On the Robustness of Most Probable Explanations