Efficient coordinate-wise leading eigenvector computation

Jialei Wang; Weiran Wang; Dan Garber; Nathan Srebro

doi:10.48550/arxiv.1702.07834

Back

Efficient coordinate-wise leading eigenvector computation

Preprint

Open access

Efficient coordinate-wise leading eigenvector computation

Jialei Wang, Weiran Wang, Dan Garber and Nathan Srebro

arXiv (Cornell University)

02/25/2017

DOI: 10.48550/arxiv.1702.07834

Files and links (1)

url

https://doi.org/10.48550/arXiv.1702.07834View

Preprint (Author's original)This preprint has not been evaluated by subject experts through peer review. Preprints may undergo extensive changes and/or become peer-reviewed journal articles. Open Access

Abstract

We develop and analyze efficient "coordinate-wise" methods for finding the leading eigenvector, where each step involves only a vector-vector product. We establish global convergence with overall runtime guarantees that are at least as good as Lanczos's method and dominate it for slowly decaying spectrum. Our methods are based on combining a shift-and-invert approach with coordinate-wise algorithms for linear regression.

Computer Science - Learning

Computer Science - Numerical Analysis

Statistics - Machine Learning

Details

Title: Subtitle: Efficient coordinate-wise leading eigenvector computation
Creators: Jialei Wang
Weiran Wang
Dan Garber
Nathan Srebro
Resource Type: Preprint
Publication Details: arXiv (Cornell University)
DOI: 10.48550/arxiv.1702.07834
eISSN: 2331-8422
Language: English
Date posted: 02/25/2017
Academic Unit: Computer Science
Record Identifier: 9984696710402771

Metrics

1 Record Views