Performance of $W$/$Z$ taggers using UFO jets in ATLAS

The identification of boosted hadronic decays of $W$ and $Z$ bosons is a key technique for a variety of searches and measurements at the Large Hadron Collider. ATLAS has recently developed a new large-radius jet collection reconstructed from Unified Flow Objects (UFOs); these new objects improve the reconstruction of jet substructure variables used in boosted $W$/$Z$ tagging. This note presents the performance of the $W$/$Z$ taggers using UFO large-radius jets, based on both cuts on substructure variables and a deep neural network combining multiple jet substructure observables. Certain analyses can further benefit by using taggers defined to be decorrelated with respect to the jet mass; this decorrelation allows analyses to make use of mass sidebands or other strategies to better control their background estimates. Two different mass decorrelation techniques are presented: an analytical method using fixed-efficiency $k$-nearest neighbors regression, and a multivariate method using an adversarial neural network. The dependence of each tagger's performance on modeling-related effects is also studied.