SMuK 2023 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
T: Fachverband Teilchenphysik
T 86: ML Methods IV
T 86.2: Vortrag
Mittwoch, 22. März 2023, 17:45–18:00, HSZ/0405
Development of novel machine learning algorithms for robust jet flavour classification for Run3 at CMS — •Annika Stein1, Judith Bennertz1, Xavier Coubez1,2, Alexander Jung1, Summer Kassem1, Ming-Yan Lee1, Spandan Mondal1, Alexandre de Moor3, Andrzej Novak1, Alexander Schmidt1, and Hendrik Schönen1 — 1III. Physikalisches Institut A, RWTH Aachen University, Aachen, Germany — 2Brown University, Providence, USA — 3Vrije Universiteit Brussel, Brussels, Belgium
Complex neural network architectures have been developed for jet tagging and play a crucial role for numerous analyses relying on this classification task. Recent advances exploit low-level information with convolutional layers, graph neural networks, or transformer models with attention mechanisms. While improving performance is one of the key components in tagger development, the capability to generalize to detector data imposes new challenges and can be probed through comparisons between the two domains, simulation and data, in different phase spaces. This talk will showcase how strategies like adversarial training can be used to improve robustness and data/MC agreement for state-of-the-art tagging algorithms. An overview of the upcoming generation of flavour tagging algorithms for Run3 will be given.