Semi-supervised learning is an approach to machine learning that combines labeled data with large amount of unlabeled data during training.