Geometric deep learning algorithms for tau lepton identification in the ATLAS experiment at the LHC.

The tau lepton is involved in many final states of Standard Model processes in proton-proton collisions at the LHC. It also appears in many signatures of predicted new physics beyond the Standard Model. As it decays predominantly to hadrons, it is essential to have a faithful reconstruction and identification of hadronic tau lepton decays, and a strong suppression of backgrounds. In the last few years geometric deep learning techniques have become the state of the art of many different tasks. For this reason in this contribution they are presented in different and recent geometric DL models to discriminate hadronic tau decays from noise given by QCD processes. A focus is given on the different approaches of the presented algorithms and on how their peculiar aspects influence the performances.