Speaker
Description
Tau leptons play a crucial role in studying electroweak processes, both in the Standard Model of particle physics and in searches for new physics. Accurate reconstruction and identification of tau leptons are essential in a high energy physics experiment. This talk presents DeepTau, the tau identification algorithm based on convolutional neural network (CNN), designed to reduce the misidentification probability of jets, muons and electrons while precisely tagging the hadronically decaying tau leptons reconstructed by the Hadron-plus-strip algorithm. The recently deployed version of DeepTau significantly enhances the performance in both efficiency and purity of tau identification as well as the robustness against mismodeling in simulation, w.r.t the previous version, by applying domain adaptation techniques through adversarial machine learning. Alternative approaches to identify hadronically decaying tau leptons as a jet flavor using universal jet taggers relying on Graph Neural Networks or Particle Transformers are also addressed. Additionally, dedicated reconstruction and identification techniques for displaced tau leptons coming from the decay of long-lived stau particles, using Graph Convolutional Neural Networks, are discussed.
