Karlsruhe 2024 – scientific programme

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

T: Fachverband Teilchenphysik

T 36: Gamma astronomy 2

T 36.8: Talk

Tuesday, March 5, 2024, 17:45–18:00, Geb. 30.22: kl. HS A

Ultra-Fast Generation of Air Shower Images for Imaging Air Cherenkov Telescopes using Generative Models — •Christian Elflein, Jonas Glombitza, and Stefan Funk for the H.E.S.S. collaboration — ECAP, Erlangen, Germany

Resource-aware simulation is an important aspect in many fields of modern physics, including astroparticle physics. The continuous progress in machine learning in this day and age and the successful application of generative models to various tasks in particle and astroparticle physics motivate our application of generative models to the simulation of air shower images in gamma astronomy.

In this contribution, we present a novel technique for the fast generation of gamma-ray air shower images from the FlashCam camera of the CT5 telescope, which is part of the High Energy Stereoscopic System (H.E.S.S.). The generative model used for the generation of images with more than 1500 pixels is based on a Wasserstein Generative Adversarial Network (WGAN) and trained using H.E.S.S. simulations. We show that our framework, in comparison to the standard simulation method, speeds up the image generation by up to five orders of magnitude while keeping a competitive image quality. The visual similarity to simulated images and the representation of physical properties in the generated images is verified by analysing several low-level and high-level parameters and their correlations.

Following this work, we additionally investigate diffusion models, which are state-of-the-art generative deep-learning models, as an alternative way to generate air shower images.

Keywords: IACT; Cherenkov Telescope; Air Shower Image; Deep Learning; Generative Model