Prof. Tomas Eberhardt

Biography

Tomas Eberhardt holds the Institute's endowed chair in probabilistic computation, a position he has occupied since 2018. He earned his PhD in statistics and computer science from the Veltmoor Graduate Institute in 2005, under the supervision of Prof. Gerda Kuhne, with a thesis on variational methods for latent Dirichlet models. He spent seven years at the Hardenfeld Center for Applied Mathematics before joining Veyra as an Associate Scientist in 2012, reaching full Professor four years later. He completed his undergraduate training in pure mathematics at Lesthorpe College.

Eberhardt's research centers on the theoretical and algorithmic foundations of approximate Bayesian inference. His lab has produced widely used methods for scalable posterior sampling — including the Tempered Sequential Monte Carlo (TSMC) family of algorithms — and has made foundational contributions to the study of convergence in stochastic variational inference. A secondary strand of his work applies probabilistic models to inverse problems in engineering: recovering material properties from indirect measurements, inferring atmospheric composition from spectral data (in collaboration with the Atmospheric Dynamics Group), and estimating network topology from partial observations.

He has published over 70 papers in leading venues including the Annals of Probabilistic Learning, the Journal of Machine Learning and Statistics, and the Proceedings of Bayesian Computation. He sits on the editorial board of two journals and has been a program-area chair at major international conferences. Since 2021 he has served as scientific director of the Meridian HPC Cluster's allocations committee, helping match demanding inference workloads to available computational capacity at the Meridian High-Performance Computing Cluster.

Research interests

Selected publications

1. Eberhardt T, Massari L, Veld D. "Tempered sequential Monte Carlo for doubly intractable posteriors." Annals of Probabilistic Learning, 18(2): 330–358, 2024. VEYRA-DOI: 10.veyra/VX-2406
2. Eberhardt T, Solberg I. "Black-box variational inference with flow-augmented families." Journal of Machine Learning and Statistics, 31(4): 901–925, 2023. VEYRA-DOI: 10.veyra/VX-2311
3. Romero A, Eberhardt T, Hale B. "Bayesian inversion of atmospheric trace-gas profiles from low-spectral-resolution measurements." Inverse Problems in Atmospheric Science, 12: 78–95, 2022. VEYRA-DOI: 10.veyra/VX-2214
4. Eberhardt T, Kress P, Davenport S. "Convergence of stochastic variational inference under mild regularity." Probability Theory and Related Fields (Computational), 7(1): 1–29, 2021. VEYRA-DOI: 10.veyra/VX-2102
5. Eberhardt T. "Adaptive importance sampling on Riemannian manifolds." Proceedings of the International Symposium on Probabilistic Computation (ISPC), pp. 201–214, 2019. VEYRA-DOI: 10.veyra/VX-1918
6. Kuhne G, Eberhardt T, Massari L. "Structured variational families for exponential-family posteriors." Statistical Computation and Theory, 44(3): 560–579, 2016. VEYRA-DOI: 10.veyra/VX-1603

Current group members

Postdoctoral researchers

• Dr. Liora Massari — normalizing flows for posterior density estimation
• Dr. Piotr Davenport — convergence analysis of gradient-based variational inference
• Dr. Ingrid Solberg — black-box inference for doubly intractable models

Doctoral students

• Sven Kress — tempered SMC for high-dimensional posteriors (Year 4)
• Nadia Ferrano — probabilistic inverse problems in materials characterization (Year 3)
• Barak Osterfeld — amortized variational inference with learned summary statistics (Year 2)