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 2409.16371cs.CL
收藏

做正确的事,只需去偏!使用 LLMs 进行多类别偏见缓解

Do the Right Thing, Just Debias! Multi-Category Bias Mitigation Using LLMs

Indian Institute of Science Education and Research (IISER), Kolkata(印度科学教育与研究学院(加尔各答));Bosch India(博世印度公司);Manipal University Jaipur(斋浦尔马尼帕尔大学);University of Manchester(曼彻斯特大学);JTP Japan(JTP日本公司);A. P. Shah Institute of Technology(A. P. Shah理工学院)

Amartya Roy, Danush Khanna, Devanshu Mahapatra, Vasanthakumar, Avirup Das, Kripabandhu Ghosh

AI总结:

本文针对语言偏见缓解模型的泛化挑战,提出了包含1507对句子、覆盖9类社会偏见的新数据集ANUBIS,并利用SFT、PPO、DPO和ICL微调T5等模型进行多类别偏见缓解实验,为构建公平AI系统提供了重要资源。

摘要:

本文探讨了为语言构建稳健且可泛化的偏见缓解模型的挑战。认识到现有数据集的局限性,我们引入了 ANUBIS,这是一个包含 1507 对精心策划的句子对的新颖数据集,涵盖九个社会偏见类别。我们评估了如 T5 等最先进的模型,利用监督微调(SFT)、强化学习(PPO、DPO)和上下文学习(ICL)来进行有效的偏见缓解。我们的分析聚焦于多类别社会偏见减少、跨数据集泛化能力以及训练模型的环境影响。ANUBIS 和我们的发现为构建更公平的 AI 系统提供了宝贵资源,并有助于开发具有广泛社会影响的负责任且无偏见的技术。

英文摘要:

This paper tackles the challenge of building robust and generalizable bias mitigation models for language. Recognizing the limitations of existing datasets, we introduce ANUBIS, a novel dataset with 1507 carefully curated sentence pairs encompassing nine social bias categories. We evaluate state-of-the-art models like T5, utilizing Supervised Fine-Tuning (SFT), Reinforcement Learning (PPO, DPO), and In-Context Learning (ICL) for effective bias mitigation. Our analysis focuses on multi-class social bias reduction, cross-dataset generalizability, and environmental impact of the trained models. ANUBIS and our findings offer valuable resources for building more equitable AI systems and contribute to the development of responsible and unbiased technologies with broad societal impact.

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。

↑