We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.DS

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Data Structures and Algorithms

Title: Improved Algorithms for Low Rank Approximation from Sparsity

Abstract: We overcome two major bottlenecks in the study of low rank approximation by assuming the low rank factors themselves are sparse. Specifically, (1) for low rank approximation with spectral norm error, we show how to improve the best known $\mathsf{nnz}(\mathbf A) k / \sqrt{\varepsilon}$ running time to $\mathsf{nnz}(\mathbf A)/\sqrt{\varepsilon}$ running time plus low order terms depending on the sparsity of the low rank factors, and (2) for streaming algorithms for Frobenius norm error, we show how to bypass the known $\Omega(nk/\varepsilon)$ memory lower bound and obtain an $s k (\log n)/ \mathrm{poly}(\varepsilon)$ memory bound, where $s$ is the number of non-zeros of each low rank factor. Although this algorithm is inefficient, as it must be under standard complexity theoretic assumptions, we also present polynomial time algorithms using $\mathrm{poly}(s,k,\log n,\varepsilon^{-1})$ memory that output rank $k$ approximations supported on a $O(sk/\varepsilon)\times O(sk/\varepsilon)$ submatrix.
Both the prior $\mathsf{nnz}(\mathbf A) k / \sqrt{\varepsilon}$ running time and the $nk/\varepsilon$ memory for these problems were long-standing barriers; our results give a natural way of overcoming them assuming sparsity of the low rank factors.
Comments: To appear in SODA 2022
Subjects: Data Structures and Algorithms (cs.DS); Numerical Analysis (math.NA)
Cite as: arXiv:2111.00668 [cs.DS]
  (or arXiv:2111.00668v1 [cs.DS] for this version)

Submission history

From: Taisuke Yasuda [view email]
[v1] Mon, 1 Nov 2021 02:52:14 GMT (46kb)

Link back to: arXiv, form interface, contact.