Isomap should be used when there is a non-linear mapping between your higher-dimensional data and your lower-dimensional manifold (e.g. data on a sphere).
Isomap is better than linear methods when dealing with almost all types of real image and motion tracking and we will now look at the example that was used in the Tenenbaum et al. of images of faces in different poses and light conditions.
The images are 4096 dimensions (64 pixel x 64 pixel).
We will reduce this down to just 2 dimensions