Poisoning the Unlabeled Dataset of Semi-Supervised Learning

Nicholas Carlini

Semi-supervised machine learning models learn from a (small) set of labeled training examples, and a (large) set of unlabeled training examples. State-of-the-art models can reach w…