过拟合

A model memorizing training data so well that performance on unseen data degrades.

Overfitting shows as great training performance but poor test performance, often from scarce data, oversized models or too many epochs. Mitigations include augmentation, regularization and early stopping.