Explaining machine-learned particle-flow reconstruction

• DOI: 10.48550/arxiv.2111.12840
• Publication Date: 2021-01-01
• Authors: Farouk Mokhtar; R. Kansal; D. Díaz; J. Duarte; Joosep Pata; M. Pierini; Jean-Roch Vlimant
• Keywords: Relevance

The particle-flow (PF) algorithm is used in general-purpose particle detectors to reconstruct a comprehensive particle-level view of the collision by combining information from different subdetectors. A graph neural network (GNN) model, known as machine-learned (MLPF) algorithm, has been developed substitute rule-based PF algorithm. However, understanding model's decision making not straightforward, especially given complexity set-to-set prediction task, dynamic building, and message-passing steps. In this paper, we adapt layerwise-relevance propagation technique for GNNs apply it MLPF gauge relevant nodes features its predictions. Through process, gain insight into decision-making.