Predicting Binary Outcomes Across Varying Environments Using Invariant Matching

The core message of this paper is that there exists a unique form of invariance in binary classification settings that allows for training models that are robust to changes in environmental conditions across multiple training environments.