Process of adjusting a model's behavior to follow specific constraints or objectives during training.