English

Improved Sparse Recovery for Approximate Matrix Multiplication

Data Structures and Algorithms 2026-02-05 v1

Abstract

We present a simple randomized algorithm for approximate matrix multiplication (AMM) whose error scales with the *output* norm ABF\|AB\|_F. Given any n×nn\times n matrices A,BA,B and a runtime parameter rnr\leq n, the algorithm produces in O(n2(r+logn))O(n^2(r+\log n)) time, a matrix CC with total squared error E[CABF2](1rn)ABF2\mathbb{E}[\|C-AB\|_F^2]\le (1-\frac{r}{n})\|AB\|_F^2, per-entry variance ABF2/n2\|AB\|_F^2/n^2 and bias E[C]=rnAB\mathbb{E}[C]=\frac{r}{n}AB. Alternatively, the algorithm can compute an *unbiased* estimation with expected total squared error nrABF2\frac{n}{r}\|{AB}\|_{F}^2, recovering the state-of-art AMM error obtained by Pagh's TensorSketch algorithm (Pagh, 2013). Our algorithm is a log-factor faster. The key insight in the algorithm is a new variation of pseudo-random rotation of the input matrices (a Fast Hadamard Transform with asymmetric diagonal scaling), which redistributes the Frobenius norm of the *output* ABAB uniformly across its entries.

Keywords

Cite

@article{arxiv.2602.04386,
  title  = {Improved Sparse Recovery for Approximate Matrix Multiplication},
  author = {Yahel Uffenheimer and Omri Weinstein},
  journal= {arXiv preprint arXiv:2602.04386},
  year   = {2026}
}
R2 v1 2026-07-01T09:35:40.035Z