Imagine having a top chef who can cook almost any dish in the world. That is your pre-trained model. Now, though, you want to teach him your grandmother’s perfect recipe for beef in cream sauce. Rather than teaching him to cook from the very beginning, you draw on his existing vast knowledge and merely show him the specifics of the new recipe. He adjusts his skills a little so that the dish tastes exactly as it should. In artificial intelligence, this process of taking a finished, capable model and merely tuning it for one particular task is called fine-tuning.
Fine-tuning is a transfer learning technique in which a pre-trained model, trained on a large general dataset, is further adapted to a specific task or domain. The process involves additional training on a smaller, narrowly focused dataset. During fine-tuning the weights of the neural network are adjusted slightly so that the model matches the characteristics of the new data better and achieves higher performance on the specialised task. An example is adapting a large language model pre-trained on general internet text for the purpose of analysing medical reports.