SMuK 2023 – wissenschaftliches Programm

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

T: Fachverband Teilchenphysik

T 63: ML Methods III

T 63.2: Vortrag

Mittwoch, 22. März 2023, 16:05–16:20, HSZ/0405

Optimising inference with binning — Phillip Keicher, Marcel Rieger, Peter Schleper, and •Jan Voss — Institut für Experimentalphysik Universität Hamburg, Hamburg, Deutschland

In order to increase the sensitivity of searches for rare processes, neural networks are nowadays a widely-spread tool to construct powerful discriminators. These discriminators are usually optimized to separate physics-motivated classes, but not necessarily on an optimal statistical inference. Consequently, the results can depend on auxiliary effects such as the exact binning choice for the distributions of the final discriminants.

This study aims to construct a setup for optimising the sensitivity with respect to the binning choice in the context of a Di-Higgs in the bb τ⁺τ⁻ final state. This setup is based on the python packages pyhf and JAX, which are used for the statistical modeling and the derivation of the inference with respect to the bin edges. This talk presents the current status of this on-going project and will highlight the challenges and possible applications of this novel technique.