Parametrizing Compton form factors with neural networks

We describe a method, based on neural networks, of revealing Compton form factors in the deeply virtual region. We compare this approach to standard least-squares model fitting both for a simplified toy case and for HERMES data.