Benefits of Over-parameterization for Out-of-Distribution Generalization

Over-parameterization improves OOD generalization in machine learning models.