I've spent the last few weeks working on efficient algorithms for self-organization. The two major kinds I'm using are called a Kohonen Self-Organizing Map and a Growing Neural Gas. I've been investigating ways to use combinations of each for creating nice 2D topologies from 3D surfaces & input space, for effective automatic UV mapping (read: not ugly and impossible to work with). Hopefully this will save many many hours in the 3d modeler's / texture artist's workflow.

Below is a link to a video showing the Growing Neural Gas learning the topology of a face surface, given only input points (no actual topology).