Research map / Learning theory and algorithms / Fairness
Sample Selection for Fair and Robust Training
Yuji Roh, Kangwook Lee, Steven Euijong Whang, Changho Suh
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 fair-and-robust sample selection with fairness-aware weighted ERM | Logistic regression · Synthetic, label flipping 10% | EO disparity: .557$\pm$.015 -> .064$\pm$.005 | Improved |
| Use fair-and-robust sample selection with fairness-aware weighted ERM | Logistic regression · Synthetic, label flipping 10% | accuracy: .665$\pm$.003 -> .727$\pm$.002 | Improved |
| Use fair-and-robust sample selection with fairness-aware weighted ERM | Logistic regression · COMPAS, label flipping 10% | EO disparity: .123$\pm$.021 -> .044$\pm$.000 | Improved |
| Use fair-and-robust sample selection with fairness-aware weighted ERM | Logistic regression · AdultCensus, label flipping 10% | EO disparity: .070$\pm$.002 -> .029$\pm$.005 | Improved |
| Add fairness constraints to sample selection (ablation) | Full fair-and-robust method · Synthetic, label flipping 10% | EO disparity: .070$\pm$.004 -> .064$\pm$.005 | Improved |
| Add fairness constraints to sample selection (ablation) | Full fair-and-robust method · COMPAS, label flipping 10% | EO disparity: .110$\pm$.005 -> .044$\pm$.000 | Improved |
| Add reweighting during model training (ablation) | Full fair-and-robust method · Synthetic, label flipping 10% | EO disparity: .233$\pm$.014 -> .064$\pm$.005 | Improved |
| Add reweighting during model training (ablation) | Full fair-and-robust method · COMPAS, label flipping 10% | EO disparity: .152$\pm$.018 -> .044$\pm$.000 | Improved |
| Use loss-based clean sample selection alone | Logistic regression on COMPAS · COMPAS, noisy labels | EO disparity | Hurt |
| Use the proposed algorithm instead of FR-Train | Logistic regression · Synthetic, label flipping 10% | EO disparity: .078$\pm$.010 -> .064$\pm$.005 | Improved |
| Use the proposed algorithm instead of FR-Train with a clean validation set | Logistic regression on COMPAS · COMPAS, label flipping 10% | accuracy: .620$\pm$.009 -> .521$\pm$.000 | Hurt |
| Use the proposed algorithm instead of FR-Train with a clean validation set | Logistic regression on COMPAS · COMPAS, label flipping 10% | EO disparity: .074$\pm$.015 -> .044$\pm$.000 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FR-Train — FR-Train: A Mutual Information-Based Approach to Fair and Robust Training (ICML 2020) · better on 5 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Optimal Fair Learning Robust to Adversarial Distribution Shift (ICML 2025 · similarity 0.87)
- Fairness without Harm: An Influence-Guided Active Sampling Approach (NeurIPS 2024 · similarity 0.86)
- Multigroup Robustness (ICML 2024 · similarity 0.86)
- Navigating Towards Fairness with Data Selection (AAAI 2025 · similarity 0.86)
- Can Private Machine Learning Be Fair? (AAAI 2025 · similarity 0.86)
- Enhancing Robust Fairness via Confusional Spectral Regularization (ICLR 2025 · similarity 0.85)
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.
