Interpreting Transformers for Jet Tagging

TL;DR

This paper interprets the attention mechanisms of the Particle Transformer model used for jet tagging in particle physics, revealing how it learns to focus on relevant particles and substructures, thereby improving understanding of its decision process.

Contribution

The study provides the first detailed analysis of attention patterns in ParT, linking them to traditional jet substructure observables and enhancing interpretability of transformer models in high-energy physics.

Findings

Attention heat maps show binary particle-to-particle attention patterns.

ParT's focus varies with decay types, indicating learning of jet substructure.

Insights suggest potential improvements for transformer architectures in physics applications.

Abstract

Machine learning (ML) algorithms, particularly attention-based transformer models, have become indispensable for analyzing the vast data generated by particle physics experiments like ATLAS and CMS at the CERN LHC. Particle Transformer (ParT), a state-of-the-art model, leverages particle-level attention to improve jet-tagging tasks, which are critical for identifying particles resulting from proton collisions. This study focuses on interpreting ParT by analyzing attention heat maps and particle-pair correlations on the $\eta$-$\phi$ plane, revealing a binary attention pattern where each particle attends to at most one other particle. At the same time, we observe that ParT shows varying focus on important particles and subjets depending on decay, indicating that the model learns traditional jet substructure observables. These insights enhance our understanding of the model's internal workings and learning process, offering potential avenues for improving the efficiency of transformer architectures in future high-energy physics applications.