If your data distribution shifts, use self-learning

Evgenia Rusak, Steffen Schneider, George Pachitariu +5

We demonstrate that self-learning techniques like entropy minimization and pseudo-labeling are simple and effective at improving performance of a deployed computer vision model und…