Research map / Learning theory and algorithms / Fairness
Fairness Reprogramming
Guanhua Zhang, Yihua Zhang, Yang Zhang, Wenqi Fan, Qing Li, Sijia Liu, Shiyu Chang
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 fairness reprogramming (Border) instead of best post-processing baseline | FairReprogram · CelebA EO | accuracy at similar negative EO: 0.951 vs 0.927 | Improved |
| use FairReprogram (Soft) instead of best post-processing baseline | FairReprogram · Civil Comments EO | negative EO and accuracy: -0.031 -> -0.011 | Improved |
| decrease number of trigger words to one (ablation) | FairReprogram Soft · Civil Comments | negative DP bias: -0.002 -> -0.011 | Hurt |
| keep a one-word trigger versus no trigger (ablation) | FairReprogram · Civil Comments | negative bias score | Improved |
| use five or more trigger words (ablation) | FairReprogram · Civil Comments | negative bias score | No clear effect |
| increase image trigger size (ablation) | FairReprogram · CelebA | negative bias score: -0.081 -> -0.054 | Improved |
| increase image trigger size (ablation) | FairReprogram · CelebA | accuracy: 0.951 -> 0.914 | Hurt |
| reduce tuning data to 1 percent | FairReprogram · Civil Comments and CelebA | negative bias score | No clear effect |
| reduce tuning data to very small | AdvPost · tuning data ratio | negative bias score | Hurt |
| transfer trigger to model trained with different seed | FairReprogram · Civil Comments and CelebA | fairness-accuracy trade-off | No clear effect |
| increase number of classes in multi-class setting | AdvPost · CelebA multi-class | fairness-accuracy trade-off | Hurt |
| replace adversarial loss with MMD regularization | MmdIn/MmdPost · Civil Comments | fairness-accuracy trade-off | Hurt |
Closest papers by meaning
- FairBatch: Batch Selection for Model Fairness (ICLR 2021 · similarity 0.88)
- Unlocking Fairness: a Trade-off Revisited (NeurIPS 2019 · similarity 0.88)
- FairNet: Dynamic Fairness Correction without Performance Loss via Contrastive Conditional LoRA (NeurIPS 2025 · similarity 0.87)
- Fairness without Harm: An Influence-Guided Active Sampling Approach (NeurIPS 2024 · similarity 0.87)
- Adversarial Fairness Network (AAAI 2024 · similarity 0.86)
- Achievable Fairness on Your Data With Utility Guarantees (NeurIPS 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.
