What is manifold learning in unsupervised learning architectures?

Manifold learning assumes data lies on a low-dimensional structure embedded in high-dimensional space.

Techniques like Isomap, UMAP, and t-SNE assume that high-dimensional data lies on a lower-dimensional manifold. The goal is to preserve local or global geometric structure when projecting into lower dimensions. This principle underlies many modern embedding methods used in unsupervised learning pipelines.