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

T: Fachverband Teilchenphysik

T 74: Flavour Physics V

T 74.2: Vortrag

Donnerstag, 3. April 2025, 16:30–16:45, VG 1.104

Transformer Model for Flavour Tagging at Belle II — Ariane Frey, Lukas Herzberg, Thibaud Humair, Benjamin Schwenker, and •Tilo Wettlaufer — II. Physikalisches Institut, Universität Göttingen

At the Belle II experiment, an entangled state of two B mesons is measured. In order to perform analyses on CP violation, one B meson is fully reconstructed (signal side). Due to entanglement, reconstructing the flavour of the other B meson (tag side) determines the flavour of the signal side. This process of flavour tagging uses the characteristics of flavour specific decay products and correlates them to the tag side flavour, without an exclusive full reconstruction. Currently, both a category based algorithm is used, which reports an effective tagging efficiency of: Q_cat. = 30.0(13) %, as well as a graph neural network reporting: Q_GNN = 37.4(6) %. An alternative deep neural network uses raw track attributes to construct numerical representations of tracks, giving: Q_DNN = 29.3(16) % for neutral B mesons.

A new flavour tagging algorithm is presented based on the model of tabular transformers. This model uses contextual embeddings which are easier to interpret, as well as provide greater robustness against noisy or missing data compared to only using multi-layer perceptrons (MLP).

Keywords: Belle II; B Physics; Flavour Tagging; Machine Learning; Transformer