I will give an oral presentation, “Multiple latent preference orderings in language models,” in a contributed session at CCS 2026.
About
I am a PhD student in Complex Systems and Data Science at the University of Vermont. I work in the Joint Lab with Jean-Gabriel Young and Laurent Hébert-Dufresne.
Before coming to Vermont, I studied liberal arts at St. John’s College, worked at MITRE, and worked on the Coherent CAPTAIN-Mills dark matter experiment at Los Alamos National Laboratory.
In summer 2026, I was a Data Science for Social Good fellow at Johns Hopkins, working with the City of Baton Rouge on early warning models for blight and structural deterioration.
Research
I study how to infer the rules of a networked system from what it does. This includes reconstructing networks and contagion processes from observations, improving inference on graphs with loops, and exploring methods for Bayesian learning that avoid sampling. I also study how language models make choices when their preferences are inconsistent.
My work brings together network science, Bayesian inference, and machine learning. A full list of papers is on the publications page.
Announcements
Our preprint “Multiple latent orderings better predict language model preferences”, with Aviral Chawla and Jean-Gabriel Young, is out.
New CCM collaboration preprint: “Axion-Like Particle Search with a Hybrid Cherenkov-Scintillation Detector”.
This summer I was a fellow in the Data Science for Social Good program at Johns Hopkins, working with the City of Baton Rouge on early warning models for blight and structural deterioration. Grateful to the GovEx team and my project team for a great 10 weeks. Video from the showcase.
We gave a parallel talk at IC2S2 2026 in Burlington: “The Parliament of Thoughts in Language Models,” with Aviral Chawla and Jean-Gabriel Young.
New preprint: “Emergent contagion complexity: Disentangling mechanistic complexity from correlated heterogeneity”, with Katerina Tang, Daniel Kaiser, Jean-Gabriel Young, Laurent Hébert-Dufresne, and Nicholas Landry.
I received a $5,000 Complexity Award Grant from the Vermont Complex Systems Institute to model strategic lying in Kenyan pastoralists with Eleonora Zanetti. The award supported participant compensation and travel for her June–July fieldwork in Kenya.
New preprint: “Simpson’s paradox explains the ubiquity of nonlinear, threshold, and complex contagions”, with Laurent Hébert-Dufresne, Antoine Allard, Jean-Gabriel Young, and Guillaume St-Onge.
I gave a talk on inferring interaction kernels in stochastic opinion dynamics models at the AMS Spring Eastern Sectional Meeting in Boston.
Our paper, “Sensitivity analysis of epidemic forecasting and spreading on networks with probability generating functions”, appeared in Journal of the Royal Society Interface.
Excited to have been invited to attend the Bridging the Interdisciplinary Gap in the Mathematics of Modeling Social Phenomena workshop at BIRS!
Had a truly life-changing experience participating in the SFI Complex Systems Summer School.
I joined Los Alamos National Laboratory as a graduate research intern, working with Andrey Lokhov on graphical models.
I presented a poster at Dynamics Days 2025 in Denver, CO.
The poster, “Sensitivity of Epidemic Forecasts with Statistical Condition Estimation for Probability Generating Functions,” is available here:
Poster PDF.
Excited to announce our paper, led by Nicholas Landry, Reconstructing networks from simple and complex contagions in Physical Review E. Check out the code and data at GitHub
I used the method of characteristics to solve a higher order voter model. Check out the slides: Slides
I presented (with co-authors) “Inferring Interaction Kernels in Stochastic Opinion Dynamics Models” at NetSci 2024 in Québec City, QC.
Slides are now available.
I presented (with T. Gebhart, L. Huynh, V. Modisette, M. Tian, A. Wiedemann, P. Chodrow, and H. Z. Brooks) “Inferring Interaction Kernels for Stochastic Agent-Based Opinion Dynamics” at the Joint Mathematics Meetings in San Francisco, CA.
I spoke at Dynamics Days 2024 in Davis, CA.
My presentation was “Understanding Polarization in the Higher-Order Non-Linear Voter Model.”
Slides are posted.
I presented at the Dartmouth Applied Math Colloquium.
My talk was titled “The Emergence of Inequality with Two Models of Social Dynamics.”
Slides are now available.
I gave an oral presentation at the 18th International Pragmatics Conference in Brussels.
The talk was “‘Her behavior was disgusting’: Emotion and Morality in Online Complaint Discourse.”
You can see a recording here: YouTube.
Selected publications
- Multiple latent orderings better predict language model preferences. A. Chawla, W. H. W. Thompson, and J.-G. Young. Preprint, 2026.
- Emergent contagion complexity: Disentangling mechanistic complexity from correlated heterogeneity. K. Tang, D. Kaiser, W. Thompson, J.-G. Young, L. Hébert-Dufresne, and N. W. Landry. Preprint, 2026.
- Simpson's paradox explains the ubiquity of nonlinear, threshold, and complex contagions. L. Hébert-Dufresne, A. Allard, J.-G. Young, W. H. W. Thompson, and G. St-Onge. Preprint, 2026.
- Sensitivity analysis of epidemic forecasting and spreading on networks with probability generating functions. M. C. Boudreau, W. H. W. Thompson, C. M. Danforth, J.-G. Young, and L. Hébert-Dufresne. Journal of the Royal Society Interface, 2026.
Selected talks
- “Multiple latent preference orderings in language models,” oral presentation at CCS 2026.
- We gave “The Parliament of Thoughts in Language Models” as a parallel talk at IC2S2 2026, with Aviral Chawla and Jean-Gabriel Young.
- “Inferring Interaction Kernels In Stochastic Opinion Dynamics Models,” AMS Spring Eastern Sectional Meeting, Boston, March 2026.
- “Sensitivity of epidemic forecasts with statistical condition estimation,” poster at Dynamics Days, 2025.
Contact
Email is the best way to get in touch with me: wthomps3 (at) uvm (dot) edu.