Research map / Learning theory and algorithms / Fairness
Fairness-constrained classification
Learn classifiers under group or individual fairness constraints, with guarantees on fairness and accuracy.
148 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Fairness, keyword cluster “fair · classifier · classification”
State History Tried Baselines 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.
27 papers in 2023–24 and 44 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 16.2% to 31.4%.
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 · On Fairness and Calibration (NeurIPS 2017 · 279 citations)
- 2018 · Learning Adversarially Fair and Transferable Representations (ICML 2018 · 184 citations)
- 2019 · Flexibly Fair Representation Learning by Disentanglement (ICML 2019 · 123 citations)
- 2020 · Pairwise Fairness for Ranking and Regression (AAAI 2020 · 82 citations)
- 2021 · FairBatch: Batch Selection for Model Fairness (ICLR 2021 · 37 citations)
- 2022 · On the Fairness of Causal Algorithmic Recourse (AAAI 2022 · 53 citations)
- 2023 · Accurate Fairness: Improving Individual Fairness without Trading Accuracy (AAAI 2023 · 23 citations)
- 2024 · FFB: A Fair Fairness Benchmark for In-Processing Group Fairness Methods (ICLR 2024 · 4 citations)
- 2025 · Searching for Unfairness in Algorithms’ Outputs: Novel Tests and Insights (AAAI 2025 · 1 citation)
- 2026 · Fair Domain Generalization: An Information-Theoretic View (AAAI 2026 · 2 citations)
Earliest papers here
- On Fairness and Calibration (NeurIPS 2017 · 279 citations)
- Recycling Privileged Learning and Distribution Matching for Fairness (NeurIPS 2017 · 52 citations)
- Learning Adversarially Fair and Transferable Representations (ICML 2018 · 184 citations)
What has been tried
109 papers here report 957 measured changes (a component added, removed or swapped, compared with a baseline): helped 60%, no clear effect 10%, hurt 24%, mixed 7%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add uniform noise to estimated p(Y|X) and p(A|X) | Adult Census and CelebA (Attractive, Male) · test accuracy | within 1% across most α's | Hurt | Post-hoc bias scoring is optimal for fair classification (ICLR 2024 · 2 citations) |
| Use true rather than inferred group membership at test time | NLP datasets · performance and fairness metrics | No clear effect | OxonFair: A Flexible Toolkit for Algorithmic Fairness (NeurIPS 2024 · 1 citation) | |
| set initial α to 0.01 instead of 0.005 | RandLaplace n=1000 k=25 · Iterations | 9.0 -> 7.6 | Helped | Accelerating Spectral Clustering under Fairness Constraints (ICML 2025 · 0 citations) |
| Reduce the minimum width p_min to 0.1 | quantity-skew scenarios · collaboration gain spread performance | 11/12 cases; overall total to 48/54 cases | Helped | Aequa: Fair Model Rewards in Collaborative Learning via Slimmable Networks (ICML 2025 · 0 citations) |
| Use excessive EF nudging | Movie reviews: horror · Δ-Joyful | Hurt | On Optimal Steering to Achieve Exact Fairness (NeurIPS 2025 · 0 citations) | |
| Increase ε̂ from 0.01 to 0.40 | Barabási–Albert graphs · runtime | 19.1% | Helped | Fair Minimum Labeling: Efficient Temporal Network Activations for Reachability and Equity (NeurIPS 2025 · 0 citations) |
| add downstream-task supervision to FRG | Adult and Health · AUC | No clear effect | Fair Representation Learning with Controllable High Confidence Guarantees via Adversarial Inference (NeurIPS 2025 · 0 citations) | |
| Use the parametric ROAD implementation instead of BROAD | Compas, Law, German Credit · Worst-1-DI and global accuracy | Helped | On the Fairness ROAD: Robust Optimization for Adversarial Debiasing (ICLR 2024 · 0 citations) | |
| Use moderate temperature τ=10 or 20 instead of very small or large temperatures τ=5 or 50 | Adult and Bank · Fairness-accuracy trade-off | Helped | On the Maximal Local Disparity of Fairness-Aware Classifiers (ICML 2024 · 0 citations) | |
| remove sex from model input | German Credit · EO bias (standardized) | 0.083 -> 0.048 | Helped | Standardized Interpretable Fairness Measures for Continuous Risk Scores (ICML 2024 · 0 citations) |
| Increase the H fairness penalty from μ₂=1 to μ₂=10⁶ | School · Bias | 0.4824 -> 0.0308 | Helped | Fair GLASSO: Estimating Fair Graphical Models with Unbiased Statistical Behavior (NeurIPS 2024 · 0 citations) |
| Use greedy initialization instead of weighted-average initialization, then run local minimization | US Census (3 labels) · DCP upper bound | 14.07% -> 7.65% | Helped | Disparate Conditional Prediction in Multiclass Classifiers (ICML 2025) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| SVM | Adult | accuracy | 0.79 – 0.846 | 2 papers, e.g. ICML 2019 and NeurIPS 2017 |
What reviewers will expect
Datasets and baselines named in the 27 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 · 2 of 27 papers
- CelebA · 2 of 27 papers
Baselines
- FCRL · 2 of 27 papers
- Fare · 2 of 27 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 71 papers; from OpenAlex and the papers' affiliations).
- University of Pennsylvania · 3 papers
- Massachusetts Institute of Technology · 3 papers
- Rice University · 3 papers
- Stanford University · 3 papers
- Amazon · 3 papers
- University of Cambridge · 3 papers
- Microsoft · 3 papers
- University of Illinois, Urbana Champaign · 2 papers
- The Ohio State University · 2 papers
- RIKEN · 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
- Post-processing and transport for fairness (101 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).