[DNN] Other Regularization Methods (Augmentation, Early Stopping)

1. Data augmentation

2. Early Stopping

• As you train more the weights become bigger.
• So you stop in the middle, ideally when the dev set error is at the minimum.