Speaker
Description
Identifying highly Lorentz-boosted Higgs bosons decaying to bottom-quark pairs is essential to the ATLAS physics programme. This poster reviews the evolution of boosted $X\to b\bar{b}$ tagging in ATLAS and the development of the corresponding calibration strategies. The Run-2 $D_{\rm Xbb}$ algorithm combined large-$R$-jet kinematics with flavour-tagging information from variable-radius track jets in a feed-forward neural network. GN2X introduced a transformer architecture that directly processes charged-particle tracks associated with the large-$R$ jet, improving the discrimination of $H\to b\bar{b}$ and $H\to c\bar{c}$ decays from top-quark and multijet backgrounds. The latest development, GN3X, further advances transformer-based identification of boosted Higgs bosons. Alongside this algorithmic evolution, we present the calibration methodology being developed for these taggers. The programme includes the extraction of signal-efficiency scale factors using $Z\to b\bar{b}+\mathrm{jets}$ events, dedicated studies of background mis-tag efficiencies and modelling, and investigations of dependencies on flavour composition and jet topology. Generator- and process-dependent differences are evaluated through MC-to-MC efficiency maps, which provide corrections for applying calibrations across simulated samples.Together, these developments aim to provide robust boosted-$H\to b\bar{b}$ identification for ATLAS Run-2 and Run-3 analyses.
