Karlsruhe 2024 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
T: Fachverband Teilchenphysik
T 31: Methods in particle physics 3 (lepton reconstruction)
T 31.3: Vortrag
Dienstag, 5. März 2024, 16:30–16:45, Geb. 20.30: 2.066
ML4Taus: Tau decay mode classification using CNNs on Calorimeter Data — •Jonathan Pampel1, Jochen Dingfelder1, Tatjana Lenz1, Christina Dimitriadi1, and Duc Bao Ta2 — 1Rheinische Friedrich-Wilhelms-Universität, Bonn, Germany — 2Johannes Gutenberg Universität, Mainz, Germany
The tau-lepton is the heaviest charged lepton with a mass of about twice the mass of the proton. It can decay leptonically into two neutrinos and another lepton or hadronically into one neutrino and some hadrons, the latter being mostly pions. In the ATLAS collaboration at CERN, there are already several algorithms for the decay mode classification of hadronically decaying tau-leptons (tau-jets).
This talk presents a novel technique based on convolutional neural networks to classify the hadronic tau-lepton decay modes. The goal is to determine the number of neutral and charged pions in a tau-jet using calorimeter information. To do this, for each calorimeter layer, an ‘image’ of the tau-jet is generated. These ‘images’ are used as input for a neural network built from several 2D convolution and pooling layers and a flattening layer followed by a number of dense layers.
The talk includes an introduction to tau decay mode classification as well as a visualization of the preprocessed data which is fed into the neural network. Finally, the best performing neural network*s architecture and its performance will be presented. All performance evaluation is done using ATLAS Run 2 γ*→ττ Monte Carlo samples.
Keywords: Tau Decay Mode Classification; Calorimeter Data; Convolutional Neural Networks; ML4Taus; ML4Pions