Statistical and machine learning applications to inference in high-energy physics

High-energy physics is currently facing many statistical issues. Quantum field theory calculations require an enormous amount of computational resources. This limits the possibility of comparing theoretical predictions in new physics models to data, in particular in global fits. It has also proven difficult to properly quantify uncertainties from quantum field theory calculations due to the cut-off of the perturbative series, the so-called scale uncertainty.

The idea behind this research area is to investigate these issues faced by physicists on a theoretical level, and to develop new statistical methodologies to help in their work. This includes statistical learning methods to emulate the likelihood function rather than fully calculating it, and an improved statistical approach to quantifying scale uncertainties. In this context, it is also important to investigate the behaviour of inferential tools, such as the likelihood ratio test statistics, in non-standard settings, in particular in the case where the assumptions of Wilks' theorem do not hold. The goal is to derive corrections to the test statistics, or understanding their asymptotic distributions. Although motivated by high-energy physics applications, the findings will be useful in general.