Making centralized (graph) computation faster, distributed and (at times) better

I will introduce a generic method for approximate inference in graphical models using graph partitioning. The resulting algorithm is linear time and provides an excellent approximation for the maximum