposed approach considers the hierarchical adaptive penalization to effectively mitigate coupling issues while improving the accuracy and interpretability of variable selection. This proposed method is particularly suited for large-scale observational and experimental datasets, with applications in domains such as gene association studies, clinical trials, and public health.

Seminar Announcement

Satabdi Saha

Postdoctoral Fellow, Department of Biostatistics, The University of Texas

MD Anderson Cancer Center

"Bayesian Variable Selection for Compositional Microbiome Data"

November 14, 2025, 1:30pm-2:20pm (CT)
145 Coates

Ongoing advancements in microbiome profiling have provided unprecedented insights into the molecular dynamics of microbial communities, sparking a surge of interest in uncovering the microbiome’s critical role in human health. Identifying microbial features linked to clinical outcomes, however, remains challenging due to the high-dimensional, sparse, and compositional nature of microbiome data. Additionally, many microbial taxa, although classified as distinct, may share functional roles, complicating traditional variable selection methods.

To address this, we developed two complementary Bayesian frameworks. Our first approach, BRACE provides a unified nonparametric approach for data-adaptive feature aggregation and selection. By clustering regression coefficients, BRACE identifies groups of taxa with similar functional impacts, yielding interpretable models and effective dimension reduction without relying on external information. In contrast, our second approach, GRACE investigates the role of external feature similarity graphs. It combines variable selection with outcome driven graph learning, characterizing how recovery of outcome relevant networks changes when guided by external, outcome-agnostic graphs such as phylogeny or co-occurrence.

Both methods explicitly respect compositionality and have shown strong performance in oral microbiome applications, together offering new strategies for interpretable regression models in compositional data settings.

Seminar Announcement

Francesco Giuseppe Cordoni, Ph.D.

Associate Professor, Laboratory of Applied Mathematics

Department of Civil, Environmental and Mechanical Engineering

University of Trento in Trento, ITALY

"Using Mathematical Models with Machine and Deep Learning to Unravel the Biological Effects of Radiation from Single-Cell Data to Patient Outcomes"

August 21, 2025, 1:30pm-2:20pm
116 Stubbs Hall

In this talk, I will explore two approaches to understanding the biological effects of radiation in particle therapy: mathematical modeling and advancements in Artificial Intelligence (AI). I will present the development of the Generalized Stochastic Microdosimetric Model (GSM2), from its foundational principles and initial in vitro validation to recent advancements. These include a multiscale extension incorporating fast chemical networks to elucidate a novel, yet unexplained, mechanism in radiobiology critical to FLASH radiotherapy and a new agent-based model for predicting patient outcomes in radiotherapy.

Additionally, I will introduce ANAKIN, a data-driven Machine and Deep Learning model designed to predict both in vitro and clinical outcomes while offering valuable insights into underlying biological mechanisms. I will demonstrate how integrating multiple spatial scales and diverse descriptions of radiation interactions with biological tissue enhances our ability to predict and understand the
biological effects of radiation.

Seminar Announcement

Katherine Thompson
Associate Professor, Dr. Bing Zhang Department of Statistics
University of Kentucky

"Artificial Intelligence and Machine Learning Use Cases: Immersive Training Examples"

May 2, 2025, 1:30pm to 2:25pm
Red River, 323 Student Union

Artificial Intelligence (AI) and Machine Learning (ML) are rapidly expanding fields transforming education, research, and everyday life. This presentation will overview AI applications from agriculture to medicine and will describe how a gathering place for AI/ML consumers, users, and developers at the University of Kentucky produced immersive learning experiences for undergraduate and graduate students alike. Details will be provided about the University of Kentucky AI/ML Hub student learning experience at the interface of healthcare and ML. Using data from a Kentucky hospital, graduate students from multiple colleges received interdisciplinary training to build ML models that identify patients in Kentucky at high risk for missing colorectal cancer screenings in the Eastern region of Kentucky.

