Modern neural networks, with billions of parameters, are so overparameterized that they can "overfit" even random, ...
Regularization in Deep Learning is very important to overcome overfitting. When your training accuracy is very high, but test accuracy is very low, the model highly overfits the training dataset set ...