Machine-learning-based inversion of nuclear responses

A microscopic description of the interaction atomic nuclei with external electroweak probes is required for elucidating aspects short-range nuclear dynamics and correct interpretation neutrino oscillation experiments. Nuclear quantum Monte Carlo methods infer response functions from their Laplace transforms. Inverting transform a notoriously ill-posed problem; Bayesian techniques, such as maximum entropy, are typically used to reconstruct original in quasielastic region. In this work, we present physics-informed artificial neural network architecture suitable approximating inverse transform. Utilizing simulated, albeit realistic, electromagnetic functions, show that outperforms entropy both low-energy transfer regions, thereby allowing robust calculations electron scattering on inclusive muon capture rates.