Semi-supervised image classification explained

Semi-supervised machine learning is getting ready for primetime. In this article we review a number of common semi-supervised algorithms, capped by a presentation of our own Mean Teacher [arxiv, github], presented at NIPS 2017. Deep learning models have delivered superhuman performance for many years. However, training with standard supervised techniques requires huge amounts of correctly labeled data. Being able to use unlabeled data would open doors to many new applications in e.g.