SSL uses a small portion of labeled data and lots of unlabeled data to train a classification model. There are several methods to label the portion of unlabeled data to be used as a training set e.g. Pseudo-labeling.
Technically, it could be viewed as performing clustering and then labeling the clusters within the labeled data. Unlabeled data with the same cluster as the labeled data share the same label as the labeled data.
https://en.m.wikipedia.org/w/index.php?title=Weak_supervision&diffonly=true
