2nd IML Machine Learning Workshop

Deep neural network-based multi-class boosted object tagger in ATLAS

Not scheduled

20m

500/1-001 - Main Auditorium (CERN)

Session 5

Speaker

Ece Akilli (Universite de Geneve (CH))

Description

A deep neural network-based multi-class boosted object tagger is developed in the context of a search for pair production of heavy vector-like quarks with hadronic final states in ATLAS. The four classes of the tagger are W/Z (V)-boson, Higgs-boson, top-quark and background jets. As the unambiguous identification of the origin of the jet is essential for this search, an identification algorithm using this four-class deep neural network is designed to allow for this. In this analysis jets from boosted objects are reconstructed with a variable cone size by re-clustering calibrated small radius jets. Both lower level information (properties of the constituent small radius jets) and the higher level features of the variable radius reclustered jets are used to train the deep neural network. By using only this information as input to the tagger, the systematic uncertainties of the tagger are obtained by a propagation of the small radius jet uncertainties. The identification algorithm developed for this final state and its performance in Monte Carlo simulation will be presented.