Transforming jet flavour tagging at the ATLAS and CMS experiments

by Philipp Gadow (Universität Hamburg)

Tuesday 30 Jun 2026, 16:00 → 17:00 Europe/Berlin

ENC-D308

The discovery of the Higgs boson in 2012 at the Large Hadron Collider (LHC) by the ATLAS and CMS experiments opened the door to investigating the least understood part of the Standard Model of particle physics. The unprecedented Run 3 dataset taken between 2022 and 2026 provides a unique opportunity to probe the structure of the Higgs potential. The most probable decay mode of Higgs bosons is into a pair of b-quarks, which hadronise into collimated sprays of particles ("jets"). Distinguishing jets initiated by b-quarks from all other jets ("flavour tagging") is therefore of greatest importance.

My talk will discuss the state-of-the-art flavour tagging algorithms of the ATLAS and CMS collaborations. Both rely on the transformer architecture, exploiting measurements in the tracking detector to classify jets based on the flavour of quarks that initiated them. The new generation of algorithms starkly contrasts with the previous, more simplistic methods and surpasses their background rejection by a factor of more than four, equivalent to the improvement of the previous two generations of models.

These gains are not limited to simulation but also carry over to real LHC data, where we see robust modelling in collision events. First studies show that model performance will keep scaling with growing training datasets. The same architecture is general enough to serve many other tasks, such as signal extraction or lepton isolation, which separates leptons coming directly from W/Z or Higgs boson decays from those produced inside jets. With a single model taking over the work of many specialised tools, the field is moving towards what looks like a foundation model for the detector. These transformer-based flavour tagging algorithms have already become common infrastructure for LHC physics, directly improving the reach of searches that probe the Higgs potential.

2026-06-30-Siegen-FlavourTagging.pdf