We recently put out our first public result for the CMS NextGen Triggers “Practical Real Time AI” task on improved jet tagging. The project started all the way back in late 2023 as the winning hackathon project at the CMS ML@L1T workshop (read more about that workshop in this post). Incorporating recent best practices from the offline world, we used a large sample of different processes and a multiclass model to tag b, c, uds, gluon, 𝜏+, 𝜏-, electron, and muon objects from constituents reconstructed by the Seeded Cone jet algorithm in the CMS Phase 2 Correlator Trigger. The model is a DeepSets Neural Network – whose output is invariant to permutations in the input particle set – with an architecture closely aligned to one from this study.

With this work we have improved b-tagging over the existing b-tagger, introduced new possibilities for tau-topology tagging with the charge categories, and introduced a multitude of new event selection opportunities for example charm-tagging, and light-vs-gluon discrimination for finding the forward jets from VBF events.

To start exploring how we can use this new jet tagger to better trigger on important processes at the HL-LHC, we created a trigger seed using total hadronic energy (HT) and the b score of the leading 4 jets in the event to select HH→bbbb events. We compared with a simple trigger requiring only an HT threshold and four jets of any flavour. At the same background trigger rate, we were able to reduce the threshold on HT by 150 GeV, at the expense of some efficiency on the plateau. A combination of these selections can eventually be used to ensure that CMS collects events from a wider spectrum of HT and di-Higgs invariant mass.

The new work has been presented by two PhD students from the team: Stella at EuCAIFCon and Duc at EPS, winning a poster award! The Detector Performance Summary with the full material is available here.

We plan to keep on improving the jet tagging performance, and to create more efficient event triggers that use the tagged jets. Stay tuned!