Manifold Learning is a set of techniques aiming at uncovering a hypothesized lower-dimensional manifold close to which most of a given dataset is located. For example: in the 3D world, points in a sheet of paper can be represented in 3D "world coordinates", but also in 2D "paper coordinates" -- which are the same regardless of how the sheet of paper is "embedded" in 3D; manifold learning techniques could learn from the 3D points that they actually "live" in an embedded 2D manifold.