Reusing learned parameters from a pre-trained model when initializing a new model to accelerate training and improve performance.