Flavour tagging with graph neural networks with the ATLAS detector

TL;DR

This paper presents new graph neural network-based algorithms for b-jet tagging in the ATLAS experiment, demonstrating promising performance on recent LHC data and discussing future prospects for HL-LHC.

Contribution

It introduces the application of graph neural networks to b-tagging in ATLAS, showcasing recent algorithm developments and performance evaluations.

Findings

Preliminary performance results on Run 3 data

Expected performance projections for HL-LHC

Demonstration of graph neural networks effectiveness in jet tagging

Abstract

The identification of jets containing a $b$-hadron, referred to as $b$-tagging, plays an important role for various physics measurements and searches carried out by the ATLAS experiment at the CERN Large Hadron Collider (LHC). The most recent $b$-tagging algorithm developments based on graph neural network architectures are presented. Preliminary performance on Run 3 data in $pp$ collisions at $\sqrt s = 13.6$ TeV is shown and expected performance at the High-Luminosity LHC (HL-LHC) discussed.