Department Seminars

See the Campus Event Calendar for details about upcoming seminars

Day and Time: Fridays from noon to 1 pm central time unless otherwise noted.

Location: Morgridge Hall Seminar Room – 7560 unless otherwise noted.

Zoom: https://uwmadison.zoom.us/j/99879638765?pwd=wbtqxoucEFIlPVCVc9SFbvKB1Av7Xk.1

Passcode: 343271

Further details should be available about a week before the seminar.

Upcoming Spring 2026 Seminars

January 23 – BDS Rotation (speakers TBD)

January 30 – Daiwei Zhou (VT)

February 6 – Claudia Solís-Lemus (UW-Madison)

February 13 – TBD

February 20 – Didong Li (UNC)

February 27 – Erin Molloy (UMD)

March 6 – Moo Kyung Chung (UW-Madison) LOCATION CHANGE to Morgridge Hall Room 2516

March 13 – Halil Kilicoglu (UIUC)

March 20 – TBD

March 27 – Spring Break (no seminar)

April 3 – Pengtao Xie (UCSD)

April 10 – Fei Zou (UNC)

April 17 – Rafael Ferreira (UW-Madison) LOCATION CHANGE: Morgridge Hall Room 2516

April 24 – George P. Tiley (NCSU) LOCATION CHANGE: Morgridge Hall Room 2516

April 30 – Jenna Wiens (UMich) – NOTE THURSDAY DATE

May 1 – BDS Rotation (Speakers TBD)

To subscribe to the BMI Seminar mailing list email join-biostat-seminar@lists.wisc.edu.

Completed 2025/26 Seminars

September 5

BDS Rotation Presentation

September 12

Speaker: Florian Willomitzer (University of Arizona)

Title: Computational Imaging on Optically Challenging Surfaces and Through Scattering Media – About the Fun of Utilizing Nature’s Limits

Poster: Willomitzer_Title_Abstract_Bio

Abstract: Computational imaging and display principles are “enabling technologies” with the potential to drive transformational changes across multiple future application scenarios: Novel breeds of cameras could see through deep tissue, fog, or smoke. Precise and fast 3D scanners could enhance medical diagnosis and therapy, and become essential for measuring dynamic scenes during robotic surgery, autonomous navigation, or additive manufacturing. Novel 3D display and eye-tracking methods could enable the next wave in AR/VR. Amidst these seemingly endless possibilities, the knowledge about fundamental physical and information-theoretical limits in computational imaging proves to be a powerful tool: Limits often manifest as uncertainty products, allowing us to optimize specific system parameters (e.g., speed) by trading off less critical information for a given application. Moreover, recognizing that our imaging system already operates at the physical limit (e.g., of resolution) can help avoid unnecessary effort and investment. In this talk, I will highlight the virtue of limits in computational imaging by discussing recent research directions of our group: Among other topics, I will introduce a set of techniques that use so-called “synthetic waves” for computational holographic imaging through scattering media, such as biological tissue, and allow the capture of “light-in-flight” information without the need for pulsed lasers or fast detectors. Furthermore, I will present novel principles for 3D imaging on surfaces with complex reflectance, enabling, for example, fundamentally new techniques for accurate and fast eye tracking.

September 19

Speaker: Yongyi Guo, University of Wisconsin-Madison

Title: Personalizing Digital Health with Reinforcement Learning: Design and Inference in Adaptive Experiments

Poster: 0919-Guo-poster.docx

Abstract: Digital health interventions offer the opportunity to deliver personalized support at scale, but designing and evaluating such interventions raises new statistical and methodological challenges. In this talk, I will describe my work on MiWaves, a mobile health study aimed at reducing cannabis use among emerging adults, where intervention messages were delivered via a just-in-time adaptive intervention (JITAI). To enable personalization, we used a reinforcement learning (RL) framework, where a mixed-effects bandit algorithm adaptively tailored micro-support messages to participants based on their individual histories.

