SCI TV - Drawing large graphs using approximate distance embedding

I will talk about recent research we have done on large graph visualization. There are two parts in this talk. First, I want to convince you that it possible to think formally about what a visualization is trying to achieve, and that it might be worthwhile to do so; the energy we try to minimize for the drawing comes directly from these considerations. Then, I want to show you how to use approximate low-rank matrix decompositions, in our case so we can compute such a drawing while taking less than O(n^2) memory and O(n^3) time, which was until then the state of the art for these problems.