Dimension-free estimators of gradients of functions with(out) non-independent variables

preprint2025arXivOpen access

Abstract

This study proposes a unified stochastic framework for approximating and computing the gradient of every smooth function evaluated at non-independent variables, using ℓp\ell_p-spherical distributions on Rd\R^d with d,p≥1d, p\geq 1. The upper-bounds of the bias of the gradient surrogates do not suffer from the curse of dimensionality for any p≥1p\geq 1. Also, the mean squared errors (MSEs) of the gradient estimators are bounded by K0N−1dK_0 N^{-1} d for any p∈[1,2]p \in [1, 2], and by K1N−1d2/pK_1 N^{-1} d^{2/p} when 2≤p≪d2 \leq p \ll d with NN the sample size and K0,K1K_0, K_1 some constants. Taking max⁡{2,log⁡(d)}<p≪d\max\left\{2, \log(d) \right\} < p \ll d allows for achieving dimension-free upper-bounds of MSEs. In the case where d≪p<+∞d\ll p< +\infty, the upper-bound K2N−1d2−2/p/(d+2)2K_2 N^{-1} d^{2-2/p}/ (d+2)^2 is reached with K2K_2 a constant. Such results lead to dimension-free MSEs of the proposed estimators, which boil down to estimators of the traditional gradient when the variables are independent. Numerical comparisons show the efficiency of the proposed approach.

Explore this paper’s authors, topics and related work.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

Reviews 0

Write a reviewWrite

No reviews yet.

Discussion 0

Add a commentWrite

No comments yet.