While adaptive algorithms improve personalization, they also complicate statistical inference: user trajectories are no longer independent, and classical estimators may fail. I will show how valid inference on key causal quantities, such as causal excursion effects, can still be achieved by leveraging the structure of the bandit algorithm. Specifically, I introduce a statistic–state decomposition of the RL policy, which separates pooled statistics across users from user-specific histories. This decomposition enables asymptotic normality of estimators, construction of hypothesis tests, and a clear connection to the posterior distribution of fixed (pooled) and random (personalized) effects. Beyond providing valid inference, this framework also offers domain scientists interpretable insights into how information is aggregated at both the population and individual levels, allowing them to critique, refine, or redesign policies.

This work underscores both the promise and the challenges of applying RL methods in real-world health settings: we can adaptively personalize support while still preserving the ability to draw reliable scientific conclusions.

September 26 – no seminar – Morgridge Hall Opening Celebration

October 3

Speaker: Benjamin Lengerich, University of Wisconsin-Madison

Title: What’s the Point of Interpretability in Biomedical AI?

Poster: 1003-Lengerich-poster.docx

Abstract: AI models, especially foundation model–based approaches, have become remarkably effective across a wide range of biomedical and scientific tasks. Their success raises a natural question: should we still care about interpretability?

In this talk, I will present a system-level argument for interpretability in biomedicine. In this view, interpretability is valuable for at least three reasons: (1) information acquisition, by identifying which measurements are most worth collecting; (2) modularity, by enabling models to be tested, swapped, and reused; and (3) value alignment, by ensuring that model reasoning connects to trustworthy interventions. I will illustrate these roles with examples from maternal health (where interpretability revealed a new metric for risk assessment), treatment effect heterogeneity (where explanations pointed to intervention strategies), and personalized gene networks (where modularity enables reuse of information across contexts).

October 10

Speaker: Wenyi Wang, University of Texas MD Anderson Cancer Center

Title: Deciphering tumor heterogeneity for benefits from immunotherapy in cancer

Poster: 1010-Wang-poster.docx

Abstract: Intra-tumor heterogeneity is characterized by a diverse population of tumor clones and subclones which are important drivers of tumor evolution and therapeutic response. However, accurate subclonal reconstruction at scale remains challenging. We developed a machine learning based method, CliPP, and surveyed 10,409 tumors from 32 cancer types. We found that high subclonal mutation fraction (sMF), the fraction of subclonal single nucleotide variants (SNVs) to all SNVs in the coding region, was prognostic of survival (progression free survival or overall survival) in 18 cancer types. In 14 cancers with low to moderate tumor mutation burden (TMB), high sMF was associated with better prognosis. In four cancers with high TMB, the opposite association was observed. In immunotherapy trials for advanced prostate cancer, a low-TMB cancer, high sMF was predictive of favorable response to ipilimumab and associated with increased CD8+ T-cell infiltration. The biphasic property of sMF that is distinct between cancers with low-moderate TMB and high TMB is further replicated within the SU2C-MARK (n=227) lung cancer cohort, where both directions of associations were observed in patients treated with immune checkpoint blockade (ICB). Our study highlights sMF as a key feature of cancer evolution, with its accurate measurement from DNA sequencing data being supported by CliPP. Our findings with response to ICB therapy advocates using sMF and TMB jointly as a marker of interplay between evolutionary dynamics and immune environments.

October 17

Speaker: Prof. Grace Y. Yi, University of Western Ontario

Title: Two Disciplines, One Mission – A Comparative View on Making Sense of Imperfect Data from Statistical Science to Machine Learning

Poster: 1017-Yi-poster.docx

Abstract: In the data-driven era, data quality plays a pivotal role in ensuring valid statistical inference and robust machine learning performance. Yet, imperfections such as measurement error in predictors and label noise in supervised learning are pervasive across a wide range of domains, including health sciences, epidemiology, economics, and beyond. These imperfections can obscure true patterns, introduce bias, and compromise the reliability of analyses. Such issues have attracted extensive attention from both the statistical and machine learning communities. In this talk, I will offer a brief comparative review of approaches in statistical science and machine learning, highlighting the importance of addressing data quality issues and developing strategies to mitigate their adverse effects on inference and prediction.

