Machine learning-based identification of highly Lorentz-boosted hadronically decaying particles at the CMS experiment

In this note, machine learning (ML) based techniques are presented to identify and classify hadronic decays of highly Lorentz-boosted W/Z/H bosons and top quarks, to be used by the CMS Collaboration. The techniques presented include the Energy Correlation Functions tagger, the Boosted Event Shape Tagger, the ImageTop tagger, and the DeepAK8 tagger. Techniques without ML have also been evaluated and are included for comparison. An alternative approach for jet clustering and identification, the Heavy Resonance Tagger with Variable-R, has been also studied. The identification performance is studied in simulated events and directly compared among algorithms. The algorithms are also validated using $35.9~\mathrm{fb}^{-1}$ of proton-proton events collected at $\sqrt{s}=13~\mathrm{TeV}$, and systematic uncertainties are assessed. The new techniques studied in this note provide significant performance improvements over non-ML techniques, reducing the background rate by up to a factor of $\sim$10 for the same signal efficiency.