A preprint on arXiv describes a machine-learning-based surrogate for charged-current muon-neutrino nucleus scattering. The model is designed to generate neutrino interaction events while being calibrated to measured differential cross sections, avoiding the need for event-level experimental samples.

The approach starts from a theoretical interaction model and then adjusts the generated events to match experimental measurements. This could help produce simulated neutrino-event samples that are consistent with real data even when only cross-section measurements are available.

This summary is based on a single arXiv preprint, so there are no other sources to compare or contrast. The abstract does not provide details on the model architecture, validation results, or limitations.