Research map / Vision-language and multimodal / Human and embodied modeling
WRING Out The Bias: A Rotation-Based Alternative To Projection Debiasing
Walter Gerych, Cassandra Parent, Quinn Perian, Rafiya Javed, Justin Solomon, Marzyeh Ghassemi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace projection with rotation (WRING) | CLIP debiasing · CelebA worst-group accuracy, Gender | worst group accuracy: 78.89 -> 80.56 | Improved |
| replace projection with rotation (WRING) | CLIP debiasing · CelebA accuracy gap, Gender | accuracy gap: 11.87 -> 9.24 | Improved |
| replace projection with rotation (WRING) | CLIP debiasing · CelebA worst-group accuracy, Age | worst group accuracy: 85.14 -> 85.59 | Improved |
| replace projection with rotation (WRING) | CLIP debiasing · CelebA accuracy gap, Skin Tone | accuracy gap: 6.36 -> 8.88 | Hurt |
| replace projection with rotation (WRING) | CLIP debiasing · CelebA accuracy gap, Nose Shape | accuracy gap: 7.33 -> 5.22 | Improved |
| use WRING instead of FairerCLIP | CLIP debiasing · CelebA worst-group accuracy, Age | worst group accuracy: 82.66 -> 85.59 | Improved |
| debias gender with FairerCLIP instead of nothing | CLIP baseline · CelebA worst-group accuracy, Age | worst group accuracy: 87.42 -> 82.66 | Hurt |
| define concept directions with image embeddings instead of text (ablation) | Projection and WRING · bias debiasing across datasets | bias for Cdebias | Improved |
| rotation instead of projection in nonlinear debiasing pipeline (ablation) | nonlinear VLM debiasing · FairFace | % change in bias for Cuncon | Improved |
| SFID instead of projection | debiasing · bias for target concept | bias reduction for Cdebias | Hurt |
Closest papers by meaning
- BendVLM: Test-Time Debiasing of Vision-Language Embeddings (NeurIPS 2024 · similarity 0.92)
- Joint Vision-Language Social Bias Removal for CLIP (CVPR 2025 · similarity 0.90)
- A Unified Debiasing Approach for Vision-Language Models across Modalities and Tasks (NeurIPS 2024 · similarity 0.90)
- Language Bias in LVLMs: From In-Depth Analysis to Simple and Effective Mitigation (ICML 2026 · similarity 0.88)
- Discovering and Mitigating Visual Biases through Keyword Explanation (CVPR 2024 · similarity 0.88)
- Two Effects, One Trigger: On the Modality Gap, Object Bias, and Information Imbalance in Contrastive Vision-Language Models (ICLR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
