Implicit Task-Driven Probability Discrepancy Measure for Unsupervised Domain Adaptation

However, most measures overlook the fact that the distributions are not the end-product of learning, but are the input of a downstream predictor.