Göttingen 2025 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

T: Fachverband Teilchenphysik

T 76: Data, AI, Computing, Electronics VII (Generative AI, MC Generators)

T 76.1: Talk

Thursday, April 3, 2025, 16:15–16:30, VG 2.101

Correcting the mis-modeling of photon energy deposits in the calorimeter using normalizing flows and flow matching — Caio Daumann, Johannes Erdmann, and •Lars Schiffeler — III. Physikalisches Institut A, RWTH Aachen University

Simulated events are key ingredients for almost all high-energy physics analyses. However, imperfections in their configuration can result in mis-modelling and discrepancies between the data and simulations. Normalizing flows are used in CMS to correct the high-level inputs to the photon identification algorithms, which have a low dimensionality. Improved identification algorithms, on the other hand, use information with an increased dimensionality, such as individual energy deposits in a calorimeter. This poses a challenge to normalizing flows, as they are more effective in lower-dimensional spaces. We investigate the influence of this increase in dimensionality on normalizing flows and compare their effectiveness to flow matching. To study these effects, simulations of a CMS-inspired toy calorimeter are used.

Keywords: high energy physics; photons; normalizing flows; flow matching