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 2402.07329cs.CV
收藏

视觉-语言模型中有害标签关联的偏见

The Bias of Harmful Label Associations in Vision-Language Models

Meta AI

Caner Hazirbas, Alicia Sun, Yonathan Efroni, Mark Ibrahim

AI总结:

本文基于含7万多个视频的 Casual Conversations 数据集评估主流 VLM,发现其有害标签关联存在显著肤色偏见,且模型扩大会增强有害预测置信度,标准视觉任务改进也无法消除差异。

摘要:

尽管基础视觉-语言模型取得了卓越性能,但文本与视觉共享的表示空间也可能编码不利于公平性的有害标签关联。尽管先前工作已经揭示了视觉-语言模型(VLM)在不同地域分类性能中的偏见,但由于缺乏丰富的带标注数据,相关研究在有害标签关联这一重要维度上仍然受限。本文研究了近期发布的、包含超过70,000个视频的 Casual Conversations 数据集中的有害标签关联。作者在多个主流 VLM 上,围绕年龄、性别、表观肤色和身体装饰物等自报标签,考察有害标签关联出现频率方面的偏见。结果发现,VLM 对肤色较深个体作出有害分类的可能性要高出4至7倍。研究还发现,扩大 transformer encoder 模型规模会提高有害预测的置信度。最后,各 VLM 在标准视觉任务上的性能提升并未解决有害标签关联中的差异问题。

英文摘要:

Despite the remarkable performance of foundation vision-language models, the shared representation space for text and vision can also encode harmful label associations detrimental to fairness. While prior work has uncovered bias in vision-language models' (VLMs) classification performance across geography, work has been limited along the important axis of harmful label associations due to a lack of rich, labeled data. In this work, we investigate harmful label associations in the recently released Casual Conversations datasets containing more than 70,000 videos. We study bias in the frequency of harmful label associations across self-provided labels for age, gender, apparent skin tone, and physical adornments across several leading VLMs. We find that VLMs are $4-7$x more likely to harmfully classify individuals with darker skin tones. We also find scaling transformer encoder model size leads to higher confidence in harmful predictions. Finally, we find improvements on standard vision tasks across VLMs does not address disparities in harmful label associations.

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。

↑