Improving Sketching Algorithms for Low-Rank Matrix Approximation via Sketch-Power Iterations

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Chang, Chao, Yang, Yuning
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914426589806592
author Chang, Chao
Yang, Yuning
author_facet Chang, Chao
Yang, Yuning
contents Power iteration can improve the accuracy of randomized SVD, but requires multiple data passes, making it impractical in streaming or memory-constrained settings. We introduce a lightweight yet effective sketch-power iteration, allowing power-like iterations with only a single pass of the data, which can be incorporated into one-pass algorithms for low-rank approximation. As an example, we integrate the sketch-power iteration into a one-pass algorithm proposed by Tropp et al., and introduce strategies to reduce its storage cost. We establish meaningful error bounds: given a fixed storage budget, the sketch sizes derived from the bounds closely match the optimal ones observed in reality. This allows one to preselect reasonable parameters. Numerical experiments on both synthetic and real-world datasets indicate that, under the same storage constraints, applying one or two sketch-power iterations can substantially improve the approximation accuracy of the considered one-pass algorithms. In particular, experiments on real data with flat spectrum show that the method can approximate the dominant singular vectors well.
format Preprint
id arxiv_https___arxiv_org_abs_2603_26298
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Improving Sketching Algorithms for Low-Rank Matrix Approximation via Sketch-Power Iterations
Chang, Chao
Yang, Yuning
Numerical Analysis
65F55, 15B33, 68W20, 65F35
Power iteration can improve the accuracy of randomized SVD, but requires multiple data passes, making it impractical in streaming or memory-constrained settings. We introduce a lightweight yet effective sketch-power iteration, allowing power-like iterations with only a single pass of the data, which can be incorporated into one-pass algorithms for low-rank approximation. As an example, we integrate the sketch-power iteration into a one-pass algorithm proposed by Tropp et al., and introduce strategies to reduce its storage cost. We establish meaningful error bounds: given a fixed storage budget, the sketch sizes derived from the bounds closely match the optimal ones observed in reality. This allows one to preselect reasonable parameters. Numerical experiments on both synthetic and real-world datasets indicate that, under the same storage constraints, applying one or two sketch-power iterations can substantially improve the approximation accuracy of the considered one-pass algorithms. In particular, experiments on real data with flat spectrum show that the method can approximate the dominant singular vectors well.
title Improving Sketching Algorithms for Low-Rank Matrix Approximation via Sketch-Power Iterations
topic Numerical Analysis
65F55, 15B33, 68W20, 65F35
url https://arxiv.org/abs/2603.26298