Saved in:
Bibliographic Details
Main Authors: You, Euijin, Lee, Hyang-Won
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.13645
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914265798017024
author You, Euijin
Lee, Hyang-Won
author_facet You, Euijin
Lee, Hyang-Won
contents Fast adversarial training (FAT) aims to enhance the robustness of models against adversarial attacks with reduced training time, however, FAT often suffers from compromised robustness due to insufficient exploration of adversarial space. In this paper, we develop a loss function to mitigate the problem of degraded robustness under FAT. Specifically, we derive a quadratic upper bound (QUB) on the adversarial training (AT) loss function and propose to utilize the bound with existing FAT methods. Our experimental results show that applying QUB loss to the existing methods yields significant improvement of robustness. Furthermore, using various metrics, we demonstrate that this improvement is likely to result from the smoothened loss landscape of the resulting model.
format Preprint
id arxiv_https___arxiv_org_abs_2601_13645
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Quadratic Upper Bound for Boosting Robustness
You, Euijin
Lee, Hyang-Won
Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
Fast adversarial training (FAT) aims to enhance the robustness of models against adversarial attacks with reduced training time, however, FAT often suffers from compromised robustness due to insufficient exploration of adversarial space. In this paper, we develop a loss function to mitigate the problem of degraded robustness under FAT. Specifically, we derive a quadratic upper bound (QUB) on the adversarial training (AT) loss function and propose to utilize the bound with existing FAT methods. Our experimental results show that applying QUB loss to the existing methods yields significant improvement of robustness. Furthermore, using various metrics, we demonstrate that this improvement is likely to result from the smoothened loss landscape of the resulting model.
title Quadratic Upper Bound for Boosting Robustness
topic Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2601.13645