Web伪影攻击破坏视觉语言模型
Web Artifact Attacks Disrupt Vision Language Models
AI总结:
本文提出基于伪影的攻击,利用不匹配文本和图形元素误导视觉语言模型,在五个数据集上达到100%攻击成功率,并扩展防御提示以降低攻击成功率。\n
摘要:
视觉语言模型(VLMs)(如CLIP、LLaVA)在大规模、轻度筛选的网络数据集上进行训练,导致它们学习到语义概念与无关视觉信号之间的意外关联。这些关联会使得模型预测依赖偶然模式而非真正的视觉理解,从而降低模型准确性。先前的工作已将这些关联武器化为一种攻击向量来操纵模型预测,例如在图像上插入欺骗性类别文本的“排版”攻击。这些攻击之所以成功,是因为VLM具有文本偏置——这是由标题回显可见文字而非描述内容所导致的。然而,这种攻击仅专注于与目标类别完全匹配的文本,忽视了更广泛的关联,包括不匹配的文本和图形符号,这些关联源于网络规模数据中大量品牌内容。为弥补这一空白,我们引入了“基于伪影”的攻击:一种新型操纵类别,利用不匹配的文本和图形元素来误导模型。与排版攻击不同,这些伪影并非预定义的,使得它们同时更难防御和更难发现。我们通过将伪影攻击构建为搜索问题来解决这一挑战,并在五个数据集上展示了其有效性,其中一些伪影相互增强,达到了100%的攻击成功率。这些攻击在模型间转移的有效性高达90%,使得攻击未见过的模型成为可能。为防御这些攻击,我们将先前工作的伪影感知提示扩展到图形设置。我们观察到,相对于标准提示,攻击成功率适度降低了最多15%,这表明了增强模型鲁棒性的一个有前景的方向。代码:https://github.com/mqraitem/Web-Artifact-Attacks
英文摘要:
Vision-language models (VLMs) (e.g. CLIP, LLaVA) are trained on large-scale, lightly curated web datasets, leading them to learn unintended correlations between semantic concepts and unrelated visual signals. These associations degrade model accuracy by causing predictions to rely on incidental patterns rather than genuine visual understanding. Prior work has weaponized these correlations as an attack vector to manipulate model predictions, such as inserting a deceiving class text onto the image in a "typographic" attack. These attacks succeed due to VLMs' text-heavy bias-a result of captions that echo visible words rather than describing content. However, this attack has focused solely on text that matches the target class exactly, overlooking a broader range of correlations, including non-matching text and graphical symbols, which arise from the abundance of branding content in web-scale data. To address this gap, we introduce "artifact-based" attacks: a novel class of manipulations that mislead models using both non-matching text and graphical elements. Unlike typographic attacks, these artifacts are not predefined, making them simultaneously harder to defend against and more challenging to find. We address this by framing artifact attacks as a search problem and demonstrate their effectiveness across five datasets, with some artifacts reinforcing each other to reach 100% attack success rates. These attacks transfer across models with up to 90% effectiveness, making it possible to attack unseen models. To defend against these attacks, we extend prior work's artifact aware prompting to the graphical setting. We see a moderate reduction of success rates of up to 15% relative to standard prompts, suggesting a promising direction for enhancing model robustness. Code: https://github.com/mqraitem/Web-Artifact-Attacks
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。