Research map / Learning theory and algorithms / Fairness
Improving Fair Training under Correlation Shifts
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Pre-processing for correlation shift (sampling to adjust (y,z) class ratios) before in-processing | FairBatch (FB) · Synthetic test (corr. 50% of train), DP | DP unfairness: 0.048 -> 0.003 | Helped |
| Pre-processing for correlation shift before in-processing | Fairness Constraints (FC) · Synthetic test, DP | DP unfairness: 0.038 -> 0.034 | Helped |
| Pre-processing for correlation shift before in-processing | Adversarial Debiasing (AD) · Synthetic test, DP | DP unfairness: 0.032 -> 0.017 | Helped |
| Pre-processing for correlation shift before in-processing | FairBatch, FC, AD · Synthetic test, DP | accuracy: FB +0.015, FC +0.071, AD +0.052 (ours higher); Ours+AD lower than RW | Mixed |
| Reweighing (RW) pre-processing before in-processing | FairBatch · Synthetic test, DP | DP unfairness: 0.048 -> 0.055 | Hurt |
| Pre-processing for correlation shift before in-processing | FairBatch, DP single metric · COMPAS test | DP unfairness: 0.038 -> 0.027 | Helped |
| Pre-processing for correlation shift before in-processing | Adversarial Debiasing, DP single metric · COMPAS test | DP unfairness: 0.054 -> 0.045 (improves) ; accuracy 0.655 -> 0.650 | Hurt |
| Pre-processing for correlation shift before in-processing | Fairness Constraints, DP single metric · COMPAS test | DP unfairness: 0.050 -> 0.037 with std 0.021 overlapping | No clear effect |
| Pre-processing for correlation shift before FairBatch, DP and EO jointly | FairBatch, multiple metrics · Synthetic test, DP & EO | max(DP, EO) unfairness: 0.091 -> 0.058 | Helped |
| Pre-processing for correlation shift before fairness constraints, DP and EO jointly | Fairness Constraints · COMPAS, DP & EO | max(DP, EO) unfairness: 0.137 -> 0.106 | Helped |
| Pre-processing for correlation shift before FairBatch, DP and EO jointly | FairBatch · COMPAS, DP & EO | max(DP, EO) unfairness: 0.187 -> 0.130 | Helped |
| Pre-processing for correlation shift before in-processing when correlation shift is large (test corr. 10% of train) (ablation) | FairBatch · Synthetic, varying test correlation | DP unfairness: 0.108 -> 0.003 | Helped |
Closest papers by meaning
- Fairness without Harm: An Influence-Guided Active Sampling Approach (NeurIPS 2024 · similarity 0.88)
- FRAPPÉ: A Group Fairness Framework for Post-Processing Everything (ICML 2024 · similarity 0.88)
- Mitigating Label Bias in Machine Learning: Fairness through Confident Learning (AAAI 2024 · similarity 0.88)
- AIM-Fair: Advancing Algorithmic Fairness via Selectively Fine-Tuning Biased Models with Contextual Synthetic Data (CVPR 2025 · similarity 0.87)
- Achievable Fairness on Your Data With Utility Guarantees (NeurIPS 2024 · similarity 0.87)
- FedFACT: A Provable Framework for Controllable Group-Fairness Calibration in Federated 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.