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 2607.02494cs.CVcs.CL
收藏

基于训练无关的概念定位实现对抗印刷体攻击的鲁棒性

Towards Robustness against Typographic Attack with Training-free Concept Localization

University of Virginia(弗吉尼亚大学)

Bohan Liu, Wenqian Ye, Guangzhi Xiong, Zhenghao He, Sanchit Sinha, Aidong Zhang

AI总结:

提出一种无需训练的可解释性方法,通过分析注意力头对词汇和语义的编码差异,定位并干预ViT中的词汇偏置电路,从而在不额外训练的情况下显著提升对印刷体攻击的鲁棒性。

摘要:

通过对比语言-图像预训练(CLIP)训练的模型作为大多数现代大型视觉语言模型(LVLM)的基础视觉编码器。尽管被广泛采用,CLIP模型表现出一个关键但未被充分探索的失败模式:图像中出现的无关文本会混淆视觉表示,使其偏向词汇意义而非真实视觉语义。这种鲁棒性问题通常被称为印刷体攻击(TA),暴露了一个对安全关键应用(如自动驾驶)构成重大风险的漏洞。为了实现可解释且有效的抗TA鲁棒性,我们提出了一种新颖的、无需训练的可解释性方法。该方法提供基于采样的隐藏状态表示解释,并定量地将语义与词汇关注度归因于各个注意力头。通过概率分析和电路挖掘,我们隔离了不成比例地编码词汇信息的特定Vision Transformer(ViT)组件,从而确定了TA的机制来源。我们进一步表明,直接应用于识别出的电路的简单干预(无需任何额外训练)可以显著提高目标分类中对印刷体攻击的鲁棒性。这些干预措施,例如选择性调整注意力权重,也优于有监督和无需训练的防御方法。我们的实验表明,将所提出的干预应用于几个最先进的LVLM的视觉编码器,在RIO-Bench上受到印刷体攻击干扰时,视觉问答准确性有显著提升。这些结果证实了我们机制方法的有效性和泛化性。代码发布在此https URL。

英文摘要:

Models trained via Contrastive Language-Image Pretraining (CLIP) serve as the foundational vision encoders for most modern Large Vision Language Models (LVLMs). Despite their widespread adoption, CLIP models exhibit a critical yet underexplored failure mode: irrelevant text appearing within images confounds visual representations, biasing them toward lexical meaning rather than true visual semantics. This robustness issue, commonly described as a Typographic Attack (TA), exposes a vulnerability that poses a significant risk to safety-critical applications such as autonomous driving. To achieve interpretable and effective robustness against TA, we propose a novel, training-free mechanistic interpretability method. Our method provides sampling-based interpretations of hidden state representations and quantitatively attributes semantic versus lexical focus to individual attention heads. Through probabilistic analysis and circuit mining, we isolate specific Vision Transformer (ViT) components that disproportionately encode lexical information, thereby identifying the mechanistic source of TA. We further show that simple interventions applied directly to the identified circuits, without any additional training, can substantially improve robustness against Typographic Attacks in object classification. These interventions, such as selective adjustment of attention weights, also outperform both supervised and training-free defense methods. Our experiments demonstrate that applying the proposed intervention to the vision encoders of several state-of-the-art LVLMs yields substantial gains in Visual Question Answering accuracy under Typographic Attack interference on RIO-Bench. These results confirm both the efficacy and the generalizability of our mechanistic approach. Code is released at https://github.com/Liu-524/SamplingTAR.

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。

↑