Research map / Learning theory and algorithms / Fairness
Task-Tailored Pre-Processing: Fair Downstream Supervised Learning
Jinwon Sohn, Qifan Song, Guang Lin
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 delta_Y=0.1 (modify labels) instead of delta_Y=0 (ablation) | Task-tailored pre-processing · ACSPublicCoverage | AUC-MP (Budget 1): .715 -> .804 | Improved |
| use delta_Y=0.1 instead of 0 (ablation) | Task-tailored pre-processing · ACSPublicCoverage | AUC-RF (Budget 1): .731 -> .816 | Improved |
| use delta_Y=0.1 instead of 0 (ablation) | Task-tailored pre-processing · ACSPublicCoverage | consistency of utility sigma_AUC-SP (Budget 1): .023 -> .015 | Improved |
| use delta_Y=0.1 instead of 0 (ablation) | Task-tailored pre-processing · ACSPublicCoverage | consistency of fairness sigma_SP (Budget 2): .292 -> .207 | Improved |
| use delta_Y=0.1 instead of 0 (ablation) | Task-tailored pre-processing · ACSPublicCoverage | sigma_KS-SP (Budget 2): .123 -> .087 | Improved |
| use Ours instead of FAAP for downstream fairness | FAAP · CelebA Blond Hair (delta_X=.010) | downstream Delta EO: approx 0 -> .229 | Improved |
| use Ours instead of FAAP for downstream fairness | FAAP · CelebA Blond Hair (delta_X=.015) | downstream Delta EO: -.067 -> .231 | Improved |
| raise fairness budget delta_X from 0 to .010 (ablation) | Task-tailored pre-processing · CelebA Smiling | P(h=1|A=1): .397 -> .427 | Improved |
| raise fairness budget delta_X from 0 to .010 (female group) (ablation) | Task-tailored pre-processing · CelebA Smiling | P(h=1|A=0): .531 -> .511 | Improved |
| raise delta_X from 0 to .015 (ablation) | Task-tailored pre-processing · CelebA Blond Hair | P_{1,1}(Y=1): .663 -> .784 | Improved |
| raise delta_X from 0 to .015 (ablation) | Task-tailored pre-processing · CelebA Blond Hair | AUC (upstream): .975 -> .963 | Hurt |
| use task-tailored pre-processing instead of data-fairness approaches | Task-tailored pre-processing · tabular benchmarks | AUC vs SP trade-off | Improved |
Closest papers by meaning
- fairret: a Framework for Differentiable Fairness Regularization Terms (ICLR 2024 · similarity 0.89)
- FairWASP: Fast and Optimal Fair Wasserstein Pre-processing (AAAI 2024 · similarity 0.88)
- FairMT: Fairness for Heterogeneous Multi-Task Learning (NeurIPS 2026 · similarity 0.88)
- FRAPPÉ: A Group Fairness Framework for Post-Processing Everything (ICML 2024 · similarity 0.87)
- FairNet: Dynamic Fairness Correction without Performance Loss via Contrastive Conditional LoRA (NeurIPS 2025 · similarity 0.87)
- Kernel-based Equalized Odds: A Quantification of Accuracy-Fairness Trade-off in Fair Representation Learning (NeurIPS 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.
