Non-linear effects such as from quantum interference pose significant challenges to the established statistical methods employed at the Large Hadron Collider (LHC). These are of particular concern in some of the most important measurements in collider physics, including that of the Higgs width. Neural Simulation-Based Inference is a powerful class of machine learning-based methods for statistical inference that naturally handle these challenges by performing high dimensional parameter estimation, without the need to bin data into low-dimensional summary histograms. I will discuss these challenges in Higgs physics and the solution developed, first in a phenomenology study, and then implemented in the ATLAS experiment. The dramatic improvement in sensitivity for this analysis promises significant gains to be had in several other studies at the LHC and more generally in the physical sciences, with the use of these newly developed methods.
If I have time, I will also show how the same story then applies to astrophysics, where these high-dimensional statistical techniques enable enhanced uncertainty propagation and interpretability for a problem of neutron stars.
