AI and the World's Hard Problems · Physics · arXiv · 2026-10-19
Neural Network Representation of Generalized Parton Distributions (NNGPD)
Extracting generalized parton distributions (GPDs) from experimental data is an inverse problem. NNGPD is a physics-informed neural-network regression method that reconstructs GPDs from integral constraints, such as Compton form factors and Mellin moments, plus symmetry and endpoint constraints, without pointwise training data. Closure tests on pseudodata from a phenomenological model reproduced the main features of the underlying distributions.