Precision measurements of theoretical parameters are a core element of the scientific program of experiments at the Large Hadron Collider (LHC) as well as other particle colliders. 

These are often performed using statistical techniques such as the method of maximum likelihood. However, given the size of datasets generated, reduction techniques, such as grouping data into bins, are often necessary. 

These can lead to a loss of sensitivity, particularly in non-linear cases like off-shell Higgs boson production and effective field theory measurements.  The non-linearity in these cases comes from quantum interference and traditional methods are unable to optimally distinguish the signal from background.

In this paper, the ATLAS collaboration pioneered the use of a neural network based technique called neural simulation-based inference (NSBI) to combat these issues. 

A neural network is a machine learning model originally inspired by how the human brain works. It’s made up of layers of interconnected units called neurons, which process information and learn patterns from data. Each neuron receives input, performs a simple calculation, and passes the result to other neurons. 

NSBI uses these neural networks to analyse each particle collision event individually, preserving more information and improving accuracy.

The framework developed here can handle many sources of uncertainty and includes tools to measure how confident scientists can be in their results.

The researchers benchmarked their method by using it to calculate the Higgs boson signal strength and compared it to previous methods with impressive results (see here for more details about this).

The greatly improved sensitivity gained from using this method will be invaluable in the search for physics beyond the Standard Model in future experiments at ATLAS and beyond.