用多视角融合让部署后的大模型更公平，不需重训也能调偏。
MPF: Aligning and Debiasing Language Models post Deployment via Multi Perspective Fusion
- 通过分解人类专家观点生成多视角输出，引导模型对齐真实偏好。
- 使模型情感分布与基准的KL散度小，校准误差降低，能泛化到新问题。
- 无需微调或提示工程，适合已上线模型的公平性优化。
多视角融合（MPF）是一种针对大型语言模型（LLMs）的后训练对齐框架，旨在应对日益增长的偏见缓解需求。基于SAGED管道——一个自动构建偏见基准并提取可解释基线分布的系统，MPF利用多视角生成来揭示并使模型输出中的偏见与细腻、类人化的基线对齐。通过将基线（如来自人力资源专业人士的情感分布）分解为可解释的视角成分，MPF通过采样和平衡响应，并根据分解中获得的概率加权生成。实证表明，MPF能够使模型的情感分布同时对齐反事实基线（绝对平等）和人力资源基线（偏向顶尖大学），实现较小的KL散度，减少校准误差，并泛化至未见过的问题。这表明MPF提供了一种可扩展且可解释的对齐与偏见缓解方法，兼容已部署的LLMs，且无需大量提示工程或微调。
原文摘要 · Abstract (English)
Multiperspective Fusion (MPF) is a novel posttraining alignment framework for large language models (LLMs) developed in response to the growing need for easy bias mitigation. Built on top of the SAGED pipeline, an automated system for constructing bias benchmarks and extracting interpretable baseline distributions, MPF leverages multiperspective generations to expose and align biases in LLM outputs with nuanced, humanlike baselines. By decomposing baseline, such as sentiment distributions from HR professionals, into interpretable perspective components, MPF guides generation through sampling and balancing of responses, weighted by the probabilities obtained in the decomposition. Empirically, we demonstrate its ability to align LLM sentiment distributions with both counterfactual baselines (absolute equality) and the HR baseline (biased for Top Univeristy), resulting in small KL divergence, reduction of calibration error and generalization to unseen questions. This shows that MPF offers a scalable and interpretable method for alignment and bias mitigation, compatible with deployed LLMs and requiring no extensive prompt engineering or finetuning.
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。