Developing group-aware priors improves machine learning models' robustness to subpopulation shifts.
Developing group-aware priors improves machine learning model robustness under subpopulation shifts.