Erlangen 2026 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
T: Fachverband Teilchenphysik
T 32: Higgs Physics IV
T 32.6: Vortrag
Dienstag, 17. März 2026, 17:30–17:45, KH 01.019
Estimate of the contribution from jets misidentified as hadronic tau decays using normalizing flows — •Matthias Moser, Tamara App, Nikita Shadskiy, Artur Monsch, Markus Klute, Günter Quast, and Roger Wolf — KIT, Karlsruhe, Germany
For H → τ τ analyses, the most accurate estimation of the background originating from falsely identified hadronic τ decays remains a major challenge. To estimate this background from data, we employ the fake-factor method, in which the rate of misidentified τ candidates is obtained from a Determination Region, which is maximally pure in a given background process, and then applied in an Application Region to determine the rate of misidentified tau candidates in the desired Signal Region.
The resulting FF transfer function exhibits non-trivial dependencies on several variables, which are difficult to model using conventional techniques. It can be shown that these dependencies are determined by the probability density functions (p.d.f.) of events in the Application and Signal Regions. These p.d.f.’s can be learned with neural-networks, like e.g., normalizing flows. While the training procedure may be computationally intensive, the evaluation of the p.d.f.’s after training is fast. This approach provides an efficient and scalable description of the multi-variable dependence of the FF transfer function.
Keywords: CMS; Higgs; tau; normalizing flow; machine learning