Göttingen 2025 – wissenschaftliches Programm

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

T: Fachverband Teilchenphysik

T 33: Data, AI, Computing, Electronics III (ML in Jet Tagging, Misc.)

T 33.4: Vortrag

Dienstag, 1. April 2025, 17:00–17:15, VG 2.101

Domain adaptation in the context of flavour tagging at the LHCb experiment — Johannes Albrecht^{1, 2}, Mirko Bunse², and •Quentin Führing^{1, 2} — ¹TU Dortmund University, Dortmund, Germany — ²Lamarr Institute for Machine Learning and Artificial Intelligence, Dortmund, Germany

Decay-time-dependent measurements of oscillating neutral B mesons at LHCb require information of the B-meson flavour at the time of its production. This information cannot be inferred from the decay products used for the reconstruction of signal candidates. Instead, multivariate algorithms are used to estimate the production flavour of B mesons, which exploit a variety of particles produced in association with the signal in the proton-proton interaction.

Simulation is often used to provide a labelled data sample for the training of these algorithms. However, known differences between simulation and recorded data are present, particularly in quantities significantly impacting the flavour tagging performance, such as the track multiplicity in fragmentation processes. As a consequence, the algorithms do not reach the same level of performance in data as in simulation.

We approach this mismatch between data and simulation with machine-learning techniques from the realm of domain adaptation. These methods prevent the multivariate algorithms from learning an implicit and undesired distinction between data and simulations. As a result, we expect improved performance on data. In this presentation, the idea and the status of the ongoing project is presented.

Keywords: Flavour Tagging; LHCb; Domain Adaptation; Neural Network; Flavour Physics