Key points are not available for this paper at this time.
We address classification problems for which the training instances are governed by an input distribution that is allowed to differ arbitrarily from the test distribution---problems also referred t...
BickelSteffen et al. (Tue,) studied this question.