October 24

Speaker: Quefeng Li (University of North Carolina)

Title: Inference on the Significance of Modalities in Multimodal Generalized Linear Models

Poster: 1024-Li-poster

Abstract: Multimodal statistical models have gained attention in recent years, yet there lacks rigorous statistical inference tools for inferring the significance of a single modality within a multimodal model. This inference problem is particularly challenging in high-dimensional multimodal models. In the context of high-dimensional multimodal generalized linear models, we propose a novel entropy-based metric, called the expected relative entropy, to quantify the information gain of one modality in addition to all other modalities in the model. We develop a deviance-based statistic to estimate the expected relative entropy and prove that this statistic is consistent and show that its asymptotic distribution can be approximated by a non-central chi-squared distribution. That enables the calculation of confidence intervals and p-values to assess the significance of the expected relative entropy for a given modality. Such an inference tool is useful for ranking the importance of data modalities and making personalized treatment recommendations based on individual patient profiles.

October 31

Speaker: Yang Lu (University of Wisconsin-Madison)

Title: Advancing Agent-Based, Data-Driven, and Trustworthy Hypothesis Generation in Biomedical Research

Poster: 1031-Lu-poster.docx

Abstract:

Rapid developments in high-throughput sequencing have enabled biologists to collect large volumes of multi-omics data with unprecedented resolution. However, interpretation of such an increasing amount of heterogeneous biological data becomes highly nontrivial. In my talk, I will present an agent-based and data-driven research paradigm to discover testable hypotheses directly from biological data in an interpretable and trustworthy fashion. In particular, the talk will focus on three recent works that address key aspects of biomedical research: analyzing data, generating and prioritizing hypotheses, and engaging with users:

(1) An interpretation method that detects non-additive interactions from any machine learning (ML) models. The detected interactions, treated as hypotheses, are rigorously controlled for statistical errors without relying on p-values. This method was the first to demonstrate to the community that higher-order interpretations of ML models can be achieved with confidence guarantees.

(2) An AI-driven agent that automatically translates biologists’ needs into actionable insights. The agent we developed enables the automatic execution of off-the-shelf Python-based bioinformatics tools, allowing researchers to generate analysis results with minimal tool-specific knowledge and coding expertise. This method was the first initiative to streamline the automatic and codeless execution of general-purpose bioinformatics tasks via conversation.

(3) A critical reevaluation of problematic statistical estimation of the Basic Alignment Search Tool (BLAST), a cornerstone tool used in daily biomedical analysis over the past 30 years. We have introduced an alternative method to address this issue, ensuring that it does not yield inflated estimates of significance. Our study has the potential to influence and reshape numerous conclusions drawn by researchers.

November 7

Speaker: Huan Sun (Ohio State University)

Title: Advancing the Capability and Safety of Computer-Use Agents Together

Poster: 1107-Sun-poster.pdf

Abstract: Nowadays, agents’ capabilities are advancing rapidly, but their safety (and security) is not keeping pace. I will argue that capability and safety are not a trade-off; they should go hand in hand: Capability without safety is risky, and safety without capability is irrelevant. To enable safe, secure, and truly wide adoption of AI agents that benefit humanity, we need to evaluate the risks rigorously and systematically in realistic, sandboxed environments, and build guardrails to enhance safety and security. I will focus on computer-use agents and discuss our recent efforts along these dimensions. The ultimate goal is to build agents that are both capable and safe for deployment.

November 14

Speaker: Gang Li (University of California Los Angeles)

Title: Prediction Performance Measures for Time-to-Event Data

Poster: 1114-Li-poster

Abstract: Evaluating and validating the performance of prediction models is a crucial task in statistics, machine learning, and their diverse applications, including precision medicine. However, developing robust prediction performance measures, particularly for time-to-event data, poses unique challenges. In this talk, I will highlight how conventional performance metrics for time-to-event data—such as the C-index, Brier Score, and time-dependent AUC—may yield undesirable results when comparing prediction models or algorithms. I will then introduce a novel time-dependent pseudo R-squared measure and demonstrate its utility as a prediction performance metric for uncensored and right-censored time-to-event data. Additionally, I will discuss its extension to competing risks scenarios and to popular epidemiologic designs, including case-cohort and nested case-control designs. Its effectiveness will be demonstrated through simulations and real-world examples.

