Low Rank Approximation and Regression in Input Sparsity Time
Abstract
We design a new distribution over matrices so that for any fixed matrix of rank , with probability at least 9/10, simultaneously for all . Such a matrix is called a \emph{subspace embedding}. Furthermore, can be computed in time, where is the number of non-zero entries of . This improves over all previous subspace embeddings, which required at least time to achieve this property. We call our matrices \emph{sparse embedding matrices}. Using our sparse embedding matrices, we obtain the fastest known algorithms for -approximation for overconstrained least-squares regression, low-rank approximation, approximating all leverage scores, and -regression. The leading order term in the time complexity of our algorithms is or . We optimize the low-order terms in our running times (or for rank- approximation, the term), and show various tradeoffs. For instance, we also use our methods to design new preconditioners that improve the dependence on in least squares regression to . Finally, we provide preliminary experimental results which suggest that our algorithms are competitive in practice.
Cite
@article{arxiv.1207.6365,
title = {Low Rank Approximation and Regression in Input Sparsity Time},
author = {Kenneth L. Clarkson and David P. Woodruff},
journal= {arXiv preprint arXiv:1207.6365},
year = {2013}
}
Comments
Included optimization of subspace embedding dimension from (d/eps)^4 to O~(d/eps)^2 in Section 4, by more careful analysis of perfect hashing, and minor improvements to regression / low rank approximation because of it