A Sketching Algorithm for Spectral Graph Sparsification
Abstract
We study the problem of compressing a weighted graph on vertices, building a "sketch" of , so that given any vector , the value can be approximated up to a multiplicative factor from only and , where denotes the Laplacian of . One solution to this problem is to build a spectral sparsifier of , which, using the result of Batson, Spielman, and Srivastava, consists of reweighted edges of and has the property that simultaneously for all , . The bound is optimal for spectral sparsifiers. We show that if one is interested in only preserving the value of for a {\it fixed} (specified at query time) with high probability, then there is a sketch using only bits of space. This is the first data structure achieving a sub-quadratic dependence on . Our work builds upon recent work of Andoni, Krauthgamer, and Woodruff who showed that bits of space is possible for preserving a fixed {\it cut query} (i.e., ) with high probability; here we show that even for a general query vector , a sub-quadratic dependence on is possible. Our result for Laplacians is in sharp contrast to sketches for general positive semidefinite matrices with bit entries, for which even to preserve the value of for a fixed (specified at query time) up to a factor with constant probability, we show an lower bound.
Cite
@article{arxiv.1412.8225,
title = {A Sketching Algorithm for Spectral Graph Sparsification},
author = {Jiecao Chen and Bo Qin and David P. Woodruff and Qin Zhang},
journal= {arXiv preprint arXiv:1412.8225},
year = {2014}
}
Comments
18 pages