Do graph neural networks learn traditional jet substructure?
This work addresses the interpretability of machine learning models in high-energy physics, showing incremental insights into how graph neural networks leverage known physical properties for jet tagging.
The study investigated whether a state-of-the-art graph neural network, ParticleNet, uses traditional jet substructure observables like the number of prongs for jet tagging at the CERN LHC, finding that the model's decision-making process aligns with these traditional features as it distinguishes between signal jets from top quarks and background jets from lighter quarks and gluons.
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.