[vonnik]

Neural nets are a great way to reduce dimensionality. In particular, deep autoencoders. here's an old Hinton paper on it: http://www.cs.toronto.edu/~hinton/science.pdf