Seminar Announcement

Dr. Jenny Price

Nigatta University, Japan

"Machine Learning for Predicting Type 2 Diabetes"

April 25, 2025, 1:30pm to 2:25pm
Room: 116 Stubbs Hall

In this talk, Dr. Price will discuss the problem of healthcare becoming more difficult as the population grows older, and the less available specialists there are for diseases, specifically in this talk for type 2 diabetes. As more people need to see specialist care, but have no availability to do so, general practitioners step in to meet the need. Which therefore leads to an increased need for machine learning to help fill the gap between these two. We will discuss the problems currently facing machine learning as we try to tackle the challenge of medical data, distrust from medical practitioners, and machine learning limitations in healthcare to hopefully help to improve the care of patients. By using a combination of models, over sampling, cross validation, and continuous learning we can create models that can create a usable model for practitioners to use as a second opinion.

Seminar Announcement

Daiqi Gao

Postdoctoral Fellow

Department of Statistics

Harvard University

"Online Reinforcement Learning for Personalized Digital Interventions"

April 11, 2025, 1:30pm to 2:25pm
Room: M-Red River 329 LSU Student Union

HeartSteps is a mobile health intervention designed to help sedentary users increase and maintain physical activity. A key component involves
delivering activity suggestions via smartphone notifications to encourage short bouts of physical activity (PA). We use online reinforcement learning (RL) algorithms to learn the optimal policy for delivering notifications based on each user's evolving needs. The reward in the RL algorithm is defined as a user's commitment to PA, measured through weekly surveys. While there are 14 decision points per week, the reward is only observed at the end of the week, creating a challenge of bagged decision times with non-Markovian and non-stationary dynamics within each bag. To address this challenge, we leverage an expert-provided causal directed acyclic graph (DAG) to construct the states and formulate the problem as a periodic Markov decision process.

Seminar Announcement

Li Hsiang Lin

Assistant Professor, Department of Mathematics and Statistics

Georgia State University

"Deep P-Spline: Fast Tuning, Theory, and Application"

March 28, 2025, 1:30pm to 2:25pm
Room: LSU Student Union, C-Capital Chamber 329

Optimizing the structure of deep neural networks (DNNs), including the number of neurons and layers, is a challenging task. This talk introduces Deep Penalized Spline (DPS), a novel framework that redefines neuron selection as a knot placement problem in spline expansions. By incorporating a diJerence penalty, DPS automates the network structure selection process, enhancing computational eJiciency and precision of conventional DNN. Embedded within a latent variable model, the method enables an eJicient algorithm for optimizing the network architectures, bridging statistical methodologies with modern network design. Numerical studies demonstrate the scalability and practical effectiveness of DPS. Theoretical properties and diverse applications of DPS will also be explored.

Seminar Announcement

Yong Zang

Showalter Scholar Associate Professor, Indiana University School of Medicine

"Curve-free dose optimization designs for oncology trials"

March 14, 2025, 1:30pm to 2:25pm
Room: LSU Student Union, M-Castilian 304

We propose two curve-free phase I/II clinical trial designs to optimize the dose of molecularly targeted agents (MTAs) and immunotherapies (ITs) by jointly modeling toxicity, efficacy, and immune response outcomes. Instead of employing complex parametric models, our approach leverages the inherent correlations among different clinical outcomes and incorporates the constrained dose-outcome order to enable efficient information sharing across doses. These designs enhance both the efficiency and transparency required for practical implementation in clinical settings. We also provide simulation studies and software demonstrations to illustrate the application of the proposed designs.

Seminar Announcement

Jeremy Brown

Professor, Department of Biological Sciences
Museum of Natural Sciences
Louisiana State University

"Bayesian Inference of the Evolutionary Past Using Genomic Data"

February 7, 2025, 1:30pm to 2:25pm
116 Stubbs Hall

Biologists have been interested in reconstructing the history of biological diversification on Earth ever since the time of Darwin. This field of study has since become known as phylogenetics. Phylogenetics was revolutionized starting in the early 1980s with the application of formal statistical approaches to inference using genetic data. Technical advances in the mid-2000s made it possible to sequence full genomes much more cheaply and quickly than ever before, which has radically increased the amount of available data. However, many challenges and open questions remain about how best to handle the scale and complexity of full-genome data in likelihood-based statistical frameworks. In this talk, I will discuss some strategies that my lab is pursuing to more efficiently and effectively reconstruct the evolutionary past from genomic data using Bayesian methods.