Research map / Learning theory and algorithms / Fairness
Post-processing and transport for fairness
Adjust or transform classifiers, often via optimal transport, to remove discrimination while preserving utility.
101 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Fairness, keyword cluster “fair · classifier · classification”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
43 papers in 2023–24 and 23 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 25.7% to 16.4%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2017 · When Worlds Collide: Integrating Different Counterfactual Assumptions in Fairness (NeurIPS 2017 · 128 citations)
- 2018 · Preventing Fairness Gerrymandering: Auditing and Learning for Subgroup Fairness (ICML 2018 · 383 citations)
- 2019 · One-Network Adversarial Fairness (AAAI 2019 · 86 citations)
- 2020 · Fair Learning with Private Demographic Data (ICML 2020 · 26 citations)
- 2021 · Robust Fairness Under Covariate Shift (AAAI 2021 · 41 citations)
- 2022 · On the Impossibility of Non-trivial Accuracy in Presence of Fairness Constraints (AAAI 2022 · 6 citations)
- 2023 · Censored Fairness through Awareness (AAAI 2023 · 19 citations)
- 2024 · Arbitrariness and Social Prediction: The Confounding Role of Variance in Fair Classification (AAAI 2024 · 18 citations)
- 2025 · Secure and Confidential Certificates of Online Fairness (NeurIPS 2025 · 1 citation)
- 2026 · Fair Decisions from Calibrated Scores: Achieving Optimal Classification While Satisfying Sufficiency (ICML 2026 · 0 citations)
Earliest papers here
- When Worlds Collide: Integrating Different Counterfactual Assumptions in Fairness (NeurIPS 2017 · 128 citations)
- Preventing Fairness Gerrymandering: Auditing and Learning for Subgroup Fairness (ICML 2018 · 383 citations)
- Why Is My Classifier Discriminatory? (NeurIPS 2018 · 153 citations)
What has been tried
82 papers here report 696 measured changes (a component added, removed or swapped, compared with a baseline): helped 57%, no clear effect 12%, hurt 22%, mixed 8%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add an initial bagging pass before ensembling with abstention | Old Adult · Wasserstein-1 distance of self-consistency distributions | .063 -> .014 | Helped | Arbitrariness and Social Prediction: The Confounding Role of Variance in Fair Classification (AAAI 2024 · 18 citations) |
| Use only the two largest sensitive groups instead of four groups | ACS datasets · equalized-odds violation | 0.38 ... 0.16 | Helped | Unprocessing Seven Years of Algorithmic Fairness (ICLR 2024 · 4 citations) |
| set α=2 rather than α=1 | NYU-v2 (3-task) · Δm% | -2.79 -> -4.96 | Helped | Fair Resource Allocation in Multi-Task Learning (ICML 2024 · 4 citations) |
| Remove the causal edge from X to the proxy label Y_P | COMPAS · FNR bound informativeness | Helped | The Fragility of Fairness: Causal Sensitivity Analysis for Fair Machine Learning (NeurIPS 2024 · 3 citations) | |
| add the weighted entropy formulation | Arrhythmia, asymmetric shift A=0 · Error % | 28.648 (3.079) -> 27.648 (2.978) | Helped | Fairness under Covariate Shift: Improving Fairness-Accuracy Tradeoff with Few Unlabeled Test Samples (AAAI 2024 · 2 citations) |
| Use both λ₁ and λ₂ rather than λ₁ alone | COMPAS classification · Utility (accuracy) | 66.45}{ (0.85) -> 66.80}{ (0.27) | Helped | Towards Harmless Rawlsian Fairness Regardless of Demographic Prior (NeurIPS 2024 · 1 citation) |
| add the fairness regularization term | D&S · MNCE | 0.585 -> 0.985 | Helped | Fair Kernel K-Means: from Single Kernel to Multiple Kernel (NeurIPS 2024 · 1 citation) |
| Add interpolated training data | Multiple group settings · fairness improvement | Helped | Who’s the (Multi-)Fairest of Them All: Rethinking Interpolation-Based Data Augmentation Through the Lens of Multicalibration (AAAI 2025 · 0 citations) | |
| Use FairICP instead of FairCP, with ICP-generated sensitive-attribute permutations | Simulation studies · prediction loss–KPC trade-off | Helped | FairICP: Encouraging Equalized Odds via Inverse Conditional Permutation (ICML 2025 · 0 citations) | |
| Add SAM to the method | DFDC · Min/Max | 0.9656 -> 0.9771 | Helped | Preserving AUC Fairness in Learning with Noisy Protected Groups (ICML 2025 · 0 citations) |
| Increase FmCF penalty-estimation batch size | Adult, German, Compas, Health · standard deviation of penalty estimator | Adult: 0.1150 -> 0.0007; German: 0.1252 -> 0.0000; Compas: 0.0924 -> 0.0007; Health: 0.1004 -> 0.0009 | Helped | Simple and Effective Specialized Representations for Fair Classifiers (NeurIPS 2025 · 0 citations) |
| Relax exact demographic-parity equalities to double-sided inequalities with ε=10⁻⁴ | COMPAS · group disparity | No clear effect | Learning with Statistical Equality Constraints (NeurIPS 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 12 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Adult · 5 of 12 papers
- COMPAS · 4 of 12 papers
- ACSIncome · 2 of 12 papers
- CheXpert · 2 of 12 papers
- Crime · 2 of 12 papers
- Law School · 2 of 12 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 66 papers; from OpenAlex and the papers' affiliations).
- Stanford University · 5 papers
- Google · 5 papers
- Harvard University · 5 papers
- ByteDance · 4 papers
- The Ohio State University · 3 papers
- University of California, Santa Cruz · 3 papers
- Microsoft · 2 papers
- Florida International University · 2 papers
- Carnegie Mellon University · 2 papers
- KRAFTON · 2 papers
Use it
Download BibTeX of 40 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Fairness-constrained classification (148 papers)
- Fair allocation and selection (69 papers)
- Minimax fairness via sampling (38 papers)
Elsewhere in the topic
- Causal and policy fairness in decisions (75 papers)
- Fairness auditing and regularization (65 papers)
- Fairness in recommendation and ranking (20 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).