Benign Overfitting

The counter-intuitive phenomenon where overparameterized models that perfectly interpolate training data, including noise, still generalize well out of sample, challenging the classical bias-variance tradeoff through mechanisms such as implicit regularization and the double descent curve.