next up previous
Next: Tamara Munzner's work Up: Other comments Previous: Other comments

Mike Murphy

Regarding the problem from Hughes, one approach is to place the points on a surface such that the length of the geodesic between two points is the weight of the edge. I've always wanted to do something like this, but I can't get past the most basic questions. For example: if the weights assumed the triangle inequality, can you always find a surface?

The other "hack" type approach to the problem I have seen is to take a singular value decoposition on the distance matrix and project along the two largest eigenvectors. But that doesn't always give good results unless the data is very highly clustered.



Steve Skiena
Thu Oct 16 16:24:00 EDT 1997