the value of the validation loss does not reach zero (it decreases and increases) while the value of the training loss decreases

Question

I used Keras to create DenseNet-121.

the number of epochs is 400

the value of the validation loss does not reach zero (it decreases and increases) while the va