Jet Tagging

Jet tagging studies how to identify the particles and processes that produced sprays of particles, or jets, in collider experiments. Models commonly learn from particle-level features using graph neural networks, transformers, Lorentz-equivariant architectures, and physics-informed representations such as jet-clustering histories; evaluation includes tasks like top-quark, Higgs, quark–gluon, and heavy-flavour tagging. Current work also explores self-supervised learning with limited labels, interpretable representations, and efficient deployment under real-time hardware constraints. A central challenge is ensuring that gains on simulated benchmarks transfer to collision data, while balancing accuracy, robustness, interpretability, and computational cost across different jet topologies.