Parts | Days | Selection | Search | Updates | Downloads | Help
T: Fachverband Teilchenphysik
T 71: Data analysis, Information technology III
T 71.3: Talk
Wednesday, March 17, 2021, 16:30–16:45, Tu
Adversarial Neural Network-based shape calibrations of observables for jet-tagging at CMS — Martin Erdmann1, •Benjamin Fischer1, Jan Middendorf1, Dennis Noll1, Yannik Alexander Rath1, Marcel Rieger2, Erwin Rudi1, and David Josef Schmidt1 — 1III. Physikalisches Institut A, RWTH Aachen University — 2CERN
Scale factors are commonly used in HEP to improve shape agreement between distributions of data and simulation. The choice of the underlying model for such corrections is of great importance, but often requires a lot of manual tuning e.g. of bin sizes or fitted functions. This can be alleviated through the use of neural networks and their inherent powerful data modeling capabilities.
We present a novel and generalized method for producing shape changing scale factors using adversarial neural networks. This method is investigated in the context of the bottom-quark jet-tagging algorithms within the CMS experiment. The scale factor of each jet is produced by the primary network using the jet's variables. The second network, the adversary, aims to differentiate between data and rescaled simulation events and facilitates the training of the former. An additional third network is used for normalization preservation with respect to correlated variables.
We present the conceptual design and resulting scale factors in comparison to the previously applied methods.