The 2009 Wolfram User Conference presentation,
DataModeler: A New Kind of Modeling Mark Kotanchek [zipped mathematica notebook],
focused on the unique benefits of DataModeler. As such, the emphasis was on ability to creatively hypothesize and refine diverse model forms and use this for variable selection and model development even when faced with ill-conditioned data or correlated inputs. The implications of this for data validation and curation are also illustrated in the Mathematica notebook used for the presentation.
Dealing with the deluge of multivariate data sets is difficult. The symbolic regression algorithms of DataModeler can convert that data into human interpretable models, handle correlated inputs, focus on the true driving variables and build models which have an associated trust metric to detect when the models is being asked to extrapolate into new operating regimes or the underlying system behavior has changed. The successful real-world applications of this technology include research acceleration, social science data analysis, production troubleshooting, active design-of-experiments, etc.