Semi-supervised learning uses partially labeled data, that is where some of the training data is labelled with expected outputs/calssifications and some is unlabelled.
Used in Chap. 9: pages 128, 129; Chap. 18: page 282
: 