English

Multiplicative Rank-1 Approximation using Length-Squared Sampling

Data Structures and Algorithms 2019-10-30 v2

Abstract

We show that the span of Ω(1ε4)\Omega(\frac{1}{\varepsilon^4}) rows of any matrix ARn×dA \subset \mathbb{R}^{n \times d} sampled according to the length-squared distribution contains a rank-11 matrix A~\tilde{A} such that AA~F2(1+ε)Aπ1(A)F2||A - \tilde{A}||_F^2 \leq (1 + \varepsilon) \cdot ||A - \pi_1(A)||_F^2, where π1(A)\pi_1(A) denotes the best rank-11 approximation of AA under the Frobenius norm. Length-squared sampling has previously been used in the context of rank-kk approximation. However, the approximation obtained was additive in nature. We obtain a multiplicative approximation albeit only for rank-11 approximation.

Cite

@article{arxiv.1909.07515,
  title  = {Multiplicative Rank-1 Approximation using Length-Squared Sampling},
  author = {Ragesh Jaiswal and Amit Kumar},
  journal= {arXiv preprint arXiv:1909.07515},
  year   = {2019}
}

Comments

A section on open problems added in the new version

R2 v1 2026-06-23T11:17:21.187Z