An implementation of neural simulation-based inference for parameter estimation in ATLAS

Abstract

Neural simulation-based inference (NSBI) is a powerful class of machine-learning-basedmethods for statistical inference that naturally handles high-dimensional parameter estimationwithout the need to bin data into low-dimensional summary histograms. Such methods arepromising for a range of measurements, including at the Large Hadron Collider, where no singleobservable may be optimal to scan over the entire theoretical phase space under consideration,or where binning data into histograms could result in a loss of sensitivity. This work develops aNSBI framework for statistical inference, using neural networks to estimate probability densityratios, which enables the application to a full-scale analysis. It incorporates a large number ofsystematic uncertainties, quantifies the uncertainty due to the finite number of events in trainingsamples, develops a method to construct confidence intervals, and demonstrates a series ofintermediate diagnostic checks that can be performed to validate the robustness of the method.As an example, the power and feasibility of the method are assessed on simulated data for asimplified version of an off-shell Higgs boson couplings measurement in the four-lepton finalstates. This approach represents an extension to the standard statistical methodology used by theexperiments at the Large Hadron Collider, and can benefit many physics analyses.