发表机构
Magellan Technology Research Institute (MTRI); National University of Singapore（麦哲伦技术研究所（MTRI）; 新加坡国立大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本研究提出GGSS方法，针对生成式VLMs设计推理时去偏方案，经实验验证其在降低人口统计学偏见的同时，能较好保留模型通用视觉语言能力。
AI 中文摘要
生成式视觉语言模型（VLMs）越来越多地用于以人为本的场景，但即使图像仅在感知种族或性别等受控属性上存在差异，它们也可能产生具有人口统计学偏见的输出。然而，现有的推理时去偏器主要是为静态嵌入或CLIP类模型设计的，而非针对生成式VLMs。我们提出GGSS——测地线门控球面引导（Geodesic-Gated Spherical Steering），这是一种保持范数的干预方法，它在单位超球面上发现反事实偏见子空间，沿测地线弧引导视觉标记，并使用自适应门将校正聚焦于携带更强人口统计学信号的标记。我们在单操作点协议下，针对分类、成对和职业-性别偏见测试，评估了四个生成式VLMs与十个适配的推理时去偏基线及基于提示的缓解方法，同时还测量了通用视觉语言能力。GGSS在所有四个模型上实现了最低的平均偏见，在配对置换检验中，四个骨干网络中的三个表现出显著效果，同时将MMStar准确率保持在未引导基线的±0.6个百分点范围内。代码可在this https URL获取。
英文摘要
Generative vision-language models (VLMs) are increasingly used in human-centered settings, yet they can produce demographically biased outputs even when images differ only in controlled attributes such as perceived race or gender. However, existing inference-time debiasers were largely designed for static embeddings or CLIP-like models rather than generative VLMs. We propose GGSS---Geodesic-Gated Spherical Steering---a norm-preserving intervention that discovers a counterfactual bias subspace on the unit hypersphere, steers visual tokens along geodesic arcs, and uses an adaptive gate to focus correction on tokens that carry stronger demographic signal. We evaluate four generative VLMs against ten adapted inference-time debiasing baselines and prompt-based mitigation under a single operating-point protocol across categorical, pairwise, and occupation-gender bias tests, while also measuring general visual-language capability. GGSS achieves the lowest average bias on all four models, significant on three of four backbones under paired permutation tests, while preserving MMStar accuracy within +/- 0.6 p.p. of the unsteered baseline. Code is available at https://github.com/dukesun99/GGSS.
CommentsAccepted to EMNLP 2026
