Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Particle Multi-Axis Transformer for Jet Tagging

Muhammad Usman, M Husnain Shahid, Maheen Ejaz, Ummay Hani, Nayab Fatima, Abdul Rehman Khan, Asifullah Khan, Nasir Majid Mirza

TL;DR

Jet tagging in high-energy physics is challenged by large, variable-length jet data. The authors introduce Particle Multi-axis Transformer (ParMAT), a transformer that combines local block and global grid attention with explicit pairwise particle interactions to produce permutation-invariant jet embeddings from per-particle features and pairwise cues. On the JetCLASS dataset of 100M jets across 10 classes, ParMAT achieves state-of-the-art accuracy and AUC, outperforming ParT, PFN, P-CNN, and ParticleNet, with notable improvements in background rejection. This work demonstrates a scalable, accurate approach to jet tagging with potential to enhance LHC discovery potential by leveraging rich particle-level information.

Abstract

Jet tagging is an essential categorization problem in high energy physics. In recent times, Deep Learning has not only risen to the challenge of jet tagging but also significantly improved its performance. In this article, we proposed an idea of a new architecture, Particle Multi-Axis transformer (ParMAT) which is a modified version of Particle transformer (ParT). ParMAT contains local and global spatial interactions within a single unit which improves its ability to handle various input lengths. We trained our model on JETCLASS, a publicly available large dataset that contains 100M jets of 10 different classes of particles. By integrating a parallel attention mechanism and pairwise interactions of particles in the attention mechanism, ParMAT achieves robustness and higher accuracy over the ParT and ParticleNet. The scalability of the model to huge datasets and its ability to automatically extract essential features demonstrate its potential for enhancing jet tagging.

Particle Multi-Axis Transformer for Jet Tagging

TL;DR

Jet tagging in high-energy physics is challenged by large, variable-length jet data. The authors introduce Particle Multi-axis Transformer (ParMAT), a transformer that combines local block and global grid attention with explicit pairwise particle interactions to produce permutation-invariant jet embeddings from per-particle features and pairwise cues. On the JetCLASS dataset of 100M jets across 10 classes, ParMAT achieves state-of-the-art accuracy and AUC, outperforming ParT, PFN, P-CNN, and ParticleNet, with notable improvements in background rejection. This work demonstrates a scalable, accurate approach to jet tagging with potential to enhance LHC discovery potential by leveraging rich particle-level information.

Abstract

Jet tagging is an essential categorization problem in high energy physics. In recent times, Deep Learning has not only risen to the challenge of jet tagging but also significantly improved its performance. In this article, we proposed an idea of a new architecture, Particle Multi-Axis transformer (ParMAT) which is a modified version of Particle transformer (ParT). ParMAT contains local and global spatial interactions within a single unit which improves its ability to handle various input lengths. We trained our model on JETCLASS, a publicly available large dataset that contains 100M jets of 10 different classes of particles. By integrating a parallel attention mechanism and pairwise interactions of particles in the attention mechanism, ParMAT achieves robustness and higher accuracy over the ParT and ParticleNet. The scalability of the model to huge datasets and its ability to automatically extract essential features demonstrate its potential for enhancing jet tagging.
Paper Structure (8 sections, 6 equations, 3 figures, 1 table)

This paper contains 8 sections, 6 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Dataset
  4. Model Architecture
  5. Experiment
  6. Evaluation Metrices
  7. Results
  8. Conclusion

Figures (3)

  • Figure 1: Cloud representation of an example of Higgs jet from JetClass. It shows all particles included in an example from Higgs jet along with the background noise. Higgs decays in to bottom quark anti quark pair which further decays into hadrons that are detected by detectors and are shown here.
  • Figure 2: The model diagram of (a) Particle Multi-Axis Transformer, (b) Particle Multi-Axis Attention block, (c) Multi-Head Attention block, (d) Multi-Axis Attention block, and (e) Class Attention Block.
  • Figure 3: (a) Training Accuracy vs Epochs and (b) Validation Accuracy vs Epochs.