Semi-supervised learning is a machine learning paradigm that leverages both limited labeled data and abundant unlabeled data to train models, occupying the middle ground between supervised and unsupervised learning. This approach is particularly valuable in domains where labeling is expensive, time-consuming, or requires expert knowledge—such as medical imaging, natural language processing, and computer vision. The core insight rests on key assumptions about data structure: the smoothness assumption (nearby points share labels), the cluster assumption (decision boundaries avoid high-density regions), and the manifold assumption (data lies on lower-dimensional manifolds). Semi-supervised methods use consistency regularization, pseudo-labeling, and graph-based propagation to extract supervisory signals from unlabeled data,enabling models to generalize better than supervised learning alone while avoiding the confirmation bias and error propagation pitfalls that arise when pseudo-labels are incorrect.