A Unified View of Label Shift Estimation
NIPS 2020, 2020.
We argue that these methods all employ calibration, either explicitly or implicitly, di ering only in the choice of calibration method and their optimization objective
Label shift describes the setting where although the label distribution might change between the source and target domains, the class-conditional probabilities (of data given a label) do not. There are two dominant approaches for estimating the label marginal. BBSE, a moment-matching approach based on confusion matrices, is provably con...More
PPT (Upload PPT)