视觉数据集中天然存在类似后门的对抗特征:ImageNet 中的“统计对手”研究
Statistical Adversaries: Natural Backdoor-like Adversarial Features in Clean Vision Datasets
AI 导读
arXiv 一篇论文提出“统计对手”概念:视觉数据集中天然存在的统计信号可表现为后门式触发器,无需恶意注入。研究者在 ImageNet 中找到与特定标签强关联的模式,用统计控制去除随机相关性,并证明这些信号能直接、可预测地改变模型预测。这类信号比一般扰动更具针对性,且可跨模型架构迁移,说明漏洞源于数据集结构而非单个模型的特性,数据集审计应将其视为潜在攻击面。
正文 · 原文
Abstract:Model-specific adversarial attacks have been extensively studied. We study a different failure mode: naturally occurring statistical signals in vision data that can behave as backdoor-like triggers without being maliciously inserted. We call these signals statistical adversaries. We analyse ImageNet to find patterns that are strongly linked to certain labels. We then use statistical controls to remove random correlations from our candidate signals. Finally, we demonstrate that these signals directly and predictably alter model predictions. These statistical adversaries are more targeted than generic corruptions and transfer across different model architectures. This suggests that some vulnerabilities are driven by dataset structure and distribution rather than a single model's idiosyncrasies. We conclude that ordinary datasets can contain exploitable adversarial surfaces even in the absence of poisoning, and suggest that dataset audits should treat spurious structure not only as a source of bias or interpretability failure, but also as a latent attack surface for vision models.
| Comments: | Author metadata correction (Malatynski -> Malatyński). No changes in paper from previous version (v1) |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Machine Learning (cs.LG) |
| MSC classes: | 68T07, 68T45, 62H30, 62H35 |
| ACM classes: | I.2.6; I.2.10; I.4.8; I.5.2; I.5.4; G.3 |
| Cite as: | arXiv:2607.05516 [cs.CV] |
| (or arXiv:2607.05516v3 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2607.05516 arXiv-issued DOI via DataCite |
Submission history
From: Paul K. Mandal [view email]
[v1]
Mon, 6 Jul 2026 18:00:25 UTC (3,092 KB)
[v2]
Mon, 17 Aug 2026 12:49:25 UTC (4,381 KB)
[v3]
Tue, 17 Nov 2026 12:23:52 UTC (4,381 KB)
来源:arXiv cs.AI · arxiv.org