Identification of electrons using a deep neural network in the ATLAS experiment

This note introduces an algorithm to identify electrons in the ATLAS experiment based on a deep neural network. Inputs to the network are high-level discriminating variables derived from the reconstructed electron track and cluster of energy depositions in the calorimeter system. The performance is estimated in simulated proton-proton collisions at $\sqrt{s}=13$ TeV and compared to the current identification algorithm which is based on a likelihood approach. Depending on the kinematics of the electron candidate, an increase in background rejection between 1.7 and 5.5 at the same signal efficiency can be observed.