10th RED LHC workshop

Transformer Neural Networks for identification of hadronically decaying $\tau$-leptons in ATLAS

Not scheduled

20m

Aula Magna, Facultat de Física (Universitat de Barcelona)

Speaker

Nadav Michael Tamir (Institut de Física d’Altes Energies (IFAE))

Description

Hadronically decaying τ-leptons (τ's) are a key experimental signature at the LHC, contributing to a wide range of Standard Model (SM) measurements and Beyond the SM searches. Reconstruction and identification ("tagging") of τ's is challenging due to substantial QCD contamination, and accurate τ-tagging is hence crucial in maximizing the potential sensitivity of physics analyses. This talk will provide an overview of τ-tagging in ATLAS, focusing on the improvements in QCD rejection rates obtained by adopting a Transformer-based architecture for the identification step. Beyond the performance gain, this approach also serves to promote a unified treatment of hadronic-object identification across ATLAS.