Heidelberg 2022 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

T: Fachverband Teilchenphysik

T 107: Data Analysis, Information Technology and Artificial Intelligence 5

T 107.8: Vortrag

Donnerstag, 24. März 2022, 18:00–18:15, T-H39

A super-resolution GAN for photon identification at collider experiments — Johannes Erdmann, Florian Mentzel, Olaf Nackenhorst, and •Aaron van der Graaf — TU Dortmund University, Department of Physics

Many processes in proton-proton collisions contain prompt photons in the final state, ranging from Standard Model (SM) measurements, such as H →   γ γ, to searches for physics beyond the SM. In order to measure these processes with a high precision, a good photon identification efficiency while retaining a high background rejection rate is important. Most misidentified photons arise from hadron decays, such as π⁰ →   γ γ, which could possibly be rejected better with a higher calorimeter granularity. This motivates the idea of artificially increasing the calorimeter granularity by training a super-resolution Generative Adversarial Network (SRGAN) with simulated low and high resolution calorimeter images. As a proof of concept, mono-energetic photons and neutral pions are simulated in an electromagnetic calorimeter and only the second calorimeter layer is used for the SRGAN training. In this presentation, the used SRGAN model, the results of the SRGAN training and the predicted super-resolution images are presented. It is shown that the predicted super-resolution images contain additional information that increases the pion rejection rate compared to the low resolution images.