Maximum Likelihood with Bias-Corrected Calibration is Hard-To-Beat at Label Shift Adaptation
Published in
International Conference on Machine Learning(2019)External Links:
TL
TL;DR
It is shown that combining maximum likelihood with a type of calibration the authors call bias-corrected calibration outperforms both BBSL and RLLS across diverse datasets and distribution shifts, and it is proved that the maximum likelihood objective is concave.