20. Dimensionality reduction

A common form of unsupervised learning is dimensionality reduction, in which we learn a mapping from the high-dimensional visible space xRD\bold{x}\in\R^D to a low-dimensional latent space zRL\bold{z}\in \R^L.

This mapping can either be parametric z=f(x;θ)\bold{z}=f(\bold{x};\theta), which can be applied to any input, or it can be a nonparametric mapping where we compute an embeddings zn\bold{z}_n for each input xn\bold{x}_n in the dataset, but not for any other point.

The former is mostly used for data visualization whereas the latter can be used as a preprocessing step for other kind of learning algorithms. For example, we can produce an embedding by mapping x\bold{x} to z\bold{z}, and then learn a simple linear classifier by mapping z\bold{z} to yy.