Vision Transformers outperform CNNs for jet classification at Large Hadron Collider

Researchers have explored the application of Vision Transformers (ViTs) for classifying quark-gluon jets using calorimeter images from the Large Hadron Collider. Their study, which utilized simulated CMS Open Data, constructed multi-channel jet-view images for an end-to-end learning approach. The findings indicate that ViT-based models, particularly hybrid architectures like ViT+MaxViT and ViT+ConvNeXt, surpass traditional Convolutional Neural Networks in accuracy, F1-score, and ROC-AUC by effectively capturing long-range spatial correlations within jet substructures. AI

IMPACT Establishes new benchmarks for applying advanced vision models to high-energy physics data analysis.

RANK_REASON The item is a research paper detailing a new application of existing models to a scientific problem. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• arXiv
• CMS Open Data
• convolutional neural network
• École cantonale d'art de Lausanne
• HCAL
• Large Hadron Collider
• Quark-Gluon Jet Classification
• vision transformer
• Vít
• ViT+ConvNeXt
• ViT+MaxViT

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →