What separates unsupervised learning from semi-supervised learning?

To construct (usually predictive) models using supervised learning, labeled data is needed. Unsupervised learning is experimental in nature when working with unlabeled data (clustering, compression).

The former is less expensive and easier to organize. The number of unlabeled or partially labeled data is generally more significant than the number of recognized samples in semi-supervised learning. Although, you can use both labeled and unlabeled data to solve a supervised learning problem.