新型攻击隐蔽地针对测试时自适应模型

作者 PulseAugur 编辑部 · [2 个来源] · 2026-05-22 09:18

研究人员开发了一种新的样本级定向对抗攻击方法，该方法专门针对测试时自适应（TTA）场景。该方法旨在仅误分类包含攻击者选择的触发器的特定输入，同时保持良性查询的整体标签分布以逃避检测。所提出的基于元学习的攻击利用了一种新颖的优先级感知梯度对齐策略，以同时优化攻击成功率和分布隐蔽性。 AI

影响这项研究突显了测试时自适应方面的一个新漏洞，可能影响更鲁棒的防御机制的开发。

排序理由该集群包含一篇详细介绍新型对抗攻击方法的学术论文。

AI 生成摘要 · Google Gemini · 来自 2 个来源。我们如何撰写摘要 →

报道来源 [2]

arXiv cs.LG TIER_1 English(EN) · Phuc Duc Nguyen, Quang Duc Nguyen · 2026-05-25 04:00

Sample-wise Targeted Adversarial Attacks on Test-time Adaptation

arXiv:2605.23411v1 Announce Type: new Abstract: Test-time adaptation (TTA) effectively counters distribution shifts but exposes models to adversarial manipulation via the unlabeled test stream. Existing class-wise targeted attacks remain impractical for stealthy exploitation in t…
arXiv cs.CV TIER_1 English(EN) · Quang Duc Nguyen · 2026-05-22 09:18

Sample-wise Targeted Adversarial Attacks on Test-time Adaptation

Test-time adaptation (TTA) effectively counters distribution shifts but exposes models to adversarial manipulation via the unlabeled test stream. Existing class-wise targeted attacks remain impractical for stealthy exploitation in this setting: since TTA operates on batches, forc…