Research

Theory of Tractable Inference. What can be efficiently determined given a model of the world? Towards this question, I'm studying the inherent tradeoff in models between expressiveness and inference tractability. For example, for the fundamental inference task of marginalization, we've shown how a common tractable modelling language captures a myriad of other models (e.g., tractable graphical models, sum-product networks, probabilistic generating circuits) up to polynomial time transformations [5,6].

Statistical Election Audits. Risk-limiting audits (RLAs) are rigorous statistical procedures used to detect errors in election results. I've conducted experiments [1] to understand the behavior of existing RLAs and developed a new statistical test, PROVIDENCE [3,4], the most efficient and secure ballot polling RLA known today. PROVIDENCE has been piloted and is implemented in the most popular election audit software in the US, open source Arlo.

Publications

Also see my Google Scholar page.

• Anji Liu, Oliver Broadrick, Mathias Niepert, Guy Van den Broeck. Discrete Copula Diffusion. Under review, 2024. (pdf, arXiv)
• Oliver Broadrick, William Cao, Benjie Wang, Martin Trapp, Guy Van den Broeck. Probabilistic Circuits for Cumulative Distribution Functions. TPM Workshop at UAI 2024. (pdf, arXiv)
• Oliver Broadrick, Honghua Zhang, Guy Van den Broeck. Polynomial Semantics of Tractable Probabilistic Circuits. UAI 2024, Oral Presentation. (pdf, arXiv, talk, slides)
• Oliver Broadrick. Risk-Limiting Audit PROVIDENCE and Round Size Considerations. MS Thesis, The George Washington University, 2023. (pdf)
• Oliver Broadrick, Poorvi Vora, and Filip Zagórski. PROVIDENCE: a Flexible Round-by-Round Risk-Limiting Audit. USENIX Security 2023. (pdf, usenix, talk, slides, earlier arXiv version)
• Hesham Fouad, Oliver Broadrick, Benjamin Harvey, Charles Peeke, and Bhagirath Narahari. Real-Time AI: Using AI on the Tactical Edge. Elsevier book chapter in "Putting AI in the Critical Loop: Assured Trust and Autonomy in Human-Machine Teams", invited chapter based on presentation at 2022 AAAI Spring Symposium. (link)
• Oliver Broadrick, Sarah Morin, Grant McClearn, Neal McBurnett, Poorvi L. Vora, and Filip Zagórski. Simulations of Ballot Polling Risk-Limiting Audits. Seventh Workshop on Advances in Secure Electronic Voting, in association with Financial Cryptography 2022. (pdf, talk, slides)

Research Talks

• May 2024, "Polynomial Semantics of Probabilistic Circuits." Invited talk, SNAIL Seminar, University of São Paulo. (slides)
• August 2023, "PROVIDENCE: a Flexible Round-by-Round Risk-Limiting Audit". USENIX Security. (talk, slides)

Teaching

I love teaching. Since high school I have tutored for hundreds of hours at the high school and college level in mathematics, computer science, and physics. At GWU, I was a teaching assistant in Discrete Mathematics and Theory of Computation courses for several semesters.

Please give me anonymous feedback on my teaching.