ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 2304.03191
23
6

Krylov Methods are (nearly) Optimal for Low-Rank Approximation

6 April 2023
Ainesh Bakshi
Shyam Narayanan
ArXivPDFHTML
Abstract

We consider the problem of rank-111 low-rank approximation (LRA) in the matrix-vector product model under various Schatten norms: \min_{\|u\|_2=1} \|A (I - u u^\top)\|_{\mathcal{S}_p} , where ∥M∥Sp\|M\|_{\mathcal{S}_p}∥M∥Sp​​ denotes the ℓp\ell_pℓp​ norm of the singular values of MMM. Given ε>0\varepsilon>0ε>0, our goal is to output a unit vector vvv such that \|A(I - vv^\top)\|_{\mathcal{S}_p} \leq (1+\varepsilon) \min_{\|u\|_2=1}\|A(I - u u^\top)\|_{\mathcal{S}_p}. Our main result shows that Krylov methods (nearly) achieve the information-theoretically optimal number of matrix-vector products for Spectral (p=∞p=\inftyp=∞), Frobenius (p=2p=2p=2) and Nuclear (p=1p=1p=1) LRA. In particular, for Spectral LRA, we show that any algorithm requires Ω(log⁡(n)/ε1/2)\Omega\left(\log(n)/\varepsilon^{1/2}\right)Ω(log(n)/ε1/2) matrix-vector products, exactly matching the upper bound obtained by Krylov methods [MM15, BCW22]. Our lower bound addresses Open Question 1 in [Woo14], providing evidence for the lack of progress on algorithms for Spectral LRA and resolves Open Question 1.2 in [BCW22]. Next, we show that for any fixed constant ppp, i.e. 1≤p=O(1)1\leq p =O(1)1≤p=O(1), there is an upper bound of O(log⁡(1/ε)/ε1/3)O\left(\log(1/\varepsilon)/\varepsilon^{1/3}\right)O(log(1/ε)/ε1/3) matrix-vector products, implying that the complexity does not grow as a function of input size. This improves the O(log⁡(n/ε)/ε1/3)O\left(\log(n/\varepsilon)/\varepsilon^{1/3}\right)O(log(n/ε)/ε1/3) bound recently obtained in [BCW22], and matches their Ω(1/ε1/3)\Omega\left(1/\varepsilon^{1/3}\right)Ω(1/ε1/3) lower bound, to a log⁡(1/ε)\log(1/\varepsilon)log(1/ε) factor.

View on arXiv
Comments on this paper