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

Jet flavour tagging is crucial in experimental high-energy physics. A tagging algorithm, DeepJetTransformer, is presented, which exploits a transformer-based neural network that is substantially faster to train. The DeepJetTransformer network uses information from particle flow-style objects and secondary vertex reconstruction as is standard for b- and c-jet identification supplemented by additional information, such as reconstructed V⁰s and K±/π± discrimination, typically not included in tagging algorithms at the LHC. The model is trained as a multiclassifier to identify all quark flavours separately and performs excellently in identifying b- and c-jets. An s-tagging efficiency of $40%$ can be achieved with a $10%$ $ud$-jet background efficiency. The impact of including V⁰s and K±/π± discrimination is presented. The network is applied on exclusive Z → qq̄ samples to examine the physics potential and is shown to isolate Z → ss̄ events. Assuming all other backgrounds can be efficiently rejected, a 5σ discovery significance for Z → ss̄ can be achieved with an integrated luminosity of 60~nb⁻¹, corresponding to less than a second of the FCC-ee run plan at the Z resonance.