arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2601.13645cs.LGcs.AIcs.CV
收藏

二次上界用于提升鲁棒性

Quadratic Upper Bound for Boosting Robustness

Department of Computer Science and Engineering, Konkuk University, Seoul, South Korea(计算机科学与工程系,konkuk大学,首尔,韩国)

Euijin You, Hyang-Won Lee

AI总结:

本文提出二次上界损失函数,用于提升对抗训练的鲁棒性,实验表明该方法能有效改善模型的鲁棒性。

摘要:

快速对抗训练(FAT)旨在通过减少训练时间增强模型对对抗攻击的鲁棒性,然而,由于对对抗空间的探索不足,FAT常常导致鲁棒性下降。在本文中,我们开发了一种损失函数以缓解FAT下的鲁棒性退化问题。具体来说,我们推导了对抗训练(AT)损失函数的二次上界(QUB),并提出利用该上界与现有的FAT方法结合。我们的实验结果表明,将QUB损失应用于现有方法可显著提升鲁棒性。此外,通过各种指标,我们证明这种改进很可能源于所得到模型的平滑损失景观。

英文摘要:

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.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