arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2312.10065cs.CYcs.AI
收藏

探究文生图基础模型下游应用中的社会偏见

Exploring Social Bias in Downstream Applications of Text-to-Image Foundation Models

University of Cambridge(剑桥大学);California Institute of Technology(加州理工学院);Carleton College(卡尔顿学院);Nvidia(英伟达)

Adhithya Prakash Saravanan, Rafal Kocielnik, Roy Jiang, Pengrui Han, Anima Anandkumar

AI总结:

针对现有研究多聚焦文生图扩散模型生成场景偏见、下游应用偏见探究不足的问题,提出用合成图像探测图像编辑与分类场景偏见的方法,在Stable Diffusion中发现显著交叉社会偏见,警示盲目下游应用的风险。

摘要:

文生图扩散模型已被应用于艺术生成、图像编辑等核心商业工作流。明确这类模型呈现的性别、种族刻板印象等内隐社会偏见,是避免歧视性结果的必要前提。现有社会偏见研究多聚焦图像生成场景,基于扩散的基础模型在其他应用中表现出的偏见仍有待充分探究。我们提出利用合成图像的方法,针对扩散模型的两大应用——图像编辑与分类——开展社会偏见探测。通过该方法,我们在前沿开源文生图模型Stable Diffusion(稳定扩散)中发现了显著且具有实质意义的交叉性社会偏见。研究结果警示,不可在未充分了解风险的情况下将文生图基础模型用于下游任务与服务。

英文摘要:

Text-to-image diffusion models have been adopted into key commercial workflows, such as art generation and image editing. Characterising the implicit social biases they exhibit, such as gender and racial stereotypes, is a necessary first step in avoiding discriminatory outcomes. While existing studies on social bias focus on image generation, the biases exhibited in alternate applications of diffusion-based foundation models remain under-explored. We propose methods that use synthetic images to probe two applications of diffusion models, image editing and classification, for social bias. Using our methodology, we uncover meaningful and significant inter-sectional social biases in \textit{Stable Diffusion}, a state-of-the-art open-source text-to-image model. Our findings caution against the uninformed adoption of text-to-image foundation models for downstream tasks and services.

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。

↑