Do graph neural networks learn traditional jet substructure?

TL;DR

This paper investigates how graph neural networks, specifically ParticleNet, learn to identify jet substructure at CERN LHC, revealing that they utilize traditional substructure features like prongs for jet classification.

Contribution

The study demonstrates that ParticleNet implicitly learns traditional jet substructure observables through analyzing relevant edge connections during training.

Findings

Relevant edges connect subjets differently for signal and background jets.

ParticleNet's decision process aligns with traditional jet substructure features.

Model behavior suggests learning of physical jet features like prongs.

Abstract

At the CERN LHC, the task of jet tagging, whose goal is to infer the origin of a jet given a set of final-state particles, is dominated by machine learning methods. Graph neural networks have been used to address this task by treating jets as point clouds with underlying, learnable, edge connections between the particles inside. We explore the decision-making process for one such state-of-the-art network, ParticleNet, by looking for relevant edge connections identified using the layerwise-relevance propagation technique. As the model is trained, we observe changes in the distribution of relevant edges connecting different intermediate clusters of particles, known as subjets. The resulting distribution of subjet connections is different for signal jets originating from top quarks, whose subjets typically correspond to its three decay products, and background jets originating from lighter quarks and gluons. This behavior indicates that the model is using traditional jet substructure observables, such as the number of prongs -- energetic particle clusters -- within a jet, when identifying jets.