Segmentation of EM showers for neutrino experiments with deep graph neural networks

We introduce a first-ever algorithm for the reconstruction of multiple showers from data collected with electromagnetic (EM) sampling calorimeters. Such detectors are widely used in High Energy Physics to measure energy and kinematics in-going particles. In this work, we consider case when many electrons pass through an Emulsion Cloud Chamber (ECC) brick, initiating electron-induced showers, which can be long exposure times or large input particle flux. For example, SHiP experiment is planning use emulsion dark matter search neutrino physics investigation. The expected full flux about 10^20 particles over five years. To reduce cost associated replacement ECC brick off-line taking (emulsion scanning), it decided increase time. Thus, expect observe lot overlapping turn EM into challenging point cloud segmentation problem. Our pipeline consists Graph Neural Network that predicts adjacency matrix clustering algorithm. propose new layer type (EmulsionConv) takes account geometrical properties shower development brick. modified hierarchical density-based method does not any prior information incoming identifies up 87% detectors. main test bench reconstructing going SND@LHC.