THURSDAY, November 20 – DeMets Lecture – Mark Gerstein (Yale)

IN PERSON ONLY!
NOTE DIFFERENT TIME AND LOCATION
12:30 PM
Morgridge Hall Seminar Room, 7th Floor

Speaker: Professor Mark Gerstein, Yale University

Title: AI Methods in Biomedicine

Poster: DeMets_LecturePoster2025-final

Abstract: This talk surveys a spectrum of AI approaches for biomedicine—from classical methods to current deep-learning techniques and emerging extensions involving agents and quantum computing. It focuses on modeling gene-regulatory and cell-communication networks to interpret brain disease, describing how cell-type–specific networks can be constructed from single-cell QTLs and co-expression data and then embedded into deep-learning frameworks to predict disease from genotype, prioritize pathways and genes, and model perturbations for drug-target discovery. The talk then highlights new LLM applications, including automatic code generation and benchmarking (Biocoder), collaborative multi-expert clinical reasoning (MedAgents), and end-to-end single-cell analysis workflows (CellForge). Finally, it addresses genomic-privacy challenges—first through classical approaches such as homomorphic encryption and hidden Markov models, and then through a quantum-computing framework that enables distributed genomic analysis without exposing individual-level data.

November 28 – Thanksgiving – no seminar

December 5

Speaker: Yuhang Zhao (University of Wisconsin-Madison)

Title: AI-assisted Vision: Context-Aware Systems to Empower People with Low Vision

Poster: 1205-Zhao-poster.docx

Abstract:

People with disabilities are marginalized by inaccessible social infrastructure and technology, facing various challenges in all aspects of their life. Conventional assistive technologies commonly provide generic solutions to a certain disability population and do not consider users’ individual and context differences, leading to high abandonment rate. My research seeks to thoroughly understand the experiences and needs of people with disabilities and create intelligent assistive technologies that are adaptive to user contexts, such as their environments and intents, providing tailored, unobtrusive support.

In this talk, I will focus on people with low vision, who have visual impairments but are not blind. I will discuss how I leverage state-of-the-art AI, augmented reality (AR), and eye-tracking technologies to design and develop context-aware systems to enhance low vision people’s visual perceptions in activities of daily living. Specifically, I divide user context into external factors (e.g., surrounding environments) and internal factors (e.g., behaviors, intents). To capture external context, I create scene-aware systems that recognize essential visual contents around the users via egocentric scene interpretation and render suitable augmentations for low vision. For example, CookAR is a wearable AR system that distinguishes and augments the affordance of kitchen tools (e.g., knife blade vs. knife handle) to facilitate safe and efficient interactions. In terms of internal context, I leverage eye-tracking techniques to understand low vision people’s unique gaze patterns and develop intent-aware systems to recognize their visual challenges and intents and provide adaptive assistance. One example is our GazePrompt system that generates gaze-aware augmentations to support reading tasks, such as highlighting the next line when a user is switching line, or verbally reading aloud a word when the user hesitates around that word. I will conclude my talk by highlighting future research directions towards AI-assisted vision for people with low vision.

December 12 – BDS Student Presentations

DIFFERENT LOCATION: The Hello World Auditorium – Morgridge Hall room 1570.

Speaker: Parth Khatri, BDS PhD Student
PhD Mentor: Huy Dinh, PhD
Title: Autocorrelation-aware resampling improves spatial transcriptomics cell-cell interaction analysis.

Speaker: Abrar Majeedi, BDS PhD Student
PhD Mentor: Yin Li, PhD
Title: Deep learning on bedside continuous signals for enhanced health monitoring in neonatal intensive care units (NICUs)

Speaker: Jie Sheng, BDS PhD Student
PhD Mentor: Daifeng Wang, PhD
Title: A variational autoencoder with transport operators to disentangle cellular gene expression dynamics of co-occuring biological processes in time and space.