ML bridges for particle physics

Breaking things is one of the most efficient — and most entertaining — ways to learn something new. Luckily for us, machine learning is currently breaking down many long‑standing boundaries in particle physics, opening the door to new, ML‑based approaches for theory‑inspired and ML‑supported data analysis.

Connecting Paris and Heidelberg, particle physics and computing, theory and experiment, my group and I work at the intersection of our favorite topics to learn more about particle physics and to develop increasingly powerful methods for analyzing LHC data. Over the past years, we have built a strong collaboration between LPNHE and ITP, leading to numerous joint research projects, workshops, and exchange programs.

Neural networks for precision predictions

Precision predictions are essential for the success of the LHC physics program. To overcome current bottlenecks in high‑precision calculations, we develop neural‑network‑based methods that improve the efficiency and accuracy of various steps in the simulation chain. Our work includes:

• Precise estimation of scattering amplitudes
• Optimized phase‑space sampling
• Data‑driven hadronization models
• Fast and accurate detector simulations

We employ a broad range of machine‑learning techniques, with a particular focus on generative models — from normalizing flows to diffusion models — including new approaches for reliable uncertainty estimation.

Inverse problems or how to unfold the LHC

In an ideal world, we would have unbiased and precise methods to constrain the fundamental parameters of the Standard Model and beyond‑the‑Standard‑Model theories using the full, high‑dimensional LHC dataset. ML‑based unfolding takes a first step toward this goal.

We develop novel methods that predict probability distributions over likely parton‑level or particle‑level configurations given a measured detector‑level event. This enables a significantly more direct interpretation of unfolded data and allows constraints on Lagrangian parameters to be derived in a transparent and statistically robust way.

If you wanna discuss the latest ML ideas for better physics or just enjoy the view on my favorite cities, you can find me either at

or at

Institut für Theoretische Physik
Universität Heidelberg
Philosophenweg 16
69120 Heidelberg, Germany

• Email: butter@thphys.uni-heidelberg.de
• Phone: +49 6221 54-9406