We have initiated a program to compute the Compton amplitude from lattice QCD with the Feynman-Hellman method. This amplitude is related to the structure function via a Fredholm integral equation of the first kind. It is known that these types of equations are inherently ill--posed - they are, e.g., extremely sensitive to perturbations of the system. We discuss two methods which are candidates to handle these problems: the model free inversion based on singular value decomposition and one Bayesian type approach. We apply the Bayesian method to currently available lattice data for
the Compton amplitude.