Quark-Gluon tagging performance at the High-Luminosity LHC using constituent-based transformer models

Abstract: Jet constituents provide a more detailed description of a jet's radiation pattern than global observables. In simulations for ATLAS Run-2 data (2015-2018), transformer-based taggers trained on low-level inputs outperformed traditional methods using high-level variables with conventional neural networks for quark-gluon discrimination. With the upcoming High-Luminosity LHC (HL-LHC), which will deliver higher luminosity and energy, the ATLAS detector will be upgraded with an extended Inner Tracker covering the forward region, previously uncovered by a tracking detector. This work studies how these upgrades will improve the accuracy and robustness of quark-gluon jet taggers.