Transforming Data into Meaning. Data-Driven approaches for Particle Physics, Nuclear Power Safety and Humanitarian Crisis Situations.

Abstract

Machine learning and data intensive methods can be applied to a plethora of research domains. We apply supervised and unsupervised machine learning, Monte Carlo simulations and statistical tools to three diverse areas of research, tackling a range of computational and data analysis challenges unique to their respective environments.

Using SHERPA-a Monte Carlo event generator-as a Standard Model machine we generate thousands of particle collision events.
We employ a range of neural network architectures to determine the most powerful discriminating features which eliminate vast numbers of background events enabling us to calculate new constraints on the charm Yukawa coupling at the Large Hadron Collider and future projections.

Hartlepool Nuclear Power Station has a rich array of instrumentation that continuously monitors reactor health as frequently as every second, at all times. We apply unsupervised machine learning and Bayesian tools to scrutinise anomalous behaviour in the data which is indicative of instrumentation degradation prior to instrumentation failure.

JUNE-an agent based epidemiological simulation-is used to extract novel social mixing matrices at Cox's Bazar, a refugee camp in Bangladesh containing displaced people. These contact matrices can be used to understand social interactions and disease spread and therefore provide better utilisation of limited resources.