he ATLAS Experiment at CERN has embraced modern AI techniques to revolutionise jet flavour tagging, a crucial process in analysing particle collisions. A new algorithm called GN2, powered by a Transformer neural network, directly analyses information from particle tracks and jets, eliminating the need for previous, hand-crafted algorithms. This advancement significantly improves the identification of b-jets and c-jets, which are vital for Standard Model measurements and the search for new physics phenomena. The ATLAS Collaboration has established robust pipelines to integrate and train these AI algorithms, leading to a substantial leap in performance and offering deeper insights into the physics signatures learned by the model. This innovative approach is already having a significant impact on ATLAS physics research, including enhancing the precision of Higgs boson studies and the search for new particles.

Paper link: https://arxiv.org/pdf/2505.19689