Fast Adversarial Training against Sparse Attacks Requires Loss Smoothing

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhong, Xuyang, Huang, Yixiao, Liu, Chen
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915587769237504
author Zhong, Xuyang
Huang, Yixiao
Liu, Chen
author_facet Zhong, Xuyang
Huang, Yixiao
Liu, Chen
contents This paper studies fast adversarial training against sparse adversarial perturbations bounded by $l_0$ norm. We demonstrate the challenges of employing $1$-step attacks on $l_0$ bounded perturbations for fast adversarial training, including degraded performance and the occurrence of catastrophic overfitting (CO). We highlight that CO in $l_0$ adversarial training is caused by sub-optimal perturbation locations of $1$-step attack. Theoretical and empirical analyses reveal that the loss landscape of $l_0$ adversarial training is more craggy compared to its $l_\infty$, $l_2$ and $l_1$ counterparts. Moreover, we corroborate that the craggy loss landscape can aggravate CO. To address these issues, we propose Fast-LS-$l_0$ that incorporates soft labels and the trade-off loss function to smooth the adversarial loss landscape. Extensive experiments demonstrate our method can overcome the challenge of catastrophic overfitting, achieve state-of-the-art performance, and narrow down the performance gap between $1$-step and multi-step adversarial training against sparse attacks.
format Preprint
id arxiv_https___arxiv_org_abs_2502_21041
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Fast Adversarial Training against Sparse Attacks Requires Loss Smoothing
Zhong, Xuyang
Huang, Yixiao
Liu, Chen
Machine Learning
Artificial Intelligence
This paper studies fast adversarial training against sparse adversarial perturbations bounded by $l_0$ norm. We demonstrate the challenges of employing $1$-step attacks on $l_0$ bounded perturbations for fast adversarial training, including degraded performance and the occurrence of catastrophic overfitting (CO). We highlight that CO in $l_0$ adversarial training is caused by sub-optimal perturbation locations of $1$-step attack. Theoretical and empirical analyses reveal that the loss landscape of $l_0$ adversarial training is more craggy compared to its $l_\infty$, $l_2$ and $l_1$ counterparts. Moreover, we corroborate that the craggy loss landscape can aggravate CO. To address these issues, we propose Fast-LS-$l_0$ that incorporates soft labels and the trade-off loss function to smooth the adversarial loss landscape. Extensive experiments demonstrate our method can overcome the challenge of catastrophic overfitting, achieve state-of-the-art performance, and narrow down the performance gap between $1$-step and multi-step adversarial training against sparse attacks.
title Fast Adversarial Training against Sparse Attacks Requires Loss Smoothing
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2502.21041