kapynResearch

Anti-Causal Domain Generalization: Leveraging Unlabeled Data

This paper introduces a novel approach to anti-causal domain generalization using unlabeled data. It addresses the limitation of existing methods that require labeled data from multiple training environments, making it applicable in data-scarce scenarios. The technique leverages the anti-causal structure where the outcome causes covariates, allowing for regularization of model sensitivity to environmental shifts.

Apple ML Research·Jul 2, 2026

Opening Kapyn…