Speaker
Description
Uncertainties in neutrino-nucleus cross-section measurements are usually evaluated by considering the spread of a measurement over an ensemble of variations of systematic parameters under the assumption they are distributed as a multivariate gaussian.
However, this cannot always be expected to be a safe assumption, in particular as we enter an era of systematic-limited measurements.
We showcase examples in which this assumption leads to incorrect conclusions when benchmarking neutrino interaction models and propose a solution to the issue.
We propose a method of directly learning the density of throws based on flow matching - a state-of-the-art generative modelling paradigm for training continuous normalizing flows.
We test our method in a realistic cross-section measurement example, showing it achieves excellent high-dimensional density estimation, surpassing the gaussian baseline and other machine learning methods.
| Poster prize | Yes |
|---|---|
| Given name | Radi |
| Surname | Radev |
| First affiliation | CERN |
| Institutional email | radi.radev@cern.ch |
| Gender | Male |
