Skip to main content
We gratefully acknowledge support from the Simons Foundation, member institutions, and all contributors. Donate
arXiv logo
Cornell University Logo

Mathematics > Numerical Analysis

arXiv:1906.04223v16 (math)
[Submitted on 10 Jun 2019 (v1), revised 13 Nov 2024 (this version, v16), latest version 27 Nov 2024 (v17)]

Title:Near-Optimal Very Low Rank Approximation at Sublinear Cost

Authors:Victor Y. Pan, Qi Luan, Soo Go
View PDF HTML (experimental)
Abstract:Eckart-Young's theorem implies that even optimal rank-r approximation of an m-by-n matrix M for m not less than n is only interesting if there is a gap in the segment of the r largest singular values of M. If a matrix M also has a gap in the segment of the next p largest singular values, then one can efficiently compute its rank-r approximation with random sampling and oversampling. For a large subclass of matrices M having such a double gap in their spectra and for r+p greatly exceeded by the square root of n, our algorithm runs in sublinear time, and for a large subclass of the class of such matrices M its output rank-r approximation of M is near-optimal under the spectral norm. We also extend the classical iterative refinement to LRA and analyze and test the resulting algorithm.
Comments: 18 pages, 1 figures, 6 tables
Subjects: Numerical Analysis (math.NA)
Cite as: arXiv:1906.04223 [math.NA]
  (or arXiv:1906.04223v16 [math.NA] for this version)
  https://doi.org/10.48550/arXiv.1906.04223

Submission history

From: Victor Pan [view email]
[v1] Mon, 10 Jun 2019 18:47:28 UTC (36 KB)
[v2] Fri, 26 Jul 2019 20:46:24 UTC (37 KB)
[v3] Tue, 5 Nov 2019 18:24:37 UTC (35 KB)
[v4] Mon, 23 Dec 2019 14:32:03 UTC (35 KB)
[v5] Mon, 30 Dec 2019 15:39:38 UTC (35 KB)
[v6] Mon, 6 Apr 2020 15:28:35 UTC (35 KB)
[v7] Wed, 31 Mar 2021 21:00:29 UTC (59 KB)
[v8] Sun, 27 Aug 2023 16:44:37 UTC (356 KB)
[v9] Tue, 31 Oct 2023 16:17:32 UTC (59 KB)
[v10] Sun, 31 Dec 2023 18:18:29 UTC (59 KB)
[v11] Tue, 2 Jan 2024 04:56:07 UTC (60 KB)
[v12] Sat, 6 Jan 2024 23:04:38 UTC (59 KB)
[v13] Fri, 5 Apr 2024 17:58:35 UTC (79 KB)
[v14] Wed, 8 May 2024 21:20:32 UTC (121 KB)
[v15] Mon, 7 Oct 2024 21:49:11 UTC (138 KB)
[v16] Wed, 13 Nov 2024 18:31:11 UTC (65 KB)
[v17] Wed, 27 Nov 2024 16:46:19 UTC (62 KB)
Full-text links:

Access Paper:

view license
Current browse context:
math.NA
< prev   |   next >
new | recent | 2019-06
Change to browse by:
cs
cs.NA
math

References & Citations

export BibTeX citation

Bookmark

BibSonomy logo Reddit logo

Bibliographic and Citation Tools

Bibliographic Explorer (What is the Explorer?)
Connected Papers (What is Connected Papers?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)

Code, Data and Media Associated with this Article

alphaXiv (What is alphaXiv?)
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Hugging Face (What is Huggingface?)
Papers with Code (What is Papers with Code?)
ScienceCast (What is ScienceCast?)

Demos

Replicate (What is Replicate?)
Hugging Face Spaces (What is Spaces?)
TXYZ.AI (What is TXYZ.AI?)

Recommenders and Search Tools

Influence Flower (What are Influence Flowers?)
CORE Recommender (What is CORE?)

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)