SMuK 2023 – wissenschaftliches Programm

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

AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz

AKPIK 8: AI Topical Day – Normalizing Flows and Invertible Neural Networks (joint session AKPIK/T)

AKPIK 8.3: Vortrag

Donnerstag, 23. März 2023, 16:15–16:30, HSZ/0004

Introspection for a normalizing-flow-based recoil calibration — •Lars Sowa, Jost von den Driesch, Roger Wolf, Markus Klute, and Günter Quast — Institute of Experimental Particle Physics (ETP), Karlsruhe Institute of Technology (KIT)

Normalizing flows (NFs) are neural networks, that preserve the prob- ability between their input and output distributions. NFs can be promising candidates either as surrogates for the fast generation of new samples or as universal approximators of arbitrary probability density functions, based on which confidence intervals may be deter- mined, both of which are interesting properties in high-energy physics (HEP). This work presents the case study of recoil calibration on LHC Run- 3 data and Monte Carlo simulation with the goal to better understand the behavior of NFs. The result of the NF is compared to a deep ensem- ble of feed-forward neural networks created to compare the calibration results and the different coverage in the value space.