Joint Statistical Meeting 2025
Join Pacific Northwest National Laboratory at the Joint Statistical Meetings in Nashville, TN
Illustration by Melanie Hess-Robinson | Pacific Northwest National Laboratory
Music City Center
One of the largest statistical events in the world, the Joint Statistical Meetings (JSM) will be held August 2–7, 2025 in Nashville, Tennessee and will feature over 600+ sessions in the areas of statistics, data science, and enriching society through AI. The event anticipates more than 5,000 attendees from around the world, including several researchers and mathematicians from Pacific Northwest National Laboratory (PNNL).
Supporting JSM on its program committee is Karl Pazdernik, senior data scientist at PNNL and invited speaker. This will be Pazdernik’s third year on the organizing committee, and this year he was elected to serve as program chair. In this role, Pazdernik has had a significant hand in shaping the sessions related to text analysis (i.e., large language models) through his selection of three topic-contributed paper sessions.
Visit PNNL at the EXPO
PNNL will also be well represented by researchers, managers, and recruiter Sarah Russel at Booth 117, August 3–6. This is a great chance to learn more about PNNL’s culture, research priorities, and what opportunities might exist for you at the Laboratory.
Booth Schedule
- Sunday, August 3: 1:00–6:00 p.m.
- Monday, August 4: 9:00 a.m.–5:30 p.m.
- Tuesday, August 5: 9:00 a.m.–5:30 p.m.
- Wednesday, August 6: 9:00 a.m.–2:30 p.m.
Featured PNNL Presentations
Sunday, August 3
Auditing the Performance and Calibration of Multi-Modal Large Language Models
Session: Invited Paper Session: Statistics for Large Language Models and Large Language Models for Statistics, 2:05–2:25 p.m.
PNNL Speaker: Brendan Kennedy
Multimodal large language models (MLLMs) are changing the way experts interact with diverse data sources, including images, graphs, multimedia, and structured data. Despite achieving high-accuracy scores on visual multiple-choice question answering tasks, further evaluation is needed to measure their readiness in critical areas like medicine and scientific research. This session will delve into two conducted analyses measuring the robustness and calibration underlying the performance of MLLMs on image QA benchmarks. Learn more.
Improved Bayesian Graphical Models
Session: Contributed Papers: Latest Research in Genomics and Microbiome with a Hint of Bayesian, 5:20–5:35 p.m.
PNNL Authors: Lisa Bramer (presenting), David Degnan, Erik VonKaenel, Moses Obiri
The study of protein–protein interactions (PPIs) provides critical insights into biological mechanisms, including antibody–antigen binding and enzyme regulation. While recent advancements in PPI research have driven significant breakthroughs, the usability of tools like graphical models remains constrained by computational limitations, particularly in the context of large, multiclass datasets. This session will explore a clustering-focused iterative methodology designed to improve both the scalability and accuracy of model estimation in high-dimensional spaces, with applications in Bayesian graphical modeling. Learn more.
Monday, August 4
Building a “Model in a Month” for Science and Defense Applications
Session: Invited Paper Session: Generative AI and Foundation Models in Defense Applications, 10:55–11:15 a.m.
PNNL Speaker: Karl Pazdernik
AI has long been a core modeling technique, but there is a recent shift toward training foundation models. Unlike narrow AI, which is designed for specific tasks, foundation models can tackle a range of applications and be re-trained or fine-tuned to enhance performance. This session will explore the development of unimodal and multimodal large language models (LLMs) for scientific and defense applications, strategies for training with limited compute, challenges in alignment, approaches to incorporating statistics into LLMs, and ensuring that outputs are both accessible and trustworthy. Learn more.
Enhancement in Coated Atmospheric Tar Balls Using Mie Theory
Session: SPAAC Poster Competition – Topic-Contributed Poster Presentations, 2:00–3:50 p.m.
PNNL Authors: Karen Magana (Intern, Presenting), Zezhen Cheng, Manish Shrivastava
Atmospheric tar balls (TBs) are solid, light-absorbing organic particles from wildfires that have the capability of disrupting Earth’s energy balance through the absorption of solar radiation. When TBs are coated with substances like water or organic matter, their optical properties will be enhanced—creating a variable optical property not currently captured in climate models.
This session will examine how Mie calculations were used to explore light absorption enhancement of different coatings on TBs. By testing multiple properties, the researchers found that clear and brown coatings enhanced light absorption through a “lensing effect,” with brown coatings showing the largest impact. Learn more.
Tuesday, August 5
Embracing the Challenges: Skills for Statisticians That Will Translate in the AI Era
Session: Test Analysis P.M. Roundtable Discussion, 12:30–1:50 p.m.
PNNL Speaker: Karl Pazdernik
AI is rapidly reshaping the landscape of statistical analysis, underscoring the need for statisticians to adapt and evolve. This roundtable discussion will explore how core statistical skill sets can be leveraged in the era of AI and highlight the steps statisticians must take to move past traditional boundaries to remain relevant and impactful in their work. Participants will gain valuable insights into blending conventional statistical techniques and cutting-edge methodologies. Learn more.