Isn’t that unsupervised learning during training or fine-tuning?