Semi-Supervised Learning with Adversarially Missing Label Information

We address the problem of semi-supervised learning in an adversarial setting. Instead of assuming that labels are missing at random, we analyze a less favorable scenario where the label information ca