Quark versus Gluon Jet Tagging Using Jet Images with the ATLAS Detector

Distinguishing quark-initiated from gluon-initiated jets is useful for many measurements and searches at the LHC. This note presents a jet tagger for distinguishing quark-initiated from gluon-initiated jets, which uses the full radiation pattern inside a jet processed as an image in a deep neural network classifier. The study is conducted using simulated dijet events in $\sqrt{s}$=13 TeV pp collisions with the ATLAS detector. Across a wide range of quark jet identification efficiencies, the neural network tagger achieves a gluon jet rejection that is comparable to or better than the performance of the jet width and track multiplicity observables conventionally used for quark-versus-gluon jet tagging.