Domain Generalization using Causal Matching
This work addresses domain generalization for machine learning models by tackling the limitation of existing invariance methods, offering a causal perspective that could improve robustness in applications like medical imaging, though it is incremental in refining theoretical foundations.
The paper identifies that class-conditional domain invariance is insufficient for domain generalization and proposes matching-based algorithms to align representations of the same object across domains, achieving competitive out-of-domain accuracy on datasets like rotated MNIST and PACS, with MatchDG showing over 50% overlap in ground-truth matches on MNIST and Fashion-MNIST.
In the domain generalization literature, a common objective is to learn representations independent of the domain after conditioning on the class label. We show that this objective is not sufficient: there exist counter-examples where a model fails to generalize to unseen domains even after satisfying class-conditional domain invariance. We formalize this observation through a structural causal model and show the importance of modeling within-class variations for generalization. Specifically, classes contain objects that characterize specific causal features, and domains can be interpreted as interventions on these objects that change non-causal features. We highlight an alternative condition: inputs across domains should have the same representation if they are derived from the same object. Based on this objective, we propose matching-based algorithms when base objects are observed (e.g., through data augmentation) and approximate the objective when objects are not observed (MatchDG). Our simple matching-based algorithms are competitive to prior work on out-of-domain accuracy for rotated MNIST, Fashion-MNIST, PACS, and Chest-Xray datasets. Our method MatchDG also recovers ground-truth object matches: on MNIST and Fashion-MNIST, top-10 matches from MatchDG have over 50% overlap with ground-truth matches.