Würzburg 2018 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

T: Fachverband Teilchenphysik

T 85: Experimentelle Methoden III

T 85.6: Talk

Thursday, March 22, 2018, 17:50–18:05, Z6 - SR 1.013

Identification of Hadronic Tau Lepton Decays with Deep Neural Networks at the ATLAS Experiment — •Christopher Deutsch, Jochen Dingfelder, and William Davey — Physikalisches Institut, Bonn, Deutschland

The tau lepton is the heaviest lepton in the Standard Model and an important probe of physics at high energy scales, such as Higgs physics and physics beyond the Standard Model. Hadronic decays make up approximately two-thirds of the total tau lepton branching ratio.

Jets originating from quarks or gluons can mimic hadronic tau decays. They are more abundant than tau leptons due to the large multijet production cross section at the LHC. Therefore, dedicated algorithms to discriminate hadronically decaying tau leptons from jets are required.

In this talk, the latest developments on a novel tau identification algorithm based on deep learning for data collected with the ATLAS detector during Run 2 of the LHC are presented. The new algorithm combines information on reconstructed objects and high-level identification variables in a neural network to build a powerful discriminant. A recurrent neural network architecture is used, allowing to process input sequences of variable length such as charged particle tracks and clusters of energy in the calorimeter. The network is expected to improve the jet rejection by a factor of two compared to the tau identification algorithm currently in use at the ATLAS experiment.