Comparison of Image Processing Models in Quark Gluon Jet Classification

TL;DR

This paper compares convolutional and transformer-based models for quark-gluon jet classification, demonstrating that fine-tuned Swin Transformers with self-supervised pretraining achieve high accuracy and robustness in simulated jet image analysis.

Contribution

It introduces a comprehensive evaluation of CNNs and transformer models, highlighting the effectiveness of fine-tuning and self-supervised learning for jet classification tasks.

Findings

Swin-Tiny transformer with fine-tuning achieves 81.4% accuracy.

Self-supervised pretraining improves feature robustness.

Hierarchical attention models show promise for jet substructure analysis.

Abstract

We present a comprehensive comparison of convolutional and transformer-based models for distinguishing quark and gluon jets using simulated jet images from Pythia 8. By encoding jet substructure into a three-channel representation of particle kinematics, we evaluate the performance of convolutional neural networks (CNNs), Vision Transformers (ViTs), and Swin Transformers (Swin-Tiny) under both supervised and self-supervised learning setups. Our results show that fine-tuning only the final two transformer blocks of the Swin-Tiny model achieves the best trade-off between efficiency and accuracy, reaching 81.4% accuracy and an AUC (area under the ROC curve) of 88.9%. Self-supervised pretraining with Momentum Contrast (MoCo) further enhances feature robustness and reduces the number of trainable parameters. These findings highlight the potential of hierarchical attention-based models for jet substructure studies and for domain transfer to real collision data.