Search papers, labs, and topics across Lattice.
This paper introduces GGSS, a novel debiasing technique for generative vision-language models that addresses demographic biases in outputs by steering visual tokens along geodesic arcs on a unit hypersphere. The method focuses on tokens with stronger demographic signals using an adaptive gating mechanism, enabling effective bias correction without compromising model accuracy. Evaluation against ten baseline debiasing methods demonstrates that GGSS consistently achieves the lowest average bias across four generative VLMs while maintaining performance within 卤0.6 percentage points of the unsteered baseline.
GGSS reduces demographic bias in generative vision-language models while preserving accuracy, outperforming existing debiasing methods.
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.
