Speaker
Description
Calorimeter clustering is a classical pattern recognition task in high energy experiments. Traditionally this is performed by aggregating neighbor cells according to the expected topology with some fine-tuned algorithms, e.g. the Particle Flow Algorithm (PFA) that commonly adopted in the future lepton collider experiment. Nowadays the deep-learning models such as Graph Neural Network (GNN) and self-attention structure have shown great potential in handling similar complex tasks.
So we attempted to introduce these deep-learning models into the calorimeter clustering task, aiming for an end-to-end reconstruction algorithm in the granular calorimeter in future lepton collider experiments. Also it can be a convenient tool for the detector design and optimization at the preparation stage, since the fine-tuning procedure for the traditional algorithms can be replaced by re-training the model. This talk/poster will cover the preliminary progress we have and the discussions about this new approach.
