AI and the World's Hard Problems · Physics · arXiv · 2026-10-22
Compositional neural networks for multimodal X-ray and neutron scattering inversion
Researchers developed a framework that combines separately trained posterior estimators at inference time to address ambiguity in reconstructing structures from scattering measurements. Reflectometry demonstrations combined multiple neutron contrasts and X-ray with neutron measurements, showing that pretrained models can support uncertainty-aware joint structural analysis without dedicated joint retraining.