Semi-Supervised Learning: More From Less Data

Semi-supervised learning (SSL) trains a model using a small set of labeled data and a large set of unlabeled data, like learning from a textbook with a few solved examples and many practice problems. This is common in computer vision, where getting millions of images is easy but labeling them is expensive. SSL leverages the unlabeled data to improve model generalization. The footgun is assuming the unlabeled data is clean; if it's from a different distribution than the labeled set, it can hurt model performance.