Out-of-distribution (OOD) generalization remains a major issue in modern deep learning. One of the main causes of poor OOD generalization is the reliance on spurious features — patterns that are predictive of the class label in the training data distribution, but that are not causally related to the target. Standard methods for reducing the reliance on spurious features typically assume that we know what the spurious feature is, and methods that attempt to alleviate this limitation are complex,…