Research map / Learning theory and algorithms / Fairness
FairMerging: Rethinking Model Merging through the Lens of Fairness
Bing Liu, Xinrui Shan, Boyu Zhang, Qiankun Zhang, Bin Yuan, Jing Wang, Xianjun Deng
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 Stage 1 sensitivity-control fine-tuning of the target model (ablation) | Task Arithmetic · Six-task merging avg | fairness gap F(D): 2.14 -> 1.78 | Improved |
| add Stage 1 sensitivity-control fine-tuning of the target model (ablation) | Ties-Merging · Six-task merging avg | fairness gap F(D): 2.11 -> 1.85 | Improved |
| add Stage 1 sensitivity-control fine-tuning of the target model (ablation) | DARE · Six-task merging avg | fairness gap F(D): 2.38 -> 2.17 | Improved |
| add Stage 1 sensitivity-control fine-tuning of the target model (ablation) | MetaGPT · Six-task merging avg | fairness gap F(D): 1.66 -> 1.43 | Improved |
| add Stage 1 sensitivity-control fine-tuning of the target model (ablation) | AdaMerging · Six-task merging avg | fairness gap F(D): 1.49 -> 1.26 | Improved |
| add Stage 2 fairness-aware coefficient optimization with orthogonal task vectors (ablation) | FairMerging without Stage 1 · Six-task merging avg | fairness gap F(D): 1.49 -> 0.58 | Improved |
| add Stage 1 on top of Stage 2 (full FairMerging) (ablation) | FairMerging Stage 2 · Six-task merging avg | fairness gap F(D): 0.58 -> 0.42 | Improved |
| replace AdaMerging with full FairMerging | AdaMerging · Six-task merging avg | fairness gap F(D): 1.49 -> 0.42 | Improved |
| replace AdaMerging with full FairMerging | AdaMerging · UTK-Face | fairness gap F(D): 2.62 -> 0.57 | Improved |
| replace AdaMerging with full FairMerging | AdaMerging · Six-task merging avg | accuracy (%): 79.25 -> 75.04 | Hurt |
| replace Task Arithmetic with full FairMerging | Task Arithmetic · Six-task merging avg | accuracy (%): 63.63 -> 75.04 | Improved |
Closest papers by meaning
- FairMT: Fairness for Heterogeneous Multi-Task Learning (NeurIPS 2026 · similarity 0.90)
- Revisiting Fairness in Multitask Learning: A Performance-Driven Approach for Variance Reduction (CVPR 2025 · similarity 0.89)
- FairNet: Dynamic Fairness Correction without Performance Loss via Contrastive Conditional LoRA (NeurIPS 2025 · similarity 0.87)
- Some Optimizers are More Equal: Understanding the Role of Optimizers in Group Fairness (NeurIPS 2025 · similarity 0.87)
- Stay Fair! Ensuring Group Fairness in Diffusion Models Across Guidance Scales (NeurIPS 2026 · similarity 0.87)
- FRAPPÉ: A Group Fairness Framework for Post-Processing Everything (ICML 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.
