Input-Sparsity Low Rank Approximation in Schatten Norm
Abstract
We give the first input-sparsity time algorithms for the rank- low rank approximation problem in every Schatten norm. Specifically, for a given matrix , our algorithm computes , which, with high probability, satisfy , where is the Schatten -norm of a matrix with singular values , and where is the best rank- approximation to . Our algorithm runs in time , where for and for and is the exponent of matrix multiplication. For the important case of , which corresponds to the more "robust" nuclear norm, we obtain time, which was previously only known for the Frobenius norm (). Moreover, since for every , our algorithm has a better dependence on than that in the singular value decomposition for every . Crucial to our analysis is the use of dimensionality reduction for Ky-Fan -norms.
Cite
@article{arxiv.2004.12646,
title = {Input-Sparsity Low Rank Approximation in Schatten Norm},
author = {Yi Li and David Woodruff},
journal= {arXiv preprint arXiv:2004.12646},
year = {2020}
}
Comments
Submitted to ICML 2020. In this version, we state the result for rectangular matrices in the abstract and introduction and give an example comparing to low-rank approximation to A^TA