Explainability in Graph Neural Networks for Jet Tagging

View on arXiv   GitHub

Abstract

Graph neural networks such as ParticleNet and transformer based networks on point clouds such as ParticleTransformer achieve state-of-the-art performance on jet tagging benchmarks at the Large Hadron Collider, yet the physical reasoning behind their predictions remains opaque. We present different methods, i.e. perturbation-based (GNNExplainer), Shapley-value-based (GNNShap), and gradient-based (GRADCam); adapted to operate on LundNet’s Lund-plane graph representation. Leveraging the fact that each node in the Lund plane corresponds to a physically meaningful parton splitting, we construct Monte Carlo truth explanation masks and introduce a physics-informed evaluation framework that goes beyond standard fidelity metrics. We perform the analysis in three transverse-momentum bins (p_T∈[500,700], [800,1000], and the inclusive region [500,1000] GeV), revealing how explanation quality and focus shift between non-perturbative and perturbative regimes. We further quantify the correlation between explainer-assigned node importance and classical jet substructure observables – N-subjettiness ratios τ₂₁ and τ₃₂ and the energy correlation functions – establishing the degree to which the model has learned known QCD features. We find that overall the weight assigned by explainability methods has a correlation with analytic observables, with expected shift across different phase space regimes, indicating that a trained neural network indeed learns some aspects of jet-substructure moments. Our open-source implementation enables reproducible explainability studies for graph-based jet taggers.

Summary

I worked on developing GNN-based jet-tagging classifiers and combined them with explainable AI tools such as GNNExplainer and Shapley values to examine how the model makes its decisions. This work showed that high-performing models do not need to remain black boxes, and that their predictions can stay physically consistent while revealing the jet substructure features they learn.