Research map / Image generation / Text-to-image diffusion
Equalized Generative Treatment: Matching f-divergences for Fairness in Generative Models
Alexandre Verine, Rafael Pinot, Florian Le Bronnec
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 |
|---|---|---|---|
| add Min-Max fine-tuning of worst-group divergence | EDM-VP pretrained · FFHQ | delta-PR: 2.2 -> 0.3 | Improved |
| add Min-Max fine-tuning | EDM-VP pretrained · FFHQ | delta-P and delta-R: 2.0/4.2 -> 0.1/0.2 | Improved |
| add Min-Max fine-tuning | EDM-VE pretrained · FFHQ | delta-PR: 2.3 -> 0.8 | Improved |
| add conditional training | EDM-VP pretrained · FFHQ | delta-PR: 2.2 -> 5.2 | Hurt |
| add group reweighting | EDM-VP pretrained · FFHQ | delta-PR: 2.2 -> 0.6 | Improved |
| add conditional training | EDM-VE pretrained · FFHQ | delta-PR: 2.3 -> 3.5 | Hurt |
| add Min-Max fine-tuning | LLaMA-3.2-Chat 1B · Wikipedia Biographies | delta-PR: 32.65 -> 27.28 | Improved |
| add Min-Max fine-tuning | Gemma-3 4B (it) · Wikipedia Biographies | delta-PR: 28.81 -> 27.93 | Improved |
| add conditional training | Gemma-3 4B (pt) · Wikipedia Biographies | delta-PR: 33.03 -> 34.25 | Hurt |
| add conditional training | Gemma-3 4B (it) · Wikipedia Biographies | delta-PR: 28.81 -> 32.57 | Hurt |
| add Min-Max fine-tuning (cost to strongest group) | EDM-VP · FFHQ | recall of strongest subgroup: 91.7 -> 86.6 | Hurt |
| add conditional training | EDM-VP · FFHQ | overall FID: 2.34 -> 2.75 | Hurt |
Closest papers by meaning
- AIM-Fair: Advancing Algorithmic Fairness via Selectively Fine-Tuning Biased Models with Contextual Synthetic Data (CVPR 2025 · similarity 0.90)
- FairRAG: Fair Human Generation via Fair Retrieval Augmentation (CVPR 2024 · similarity 0.90)
- FairImagen: Post-Processing for Bias Mitigation in Text-to-Image Models (NeurIPS 2025 · similarity 0.89)
- Fair Text-to-Image Diffusion via Fair Mapping (AAAI 2025 · similarity 0.88)
- On Optimal Steering to Achieve Exact Fairness (NeurIPS 2025 · similarity 0.88)
- All Should Be Equal in the Eyes of LMs: Counterfactually Aware Fair Text Generation (AAAI 2024 · 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.
