Würzburg 2018 – scientific programme

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

T: Fachverband Teilchenphysik

T 85: Experimentelle Methoden III

T 85.10: Talk

Thursday, March 22, 2018, 18:50–19:05, Z6 - SR 1.013

Track classification in hadronic tau decays with recurent neural networks — •Max Märker, Dirk Duschinger, Richard Hartmann, Wolfgang Mader, and Arno Straessner — IKTP TU Dresden

Tau leptons often play an important role in searches for new physics, not only because the Higgs decay probability into tau leptons is magnitudes larger than that for decays into muons or electrons, but also physics beyond the Standard Model can introduce enhanced couplings to tau leptons. However, their short lifetime makes it hard to detect tau leptons directly. In fact, tau decays in the ATLAS detector at the LHC often take place before any detector component. The majority of these decays are those into hadrons and additional neutrinos, where the hadronic constituents are most often 1 or 3 charged pions plus additional neutral pions. The classification of tracks of hadronic tau decays plays a crucial role in ATLAS tau reconstruction in terms of rejection against QCD jets and electrons.

In previous ATLAS analyses Boosted Decision Trees (BDT) where used successfully to separate tracks from hadronic tau decays and tracks from pile-up, conversions and underlying event. With recent developments in the field of artificial neural networks, new approaches are investigated utilizing the higher flexibility of neural networks to further improve the reconstruction of the charge multiplicity of hadronic tau decays. The focus is set on architectures using recurrent neural networks in order to learn the the special kinematic properties of tau decays.