Research map / Learning theory and algorithms / Fairness
Safe Fairness Guarantees Without Demographics in Classification: Spectral Uncertainty Set Perspective
Ainhize Barrainkua, Santiago Mazuelas, Novi Quadrianto, Jose Antonio Lozano
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 |
|---|---|---|---|
| tune hyperparameters with WCE+T+A instead of WCE (ablation) | SPECTRE · COMPAS (race) | worst-group accuracy (%): 56.1 -> 59.0 | Improved |
| tune hyperparameters with TOPN+WCE instead of ACC (ablation) | SPECTRE · COMPAS (race) | worst-group accuracy (%): 56.0 -> 53.9 | Hurt |
| tune hyperparameters with ACC instead of WCE (ablation) | SPECTRE · ACSEmployment state CT | worst-group accuracy (%): 61.7 -> 62.7 | No clear effect |
| use SPECTRE instead of XGBoost | XGBoost · ACS datasets, 20 states | worst-group accuracy: fairness gains up to 20-40% with accuracy drop of 1-4% | Improved |
| use SPECTRE instead of XGBoost | XGBoost · ACSEmployment state CT | worst-group accuracy (%): 58.0 -> 62.7 | Improved |
| use SPECTRE (WCE+T+A) instead of ARL | ARL · COMPAS (race) | worst-group accuracy (%): 55.3 -> 59.0 | Improved |
| use SPECTRE instead of LR | LR · COMPAS (race) | worst-group accuracy (%): 59.5 -> 59.0 | No clear effect |
| evaluate intersectional instead of single-attribute groups on German Credit | ARL, LR, XGB · German Credit | worst-group accuracy: decrease of approximately 20-30% | Hurt |
| use random Fourier feature mapping instead of polynomial degree 9 mapping (ablation) | MRC · toy dataset | worst-group accuracy | Improved |
| increase sigma from small to very large values (ablation) | SPECTRE · toy dataset | worst-group accuracy | Mixed |
Closest papers by meaning
- Fairness guarantees in multi-class classification with demographic parity (JMLR 2024 · similarity 0.90)
- Fairness risk measures (ICML 2019 · similarity 0.89)
- Blind Pareto Fairness and Subgroup Robustness (ICML 2021 · similarity 0.88)
- Towards Harmless Rawlsian Fairness Regardless of Demographic Prior (NeurIPS 2024 · similarity 0.88)
- Geometry of Relaxed Fair Regression: A Unified Framework for Aware and Unaware Settings (NeurIPS 2026 · similarity 0.87)
- Utility-Fairness Trade-Offs and How to Find Them (CVPR 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.
