Machine Learning Methods for Histogram Deconvolution in High Energy Physics

Keras sequential neural networks are developed to perform histogram deconvolution on Z-boson mass spectra generated by the MadGraph5_aMC@NLO event generator using Pythia8 and Delphes. Three ways of interpreting the problem with neural networks are presented, tested and then compared with each other and the popular unfolding method TUnfold. A bin classification method is identified as a robust deconvolution method, with results comparable to that of TUnfold.