Heidelberg 2022 – scientific programme

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

T: Fachverband Teilchenphysik

T 79: Data Analysis, Information Technology and Artificial Intelligence 4

T 79.2: Talk

Wednesday, March 23, 2022, 16:30–16:45, T-H38

Clustering Energy Depositions in the Electromagnetic Calorimeter at Belle II using Graph Neural Networks — •Florian Wemmer, Pablo Goldenzweig, and Torben Ferber for the Belle II collaboration — Karlsruher Institut fuer Technologie

Electromagnetic calorimeters in particle detectors like at the Belle II Experiment consist of almost ten thousand sensitive crystals providing detailed energy deposition information in space. The correct assignment of energy depositions in those crystals to clusters originating from a distinct particle imposes a huge challenge especially in the presence of beam induced backgrounds, electronic noise and overlapping clusters. Graph Neural Networks (GNNs) allow for a machine learning algorithm to unrestrictedly and elegantly learn a feature space best suited to solve a problem. Using readily available Monte Carlo data we apply a GNN to try and cluster crystalwise energy information as well as distinguishing physics signals from beam background in the Belle II electromagnetic calorimeter. As a starting point to the development of more capable algorithms the - in actuality complex - detector data is simplified to two possibly overlapping clusters and beam background. We give insight to possible loss functions and metrics of the GNN as well as presenting first results of the clustering process.