analiza aiqbench
GGSS introduces a novel inference-time intervention that corrects demographic bias in generative vision-language models by steering visual tokens along geodesic arcs on a unit hypersphere. This approach effectively reduces bias across multiple model backbones while maintaining near-baseline performance, addressing a critical gap where previous debiasers failed to handle the dynamic generation process of VLMs. The method offers a practical pathway for deploying more equitable AI systems in human-centered applications without sacrificing visual-language capabilities.
GGSS: Geodesic-Gated Spherical Steering for Inference-Time Debiasing of Generative Vision-Language Models
arXiv:2608.25375v1 Announce Type: cross Abstract: 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.
Abstract: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 this https URL.
Computers and Society (cs.CY); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
arXiv:2608.25375 [cs.CY]
(or arXiv:2608.25375v1 [cs.CY] for this version)
https://doi.org/10.48550/arXiv.2608.25375 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 262 cuvinte
Subiecte conexe
Gated Activation Steering for Reducing Sycophancy & Hallucination in Medical Question Answering
arXiv · 27 aug. 2026
[AINews] Claude Opus 5.5, the new default model for AINews — and everybody cuts prices 40-50%
Latent Space · 23 sept. 2026
Query-Adaptive Neuron Steering for LLMs
arXiv · 14 oct. 2026
The Density of Thought: Predicting and Steering Reasoning in Large Language Models
arXiv · 17 oct. 2026