Learning from virtual experiments to assist users of Small Angle Neutron Scattering in model selection

Abstract

In this work we combine the advantages of Small Angle Neutron Scattering (SANS) Monte Carlo simulations with the recent advances in computer vision to generate a tool that can assist SANS users in form factor model selection. We generate adataset of almost 260.000 SANS virtual experiments of the SANS beamline KWS-1 at FRM-II, Germany, intended for Machine Learning purposes. Then, we train a recommendation system based on an ensemble of Convolutional Neural Networks to predict the form factor model from the two-dimensional SANS scattering pattern measured at the position-sensitive detector of the beamline. The results show that the CNNs can learn the form factor prediction task, and that this recommendation system has a high accuracy in the classification task on 46 different form factor models. We also test the network with real data and explore the outcome. Finally, we discuss the reach of counting with the set of virtual experimental data presented here, and of such a recommendation system in the SANS user data analysis procedure.