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 2601.12430cs.CL
收藏

系统介导的注意力失衡使视觉-语言模型倾向于说‘是’

System-Mediated Attention Imbalances Make Vision-Language Models Say Yes

Saarland Informatics Campus, Saarland University, Germany(萨尔兰州信息学校区,萨尔兰州大学,德国);Max Planck Institute for Informatics, Germany(马克斯·普朗克信息研究所,德国)

Tsan Tsai Chan, Varsha Suresh, Anisha Saha, Michael Hahn, Vera Demberg

AI总结:

本文研究了视觉-语言模型中系统介导的注意力失衡对‘是’偏见的影响,提出通过重新分配注意力以减少这种偏见,从而提升模型可靠性。

摘要:

视觉-语言模型(VLM)幻觉通常与输入模态间注意力分配不平衡有关:系统、图像和文本。然而,现有缓解策略多倾向于图像中心的解释,往往优先增加图像注意力而忽视其他模态的作用。本文评估了更全面的系统介导解释,认为这些不平衡源于功能冗余的系统权重,减少了对图像和文本输入的关注。研究显示,该框架为理解‘是’偏见(VLMs indiscriminately respond 'yes')提供了有用的实证视角。因果性地将注意力从系统模态转移到图像和文本输入可显著抑制此偏见,常优于现有方法。进一步证据表明,系统介导的注意力失衡通过鼓励对粗粒度输入表示的默认依赖,导致某些任务有效但其他任务不适用。综合来看,这些发现牢固确立了系统注意力在VLM幻觉中的关键作用,并突显其作为缓解手段的潜力。

英文摘要:

Vision-language model (VLM) hallucination is commonly linked to imbalanced allocation of attention across input modalities: system, image and text. However, existing mitigation strategies tend towards an image-centric interpretation of these imbalances, often prioritising increased image attention while giving less consideration to the roles of the other modalities. In this study, we evaluate a more holistic, system-mediated account, which attributes these imbalances to functionally redundant system weights that reduce attention to image and textual inputs. We show that this framework offers a useful empirical perspective on the yes-bias, a common form of hallucination in which VLMs indiscriminately respond `yes'. Causally redistributing attention from the system modality to image and textual inputs substantially suppresses this bias, often outperforming existing approaches. We further present evidence suggesting that system-mediated attention imbalances contribute to the yes-bias by encouraging a default reliance on coarse input representations, which are effective for some tasks but ill-suited to others. Taken together, these findings firmly establish system attention as a key factor in VLM hallucination and highlight its potential as a lever for mitigation.

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。

↑