Fine-tuning is a transfer learning technique to update the parameters of a pre-trained model for a specific use-case.
In fine-tuning, one epoch is used to train a new head on top of the pre-trained model, so the neural network works for the desired use-case.
Then, it uses an specified number of epochs to train the entire network, more quickly updating the last layers (specially the head) and then the earlier layers (which usually don’t require many changes).