Predictors from causal features do not generalize better to new domains

1 code implementation • 15 Feb 2024 • Vivian Y. Nastl, Moritz Hardt

Our goal is to test the hypothesis that models trained on causal features generalize better across domains.