Finetuning

Using a pre-trained network as a starting point, further training by progressively unfreezing layers, allowing weights in those layers to be updated during re-training