NeurIPS 2020

Self-training Avoids Using Spurious Features Under Domain Shift

Meta Review

The authors prove that when the initial classifier is sufficiently accurate, self-training, via either pseudo-labels or entropy minimization, will improve robustness of the classifier against domain shift caused by spurious features. It is a piece of solid theory paper. Referees agree that this work is above the threshold and do not have further concerns after reading the authors' rebuttal.