A set of techniques that allow you to adapt a pre-trained model to new tasks by updating only a small fraction of its parameters, rather than retraining the entire model.