Salvato in:
Dettagli Bibliografici
Autori principali: Shapira, Yuval, Drachsler-Cohen, Dana
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2511.10576
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866917320410005504
author Shapira, Yuval
Drachsler-Cohen, Dana
author_facet Shapira, Yuval
Drachsler-Cohen, Dana
contents Few-pixel attacks mislead a classifier by modifying a few pixels of an image. Their perturbation space is an $\ell_0$-ball, which is not convex, unlike $\ell_p$-balls for $p\geq1$. However, existing local robustness verifiers typically scale by relying on linear bound propagation, which captures convex perturbation spaces. We show that the convex hull of an $\ell_0$-ball is the intersection of its bounding box and an asymmetrically scaled $\ell_1$-like polytope. The volumes of the convex hull and this polytope are nearly equal as the input dimension increases. We then show a linear bound propagation that precisely computes bounds over the convex hull and is significantly tighter than bound propagations over the bounding box or our $\ell_1$-like polytope. This bound propagation scales the state-of-the-art $\ell_0$ verifier on its most challenging robustness benchmarks by 1.24x-7.07x, with a geometric mean of 3.16.
format Preprint
id arxiv_https___arxiv_org_abs_2511_10576
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Tight Robustness Certification Through the Convex Hull of $\ell_0$ Attacks
Shapira, Yuval
Drachsler-Cohen, Dana
Machine Learning
Few-pixel attacks mislead a classifier by modifying a few pixels of an image. Their perturbation space is an $\ell_0$-ball, which is not convex, unlike $\ell_p$-balls for $p\geq1$. However, existing local robustness verifiers typically scale by relying on linear bound propagation, which captures convex perturbation spaces. We show that the convex hull of an $\ell_0$-ball is the intersection of its bounding box and an asymmetrically scaled $\ell_1$-like polytope. The volumes of the convex hull and this polytope are nearly equal as the input dimension increases. We then show a linear bound propagation that precisely computes bounds over the convex hull and is significantly tighter than bound propagations over the bounding box or our $\ell_1$-like polytope. This bound propagation scales the state-of-the-art $\ell_0$ verifier on its most challenging robustness benchmarks by 1.24x-7.07x, with a geometric mean of 3.16.
title Tight Robustness Certification Through the Convex Hull of $\ell_0$ Attacks
topic Machine Learning
url https://arxiv.org/abs/2511.10576