ParticleNet: Jet Tagging via Particle Clouds

TL;DR

ParticleNet introduces a jet representation as an unordered particle cloud and applies EdgeConv-based dynamic graph CNNs to jet tagging. The approach delivers state-of-the-art performance on top tagging and quark-gluon tagging benchmarks, highlighting the advantage of leveraging local particle neighborhoods and permutation symmetry. The inclusion of particle identification information further boosts performance, especially for quark-gluon tagging, while ParticleNet-Lite provides a favorable speed-accuracy trade-off. Overall, the particle cloud representation emerges as a natural, scalable framework with potential applications across jet physics and related collider tasks.

Abstract

How to represent a jet is at the core of machine learning on jet physics. Inspired by the notion of point clouds, we propose a new approach that considers a jet as an unordered set of its constituent particles, effectively a "particle cloud". Such a particle cloud representation of jets is efficient in incorporating raw information of jets and also explicitly respects the permutation symmetry. Based on the particle cloud representation, we propose ParticleNet, a customized neural network architecture using Dynamic Graph Convolutional Neural Network for jet tagging problems. The ParticleNet architecture achieves state-of-the-art performance on two representative jet tagging benchmarks and is improved significantly over existing methods.

Figures (4)

• Figure 1: The structure of the EdgeConv block.
• Figure 2: The architectures of the ParticleNet and the ParticleNet-Lite networks.
• Figure 3: Performance comparison in terms of ROC curves on the top tagging benchmark dataset.
• Figure 4: Performance comparison in terms of ROC curves on the quark-gluon tagging benchmark dataset.