Nonsmooth Optimization with Zeroth Order Comparison Feedback

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Bakkali, Taha El, Chayti, El Mahdi, Saadi, Omar
Format: Preprint
Publié: 2026
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866917256582135808
author Bakkali, Taha El
Chayti, El Mahdi
Saadi, Omar
author_facet Bakkali, Taha El
Chayti, El Mahdi
Saadi, Omar
contents We study unconstrained optimization problems of nonsmooth, nonconvex Lipschitz functions, using only noisy pairwise comparisons governed by a known link function. Our goal is to compute a $(δ,\varepsilon)$-Goldstein stationary point. We combine randomized smoothing with a novel unbiased reduction from comparisons to local value differences. By leveraging a Russian-roulette truncation on the Bernoulli-product expansion of the inverse link, we construct an exactly unbiased estimator for directional differences. This estimator has finite expected cost and variance scaling quadratically with the function gap, $\mathcal{O}(B^2)$, under mild conditions. Plugging this into the smoothed gradient identity enables a standard nonconvex SGD analysis, yielding explicit comparison-complexity bounds for common symmetric links such as logistic, probit, and cauchit.
format Preprint
id arxiv_https___arxiv_org_abs_2602_05622
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Nonsmooth Optimization with Zeroth Order Comparison Feedback
Bakkali, Taha El
Chayti, El Mahdi
Saadi, Omar
Optimization and Control
We study unconstrained optimization problems of nonsmooth, nonconvex Lipschitz functions, using only noisy pairwise comparisons governed by a known link function. Our goal is to compute a $(δ,\varepsilon)$-Goldstein stationary point. We combine randomized smoothing with a novel unbiased reduction from comparisons to local value differences. By leveraging a Russian-roulette truncation on the Bernoulli-product expansion of the inverse link, we construct an exactly unbiased estimator for directional differences. This estimator has finite expected cost and variance scaling quadratically with the function gap, $\mathcal{O}(B^2)$, under mild conditions. Plugging this into the smoothed gradient identity enables a standard nonconvex SGD analysis, yielding explicit comparison-complexity bounds for common symmetric links such as logistic, probit, and cauchit.
title Nonsmooth Optimization with Zeroth Order Comparison Feedback
topic Optimization and Control
url https://arxiv.org/abs/2602.05622