Hands-On Transfer Learning with Python

Overfitting and underfitting in neural networks

As in any other ML training, the dataset for training a deep learning model is split into train, test, and validation. During iterative training of the model, typically the validation error is slightly more than the training error. If the gap between the test error and the validation error increases over iterations, it's a case of overfitting. If the training error stops decreasing to a sufficiently low value, we can conclude that the model is underfitting.