Identification of hadronic tau lepton decays using neural networks in the ATLAS experiment

This note describes a novel algorithm to identify the visible decay products of hadronic tau decays ($\tau_\text{had-vis}$) used by the ATLAS experiment for Run 2 of the LHC. The algorithm is based on recurrent neural networks (RNN) employing information from reconstructed charged-particle tracks and clusters of energy in the calorimeter associated to $\tau_\text{had-vis}$ candidates as well as high-level discriminating variables. The expected performance of this algorithm is evaluated in simulated proton-proton collisions at $\sqrt{s} = 13 \, \text{TeV}$ and compared to a BDT-based approach.