Parts | Days | Selection | Search | Updates | Downloads | Help
T: Fachverband Teilchenphysik
T 26: Data Analysis, Information Technology and Artificial Intelligence
T 26.3: Talk
Monday, March 21, 2022, 16:45–17:00, T-H39
Charm tagger shape calibration for BDT-based signal-background discrimination — •Spandan Mondal1, Xavier Coubez1,2, Alena Dodonova1, Ming-Yan Lee1, Luca Mastrolorenzo1, Andrzej Novak1, Andrey Pozdnyakov1, Alexander Schmidt1, and Annika Stein1 — 1RWTH Aachen University — 2Brown University, USA
Identification of charm-quark-initiated jets at the LHC is especially challenging. Usage of deep learning based algorithms have enabled several CMS analyses to efficiently discriminate charm jets simultaneously from bottom and light jets. The charm probability scores yielded by such charm tagging algorithms can play a powerful role when used as inputs to a machine learning based algorithm for discrimination between signal and background. However, as jet identification algorithms are trained strictly on simulated jets, a direct usage of charm tagger output values requires calibrating the entire output probability distributions using real jets reconstructed from CMS data. This talk focuses on the calibration of the output discriminator values of charm-tagging algorithms using flavour-enriched selections of jets in data. Additionally, the improvement resulting from a shape calibration approach, over the traditional approach of calibrating efficiencies at fixed c-tagger working points, is exemplified in the context of the resolved VHcc analysis.