Publication Highlights

For a complete list of publications, please refer to my Google Scholar profile.

Preprint Can Microcanonical Langevin Dynamics Leverage Mini-Batch Gradient Noise?

Emanuel Sommer*, Kangning Diao*, Jakob Robnik, Uros Seljak, David Rügamer

Paper

Bayesian Deep Learning Microcanonical Langevin Scalability

AISTATS 2026 (spotlight) On the Interplay of Priors and Overparametrization in Bayesian Neural Network Posteriors

Julius Kobialka*, Emanuel Sommer*, Chris Kolb, Juntae Kwon, Daniel Dold, David Rügamer

Paper | Conference

Bayesian Deep Learning Sampling-based Inference BNN Priors Overparameterization

ICLR 2025 Microcanonical Langevin Ensembles: Advancing the Sampling of Bayesian Neural Networks

Emanuel Sommer, Jakob Robnik, Giorgi Nozadze, Uros Seljak, David Rügamer

Paper | Conference

Bayesian Deep Learning Sampling-based Inference MCMC UQ

ICML 2024 Connecting the Dots: Is Mode-Connectedness the Key to Feasible Sample-Based Inference in Bayesian Neural Networks?

Emanuel Sommer*, Lisa Wimmer*, Theodore Papamarkou, Ludwig Bothmann, Bernd Bischl, David Rügamer

Paper | Conference

BNNs Mode-Connectedness Sampling-based Inference Symmetries

Econometrics and Statistics Vine Copula based Portfolio Level Conditional Risk Measure Forecasting

Emanuel Sommer, Karoline Bax, Claudia Czado

Paper | Journal

Dependence Modeling Vine Copulas Portfolio Risk Stress Testing Time Series

* Equal contribution

---

Software

{portvine} Package

Portfolio-level unconditional as well as conditional risk measure estimation (VaR/Expected Shortfall) using Vine Copulas and ARMA-GARCH models. Designed for backtesting and stress testing.

CRAN | GitHub | Docs

{bde} Package

Bayesian Deep Ensembles (specifically MILE) implementation fully compatible with scikit-learn (part of scikit-learn-contrib) and focused on tabular data. Leverages JAX for accelerator-backed training and inference.

GitHub | Docs