Check if you have access. The most basic assumption used in statistical learning theory is that training data and test data are drawn from the same underlying distribution. Domain adaptation for statistical classifiers daume h iii.
