Transforming Flavour Tagging at the ATLAS Experiment
by
Physics West Lecture Theatre (PW117)
Poynting
Identifying the flavour of jets produced in proton–proton collisions is crucial for many LHC measurements and searches, from precision Higgs studies to the hunt for new heavy particles. The ATLAS flavour tagging group has recently developed a new generation of machine-learning–based taggers that greatly enhance the identification of jets containing b- and c-hadrons. The latest of these GN2, represents a significant advance in both performance and architecture. GN2 applies a transformer architecture to the charged-particle tracks within each jet, allowing it to capture both local and global correlations in the jet’s internal structure. In this seminar I will review the evolution of ATLAS flavour tagging leading up to GN2, outline the key ideas behind its design and training, and present its measured performance on ATLAS data compared to earlier deep-learning taggers.