无监督提取性别概念向量，精准控制大模型偏见。
Unsupervised Concept Vector Extraction for Bias Control in LLMs
- 用概率加权无监督提取概念向量，无需标签数据。
- 通过投影方法实现对模型输出的精确调控，有效降低性别与种族偏见。
- 方法通用性强，适合需可控生成的场景研究者使用。
大型语言模型（LLMs）常延续刻板印象并表现出偏见。现有方法多将偏见视为黑箱问题，未关注概念在模型中的表征方式。本文借鉴表示工程方法，研究模型中'性别'概念的表征。提出一种无需标注数据的概率加权方法，提取概念向量，并高效选取用于测量和操控模型表征的调节向量。开发基于投影的方法，实现对模型预测的精准调控，在缓解性别偏见方面表现有效，并可推广至种族偏见。代码已开源：https://github.com/hannahxchen/gender-bias-steering。
原文摘要 · Abstract (English)
Large language models (LLMs) are known to perpetuate stereotypes and exhibit biases. Various strategies have been proposed to mitigate these biases, but most work studies biases as a black-box problem without considering how concepts are represented within the model. We adapt techniques from representation engineering to study how the concept of "gender" is represented within LLMs. We introduce a new method that extracts concept representations via probability weighting without labeled data and efficiently selects a steering vector for measuring and manipulating the model's representation. We develop a projection-based method that enables precise steering of model predictions and demonstrate its effectiveness in mitigating gender bias in LLMs and show that it also generalizes to racial bias. Our code is available at: https://github.com/hannahxchen/gender-bias-steering
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。