English

A note on estimating the dimension from a random geometric graph

Machine Learning 2023-11-23 v1 Machine Learning Statistics Theory Statistics Theory

Abstract

Let GnG_n be a random geometric graph with vertex set [n][n] based on nn i.i.d.\ random vectors X1,,XnX_1,\ldots,X_n drawn from an unknown density ff on Rd\R^d. An edge (i,j)(i,j) is present when XiXjrn\|X_i -X_j\| \le r_n, for a given threshold rnr_n possibly depending upon nn, where \| \cdot \| denotes Euclidean distance. We study the problem of estimating the dimension dd of the underlying space when we have access to the adjacency matrix of the graph but do not know rnr_n or the vectors XiX_i. The main result of the paper is that there exists an estimator of dd that converges to dd in probability as nn \to \infty for all densities with f5<\int f^5 < \infty whenever n3/2rndn^{3/2} r_n^d \to \infty and rn=o(1)r_n = o(1). The conditions allow very sparse graphs since when n3/2rnd0n^{3/2} r_n^d \to 0, the graph contains isolated edges only, with high probability. We also show that, without any condition on the density, a consistent estimator of dd exists when nrndn r_n^d \to \infty and rn=o(1)r_n = o(1).

Keywords

Cite

@article{arxiv.2311.13059,
  title  = {A note on estimating the dimension from a random geometric graph},
  author = {Caelan Atamanchuk and Luc Devroye and Gabor Lugosi},
  journal= {arXiv preprint arXiv:2311.13059},
  year   = {2023}
}