Thursday, May 28, 2026, 16:00
WBGB/019
Anja Butter, Heidelberg University & LPNHE, France
Abstract:
Modern particle physics experiments produce enormous amounts of data, and
understanding them requires complex and computationally expensive precision
simulations. Neural networks are transforming the way we simulate and analyze
this data by making many intermediate steps faster, more efficient, and even
invertible.
Neural networks can be used throughout the full simulation chain: from the
interpolation of interaction cross sections, via the generation of artificial
particle events, to modeling the detector response, machine learning enables
high-precision predictions. In the inverse direction, neural networks can
reconstruct the underlying physical processes from the measured high-dimensional
data. Along the way, we focus on the question how we can incorporate physics
knowledge — such as exact and approximate symmetries — into network
architectures and training objectives, alongside the precise quantification of
uncertainties to ensure physically meaningful predictions.