Societal biases in vision and language applications

视觉和语言应用中的社会偏见

• 批准号: 22K12091
• 负责人: GARCIA・DOCAMPO NOA
• 金额: $ 2.75万
• 依托单位: Osaka University
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Scientific Research (C)
• 财政年份: 2022
• 资助国家: 日本
• 起止时间: 2022-04-01 至 2026-03-31
• 项目状态: 未结题
• 来源: https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-22K12091/
• 关键词: bias in computer vision; computer vision; image captioning; vision and language; ethical ai; bias in machine learning; fairness

In this project, we have investigated the problem of societal bias in image captioning and multimodal vision and language models. As an outcome of the first year of research:* We have shown that captioning models encode gender and racial bias.* We have proposed a new metric to measure societal bias.* We have annotated a dataset to study and mitigate societal bias.* We have designed a bias mitigation method for image captioning.

在这个项目中，我们研究了图像字幕以及多模态视觉和语言模型中的社会偏见问题。作为第一年研究的成果：*我们已经证明字幕模型编码了性别和种族偏见。*我们提出了一种新的衡量社会偏见的指标。*我们注释了一个数据集来研究和减轻社会偏见。*我们设计了一种用于图像字幕的偏见缓解方法。

项目成果

Gender and Racial Bias in Visual Question Answering Datasets

视觉问答数据集中的性别和种族偏见

• 发表时间: 2022
• 作者: Chihiro Hoshizawa;Takashi Komuro;Yusuke Hirota
• 通讯作者: Yusuke Hirota

Societal biases in vision and language

视觉和语言方面的社会偏见

• 发表时间: 2022
• 作者: Chihiro Hoshizawa;Takashi Komuro;Yusuke Hirota;Noa Garcia
• 通讯作者: Noa Garcia