We continue with our previous lecture on graph partitioning, starting with a review of Spectral methods. We then describe the idea of multi-level methods, which improve performance of other techniques by working on coarsened versions of the graph. Although existing software is very good for the classical graph partitioning problems, we describe some of the open questions in graph partitioning. Time permitting, we will also talk about load balancing in general, and how graph partitioning fits into the general area.
These notes steal liberally from lectures by previous CS267 instructors (in particular, Jim Demmel) and from Bruce Hendrickson for the open problems.
PowerPoint, Postscript, PDF