Accurately inferring spurious attribute groups can significantly improve the worst-group accuracy of machine learning models by mitigating the negative impact of spurious correlations.
