Selective training

Transfer learning makes the training phase a bit more complex because we should first restore the pretrained layers and define which ones should be frozen. Thankfully, several tools are available that simplify these operations.