Partenaires

CNRS IN2P3
Sorbonne Universite
Universite de Paris
Initiative Physique des Infinis
UPMC


Rechercher

Sur ce site

Sur le Web du CNRS


Accueil > Thèses, Stages, Formation et Enseignement > Propositions de thèses 2023 > Quantifying ML uncertainties in searches for new physics at the LHC

Quantifying ML uncertainties in searches for new physics at the LHC

par Tristan Beau - 12 décembre 2022

Titre : Quantifying ML uncertainties in searches for new physics at the LHC

Directrice/directeur de thèse : Bertrand Laforge

Co-encadrant.e : Anja Butter

Groupe d’accueil :ATLAS

Description :

Precise estimation of uncertainties is a crucial asset in the search for new physics at the LHC. While neural network based simulation and analysis methods have enable a more efficient treatment of high-dimensional data, a rigorous treatment of network induced uncertainties remains elusive.

In the research project the PhD student will explore different methods to estimate network induced uncertainties. Starting from toy examples that highlight limitations of interpolation and extrapolation the student will analyse the properties of multiple methods including Bayesian Neural Networks, ensemble methods and mutual information. Once advantages and limitations of each method are understood, the student will apply them to complex high energy physics problems like jet unfolding and event simulation.

Lieu(x) de travail : LPNHE

Stage proposé avant la thèse : Non

Facebook

Dans la même rubrique :

Enregistrer au format PDF