Leiden University Student Repository

Sluijter, Benjamin (2025)

Classifiers versus Normalizing Flows for uncertainty-aware estimation of a signal ratio

Master thesis | Physics (MSc)

In recent years, many advancements have been made in the field of Simulation-Based Inference (SBI), due to the lack of tractable likelihoods in modern physics experiments. In the High-Energy Physics (HEP) literature, a popular choice for doing SBI is by using binary classifiers, which can be used to obtain likelihood ratios by means of the likelihood-ratio trick. In the Astrophysics literature, on the other hand, more research is done on Normalizing Flows, which directly model the likelihoods. In this thesis, we compare the two methods, assessing their performance on a general HEP problem: inference of a signal ratio in the presence of a nuisance parameter. We perform this comparison on both a toy Gaussian example and a realistic Higgs decay example and do not find a clear winner over the two cases. We do find interesting qualitative differences, especially for poorly performing models, suggesting that it may be beneficial to implement both methods rather than selecting just one.

Author

Sluijter, Benjamin

Faculty

Faculty of Science

Specialisation

Classical/Quantum Information

Supervisors

Dorothea Samtleben

ECTS Credits

24

Language

en