arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2605.08191cs.CVcs.AI
收藏

通过协同平滑实现稳健的分布外检测框架

A Robust Out-of-Distribution Detection Framework via Synergistic Smoothing

Imperial College London(伦敦帝国学院);Zeroth Research(Zeroth研究)

Maria Stoica, Abdelrahman Hekal, Alessio Lomuscio

AI总结:

本文提出ROSS框架,通过基线分数的不稳定性区分分布内和分布外样本,实现对对抗攻击的强鲁棒性,实验表明其在多个数据集上表现优异。

摘要:

可靠的分布外(OOD)检测是安全部署机器学习系统的关键要求。尽管近期有进展,最先进的OOD检测器对对抗攻击高度敏感,这影响了其在自动化系统中的可信度。为了解决这一漏洞,我们应用中位数平滑处理基线OOD检测分数,平衡干净和对抗准确率。我们的关键见解是,用于中位数平滑生成的噪声样本可以被重新利用来量化基线分数的局部不稳定性。我们观察到,OOD样本在扰动下表现出更高的不稳定性。基于此,我们提出ROSS,一种新颖且稳健的后处理OOD检测器,利用基线分数的不稳定性进一步区分ID和OOD样本。ROSS实现了对称鲁棒性,对分数最小化和最大化攻击均表现强劲,不同于先前工作。这种对称防御导致了最先进的鲁棒性,比先前方法高出多达40 AUROC点。我们在CIFAR-10、CIFAR-100和ImageNet上进行了广泛的实验。代码可在:https://github.com/Abdu-Hekal/ROSS获取。

英文摘要:

Reliable out-of-distribution (OOD) detection is a critical requirement for the safe deployment of machine learning systems. Despite recent progress, state-of-the-art OOD detectors are highly susceptible to adversarial attacks, which undermines their trustworthiness in automated systems. To address this vulnerability, we apply median smoothing to baseline OOD detection scores, balancing clean and adversarial accuracies. Our key insight is that the noisy samples generated for median smoothing can be repurposed to quantify the local instability of the base score. We observe that OOD samples exhibit higher instability under perturbation. Based on this, we propose ROSS, a novel and robust post-hoc OOD detector that leverages the instability of baseline scores to further distinguish between in-distribution (ID) and OOD samples. ROSS achieves symmetric robustness, performing strongly against both score-minimising and score-maximising attacks, unlike prior work. This symmetric defence leads to state-of-the-art robustness, outperforming prior methods by up to 40 AUROC points. We demonstrate ROSS's effectiveness on extensive experiments across CIFAR-10, CIFAR-100, and ImageNet. Code is available at: https://github.com/Abdu-Hekal/ROSS.

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。

↑