Research map / Vision-language and multimodal / Multimodal LLMs
Gender Bias in Vision-Language In-Context Learning
Tong Xiang, Yuta Nakashima, Noa Garcia
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 |
|---|---|---|---|
| use male-only demonstrations instead of random (ablation) | LVLM ICL · captioning and pronoun prediction | ER_m-f | Improved |
| use female-only demonstrations instead of random (ablation) | LVLM ICL · captioning and pronoun prediction | ER_m-f: -27.5 -> 27.2 (MiniCPM, VisoGender-OO 8-shot) | Hurt |
| use gendered (single-gender) demonstrations (ablation) | LVLM ICL · VisoGender-OP 8-shot | opposite-gender error rate: ER_f 9.7 (FS) -> 85.5 (MS), Qwen-VL | Hurt |
| use gendered demonstrations (ablation) | LVLM ICL · VQA (VisualCoT) | ER_m-f: differs by at most 1.82 across settings | No clear effect |
| use similarity-based retrieval (SIIR/SITR) instead of random selection | LVLM ICL · COCOBias captioning 8-shot | ER_m-f: -1.01 (RS) -> -2.25 (SIIR), -2.35 (SITR), Phi-3.5-vision | Hurt |
| use a gender-balanced pool for similarity retrieval (ablation) | SIIR/SITR · COCOBias captioning | ER_m-f | Mixed |
| use balanced demonstrations (BS) instead of random (ablation) | LVLM ICL · COCOBias captioning 8-shot | ER_m-f: -1.01 -> -0.75 (Phi-3.5-vision) | Improved |
| vary gender composition of demonstrations (ablation) | LVLM ICL · COCOBias captioning | CLIPScore/BLEU-4: CLIPScore range at most 0.67 | No clear effect |
| replace real in-context images with synthetic FLUX/SD3.5L images (ablation) | LVLM ICL · COCOBias captioning MS/FS | |ER_m-f| on revealed samples: Phi-3.5-vision MS 4-shot 2.24 -> gap -0.36 (FLUX) | Improved |
| replace real in-context images with synthetic images (ablation) | LVLM ICL · COCOBias captioning | CLIPScore: differences within 0.5 | No clear effect |
| replace real in-context images with synthetic images (ablation) | LVLM ICL · COCOBias captioning | reveal rate | Hurt |
Closest papers by meaning
- Revealing and Reducing Gender Biases in Vision and Language Assistants (VLAs) (ICLR 2025 · similarity 0.90)
- Are Models Biased on Text without Gender-related Language? (ICLR 2024 · similarity 0.88)
- Language Bias in LVLMs: From In-Depth Analysis to Simple and Effective Mitigation (ICML 2026 · similarity 0.87)
- See It from My Perspective: How Language Affects Cultural Bias in Image Understanding (ICLR 2025 · similarity 0.87)
- On Fairness of Unified Multimodal Large Language Model for Image Generation (NeurIPS 2025 · similarity 0.86)
- SocialCounterfactuals: Probing and Mitigating Intersectional Social Biases in Vision-Language Models with Counterfactual Examples (CVPR 2024 · similarity 0.86)
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.
