# PCN: A Deep Learning Approach to Jet Tagging Utilizing Novel Graph Construction Methods and Chebyshev Graph Convolutions

**Authors:** Yash Semlani, Mihir Relan, Krithik Ramesh  
**arXiv:** [2309.08630](https://arxiv.org/abs/2309.08630)  
**Date:** 2023-09-12  
**Summary page:** https://www.summarizepaper.com/en/paper/2309.08630v3/

## Key points

- Jet tagging is a crucial classification problem in high-energy physics experiments involving identifying and associating collimated sprays of subatomic particles known as jets with their emitter particle.
- Current approaches use deep learning techniques to analyze collision data, but inconsistent representation of jets as inputs leads to the omission of informative features.
- A novel graph-based representation of jets called Particle Chebyshev Network (PCN) is proposed in this study to encode maximum information.
- PCN utilizes Chebyshev graph convolutions (ChebConv), which have not been explored in jet tagging before, leading to substantial improvement in accuracy compared to existing taggers.
- The research opens up possibilities for future investigations into graph-based representations of jets and the utilization of ChebConv layers in high-energy physics experiments.
- Detailed insights are provided into the methodology used, including graph construction and model architectures incorporating localized interaction analysis through Chebyshev Graph Convolutions and global jet structure analysis using Edge Convolutions.
- Results section discusses dataset used for experimentation and compares PCN performance with state-of-the-art models in jet tagging.
- Study emphasizes the significance of advancements in jet tagging methodologies tailored for high-energy physics experiments at facilities like the Large Hadron Collider at CERN.

## Summary

In high-energy physics experiments, jet tagging is a crucial classification problem that involves identifying the collimated sprays of subatomic particles known as jets from particle collisions and associating them with their emitter particle. This process plays a significant role in searching for new physics beyond the Standard Model. Current approaches utilize deep learning techniques to analyze complex collision data and uncover hidden patterns. However, the representation of jets as inputs to deep learning models has been inconsistent, leading to the omission of informative features. To address this issue, a novel graph-based representation of jets is proposed in this study to encode maximum information. The researchers introduce the Particle Chebyshev Network (PCN), a graph neural network (GNN) that leverages Chebyshev graph convolutions (ChebConv). ChebConv has shown effectiveness as an alternative to traditional graph convolutions in GNNs but has not been explored in jet tagging before. By utilizing PCN, the study achieves a substantial improvement in accuracy compared to existing taggers. The research also opens up possibilities for future investigations into graph-based representations of jets and the utilization of ChebConv layers in high-energy physics experiments. The code for PCN-Jet-Tagging is available on GitHub for further exploration. The study provides detailed insights into the methodology used, including graph construction and model architectures incorporating localized interaction analysis through Chebyshev Graph Convolutions and global jet structure analysis using Edge Convolutions. The results section discusses the dataset used for experimentation and compares the performance of PCN with state-of-the-art models in jet tagging. The discussion highlights the significance of advancements in jet tagging methodologies and emphasizes how current approaches often fall short due to limitations in capturing intricate interactions within particle collisions. Overall, this study contributes valuable insights into improving jet tagging accuracy through innovative graph-based representations and advanced neural network architectures tailored specifically for high-energy physics experiments at facilities like the Large Hadron Collider at CERN.