JetParticle-JEPA: An Efficient Self-Supervised Representation Learning method for Jet Tagging in High-Energy Physics

JetParticle-JEPA (JP-JEPA) is a self-supervised Joint-Embedding Predictive Architecture designed for jet tagging in high-energy physics, utilizing a Particle Transformer backbone to learn representations from continuous particle clouds without the need for tokenization. It achieves competitive performance on the JetClass benchmark compared to fully supervised methods and excels in low-label scenarios, while also demonstrating robustness to detector mismodeling and improved uncertainty quantification. This framework presents a significant advancement for practitioners in LHC physics, offering a data-efficient approach to jet representation learning.