Comparison between interactive (subjective) and traditional (numerical) inversion by genetic algorithms

Inversion algorithms employ numerical evaluation of the mismatch between model and data to guide the search for minima in parameter spaces. In an alternative approach, the numerical evaluation of data misfit can be replaced by subjective judgement of the solution quality. This widens the class of problems that can be treated within the framework of formal inverse theory, in particular including various applications in which "structural similarity" between model and data determines the quality of the fit. In this paper we compare the performance of a traditional numerical inversion with an interactive inversion, in which a priori knowledge, experience and even personal intuition are provided by the user via subjective judgement. The comparison is performed on a geological application and shows that user expertise can partly compensate for lack of sufficient constraints in the numerical inversion.