Speaker
Description
The CMS electromagnetic calorimeter (ECAL) consists of more than 75,000 lead tungstate crystals located inside the magnetic field of the CMS solenoid magnet. The combined effect of the magnetic field and the material between the collision point and the ECAL leads to electrons and photons depositing their energy over several crystals. To achieve the best energy resolution it is essential to cluster together all channels containing energy deposits originating from the initial particle. The traditionally used topological energy clustering methods will see performance degradations during the LHC Run 3 (2022-2025) due to the larger number of particles from additional interactions overlapping in the detector and due to the ageing of the ECAL, resulting in higher noise levels. Different deep learning techniques such as graph neural networks and self-attention algorithms have been studied to make the cluster energy reconstruction more robust against the increased number of simultaneous interactions and the higher noise. This talk will present the concepts of the new clustering methods, the experience with the training and implementation of the models in the CMS reconstruction, and the achieved performance under Run 3 conditions.
| Collaboration / Activity | CMS |
|---|
