Event Details

Many new physics searches at the LHC now wish to exploit substructure features of hadronic jets in order to increase sensitivity to signals involving boosted particle decays. Although theorists have developed highly useful variables to this end, such as N-subjettiness, in practice these variables can be highly correlated to observables of interest, such as jet mass. Such correlations introduce experimental difficulties in modeling signal and background distributions. I will discuss a method of using adversarially-trained neural networks to construct an optimal substructure tagger while minimizing correlation to the observable of interest. I will also discuss future applications of this method specifically in the context of boosted hadronic resonance searches, such as a low-mass Z’.