English

A unified framework for linear dimensionality reduction in L1

Data Structures and Algorithms 2015-06-03 v5 Numerical Analysis Metric Geometry Probability

Abstract

For a family of interpolation norms 1,2,s\| \cdot \|_{1,2,s} on Rn\mathbb{R}^n, we provide a distribution over random matrices ΦsRm×n\Phi_s \in \mathbb{R}^{m \times n} parametrized by sparsity level ss such that for a fixed set XX of KK points in Rn\mathbb{R}^n, if mCslog(K)m \geq C s \log(K) then with high probability, 12x1,2,sΦs(x)12x1,2,s\frac{1}{2} \| x \|_{1,2,s} \leq \| \Phi_s (x) \|_1 \leq 2 \| x\|_{1,2,s} for all xXx\in X. Several existing results in the literature reduce to special cases of this result at different values of ss: for s=ns=n, x1,2,nx1\| x\|_{1,2,n} \equiv \| x \|_{1} and we recover that dimension reducing linear maps can preserve the 1\ell_1-norm up to a distortion proportional to the dimension reduction factor, which is known to be the best possible such result. For s=1s=1, x1,2,1x2\|x \|_{1,2,1} \equiv \| x \|_{2}, and we recover an 2/1\ell_2 / \ell_1 variant of the Johnson-Lindenstrauss Lemma for Gaussian random matrices. Finally, if xx is ss-sparse, then x1,2,s=x1\| x \|_{1,2,s} = \| x \|_1 and we recover that ss-sparse vectors in 1n\ell_1^n embed into 1O(slog(n))\ell_1^{\mathcal{O}(s \log(n))} via sparse random matrix constructions.

Keywords

Cite

@article{arxiv.1405.1332,
  title  = {A unified framework for linear dimensionality reduction in L1},
  author = {Felix Krahmer and Rachel Ward},
  journal= {arXiv preprint arXiv:1405.1332},
  year   = {2015}
}

Comments

18 pages, 1 figure

R2 v1 2026-06-22T04:07:23.310Z