# Dortmund 2021 – wissenschaftliches Programm

## Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

# T: Fachverband Teilchenphysik

## T 5: Higgs decay in fermions I

### T 5.5: Vortrag

### Montag, 15. März 2021, 17:00–17:15, Te

**Optimization of neural networks considering systematic uncertainties** — •Gessi Risto^{1}, Stefan Wunsch^{1,2}, Roger Wolf^{1}, and Guenter Quast^{1} — ^{1}Karlsruhe Institute of Technology, Institute of Experimental Particle Physics, Karlsruhe, Germany — ^{2}CERN, Geneva, Switzerland

Machine learning based data analysis strategies have shown an improved performance for many measurements in high-energy physics. This work presents a novel method of neural network optimization based on binned Poisson likelihoods with nuisance parameters to integrate the influence of systematic uncertainties in the training objective. We show with simple examples using pseudo-experiments and examples from high-energy physics that such an analysis strategy can result in an optimal measurement, and demonstrate an application of this method on a reduced CMS dataset used for the machine learning based SM STXS analysis of the Higgs to two tau leptons channel of CMS.