Bonn 2020 – wissenschaftliches Programm
Die DPG-Frühjahrstagung in Bonn musste abgesagt werden! Lesen Sie mehr ...
T 47.2: Vortrag
Mittwoch, 1. April 2020, 16:45–17:00, H-HS IV
New methods for the application of neural networks in the presence of systematic uncertainties — Simon Jörger1, Günter Quast1, Roger Wolf1, and •Stefan Wunsch1,2 — 1KIT — 2CERN
In this talk, we introduce new methods for the application of neural networks in the presence of systematic uncertainties, common in HEP analysis. We discuss a new technique to identify those features of the multi-dimensional input space, which the response of the network is most sensitive to, as a crucial input to the discussion of the propagation of systematic uncertainties onto the neural network output and therefore the measurement. A second method is presented to penalize such propagations of systematic variations controlling the impact of systematic uncertainties on the measurement.