Reducing the Biases in Machine Learning Algorithms for Higgs Physics

This report examines the reduction of classification uncertainty for Higgs bosons produced in vector boson fusion (VBF) which decay via the diphoton channel. ATLAS reports a high uncertainty in the measured standard model (SM) Higgs to vector boson coupling strengths. An adversarial neural network (ANN) is developed to address VBF Higgs background modelling uncertainty. The ANN achieves higher signal purity over the ATLAS implementation through a 30% decrease in background efficiency. Systematic uncertainties between different simulation parameters are studied. The sole classification between VBF and gluon-gluon fusion Higgs production modes is excluded as the direct source of this uncertainty. These results provide a means for reducing the Higgs to vector boson coupling uncertainty, allowing for a more stringent test of the SM.