Speaker
Description
The increasing instantaneous luminosity and pile-up at the Large Hadron Collider impose stringent constraints on the online reconstruction and selection of electrons and photons in the ATLAS High-Level Trigger (HLT). To address these challenges, machine learning techniques are being integrated at multiple stages of the electromagnetic trigger sequence, from early energy reconstruction to object identification. This contribution reports recent advances across the HLT electron and photon reconstruction chain. A gradient-boosted decision tree-based energy calibration using calorimeter ring information has been developed for the FastCalo step, improving energy resolution while remaining pile-up robust and computationally efficient. The NeuralRinger algorithm, based on calorimeter rings, has been deployed early in the trigger sequence, enabling efficient background rejection for photons and significant rate reductions at high transverse energies. In parallel, new electron identification strategies combining calorimeter and tracking information in neural networks show improved background rejection relative to the current online selection. First studies of deep neural network–based electron identification at the HLT are also presented, compared with the established likelihood-based approach and deployment challenges. Together, these results demonstrate a coherent machine-learning-driven strategy to enhance the performance and scalability of electron and photon triggering in ATLAS for the High-Luminosity LHC era.
| I read the instructions above | Yes |
|---|