Accuracy on the Line: On the Strong Correlation
      Between Out-of-Distribution and In-Distribution Generalization

   John Miller 1 Rohan Taori 2 Aditi Raghunathan 2 Shiori Sagawa 2 Pang Wei Koh 2 Vaishaal Shankar 1
                Percy Liang 2 Yair Carmon 3 Ludwig Schmidt 4

             Abstract                 standard paradigm to measure generalization is to evaluate a
   For machine learning systems to be reliable, we        model on a single test set drawn from the same distribution
   ...