Characterizing Out-of-Distribution Error via Optimal Transport Y uzhe Lu

Out-of-distribution (OOD) data poses serious challenges in deployed machine learning models, so methods of predicting a model's performance on OOD data