Jet Flavour Tagging at FCC-ee with a Transformer-based Neural Network: DeepJetTransformer

• DOI: 10.48550/arxiv.2406.08590
• Publication Date: 2024-06-12
• Authors: F. Blekman; F. Canelli; A. De Moor; Kunal Gautam; A. Ilg; A. Macchiolo; E. Ploerer
• Keywords: Flavour

Jet flavour tagging is crucial in experimental high-energy physics. A algorithm, \texttt{DeepJetTransformer}, presented, which exploits a transformer-based neural network that substantially faster to train. The \texttt{DeepJetTransformer} uses information from particle flow-style objects and secondary vertex reconstruction as standard for $b$- $c$-jet identification supplemented by additional information, such reconstructed V$^0$s $K^{\pm}/\pi^{\pm}$ discrimination, typically not included algorithms at the LHC. model trained multiclassifier identify all quark flavours separately performs excellently identifying $c$-jets. An $s$-tagging efficiency of $40\%$ can be achieved with $10\%$ $ud$-jet background efficiency. impact including discrimination presented. applied on exclusive $Z \to q\bar{q}$ samples examine physics potential shown isolate s\bar{s}$ events. Assuming other backgrounds efficiently rejected, $5\sigma$ discovery significance an integrated luminosity $60~\text{nb}^{-1}$, corresponding less than second FCC-ee run plan $Z$ resonance.