Speakers
Sanmay Ganguly
(University of Tokyo (JP))
Etienne Dreyer
(Weizmann Institute of Science (IL))
Description
Neural network models that rely on jet substructure are commonly trained assuming jet constituents at truth level or smeared by parameterized detector response. However, the performance in such simplified circumstances may translate poorly to actual collider experiments. We investigate the impact by comparing large-R jet tagging using smeared particle-level jets versus jets built using detector-level inputs. Our tool for the study is an open-source, state-of-the-art calorimeter simulation [1] which realistically mimics the characteristics of a large general-purpose detector.
[1] https://iopscience.iop.org/article/10.1088/2632-2153/acf186
Authors
Sanmay Ganguly
(University of Tokyo (JP))
Etienne Dreyer
(Weizmann Institute of Science (IL))
Patrick Rieck
(New York University (US))
