Mass decorrelated taggers with adversarial neural networks


  Davide Melini  ,  Enrique   ,  Kajomovitz  
Technion

Jet substructure observables can be used to classify jets based on their radiation patterns and distinguish the signal of hadronically decaying resonances from non-resonant multijets background. The information given by such observables can be combined using machine learning techniques into a single-valued powerful classifier.
However, the jet substructure observables may be correlated with the jet mass, which may lead to a sculpting of the jet mass distributions when performing a selection based on substructure information. Mass-decorrelated jet substructure observables have the potential to increase the sensitivity of searches for new physics in final states with high-pT hadronically decaying resonances, both by minimising the sculpting of background jet mass distributions and enabling more robust background estimation