Jet tagging in the Lund plane with graph networks

A bstract The identification of boosted heavy particles such as top quarks or vector bosons is one the key problems arising in experimental studies at Large Hadron Collider. In this article, we introduce LundNet, a novel jet tagging method which relies on graph neural networks and an efficient description radiation patterns within to optimally disentangle signatures objects from background events. We apply framework number different benchmarks, showing significantly improved performance for compared existing state-of-the-art algorithms. study robustness LundNet taggers non-perturbative detector effects, show how kinematic cuts Lund plane can mitigate overfitting network model-dependent contributions. Finally, consider computational complexity its scaling function cuts, order magnitude improvement speed over previous graph-based taggers.