Gießen 2024 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
HK: Fachverband Physik der Hadronen und Kerne
HK 1: Computing I
HK 1.6: Vortrag
Montag, 11. März 2024, 18:00–18:15, HBR 14: HS 1
Multi-particle Reconstruction in Detector Data using object-centric Machine Learning — •Sara Aumiller1, Nicole Hartman1, Lukas Heinrich1, Florian Kaspar1, Karina-Sânziana Stelea1, Stefan Wallner1, Dominik Ecker1, Luise Meyer-Hetling1, Andrii Maltsev1, and Thomas Pöschl2 — 1Technical University of Munich, Germany — 2CERN, Geneve, Switzerland
High-energy physics experiments require high performance on separation and reconstruction of multi-particle signals in detector data. This task gets especially challenging if signals of multiple, quasi-simultaneous particles overlap. Such events occur in various forms like calorimetric clusters, Cherenkov light rings or track patterns. Traditional reconstruction methods of particle-detector data are regularly pushed to their limits when confronted with this challenge as they require specific development for each task and get computationally expensive.
In this talk, I will explore a universal approach of multi-particle reconstruction using object-centric machine learning. This includes state-of-the-art artificial-neural-network methods like Invariant Slot Attention and Variational Autoencoding which are applied on simulated, particle-detector data. The research holds potential for future application in experiments like COMPASS or AMBER at CERN.
Keywords: Machine Learning; Neural Networks; Deep Learning