Pith. sign in

REVIEW 5 major objections 4 minor 66 references

Whence Is A Model Fair? Fixing Fairness Bugs via Propensity Score Matching

T0 review · 5 major / 4 minor · reviewed 2026-08-16 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A model's reported fairness can be a sampling artifact: propensity score matching lets FairMatch locate the already-fair test samples and fix only the rest, removing most measured bias without sacrificing predictive performance.

desk verdict FairMatch's core claim is undercut by an algorithm that fits to ytrain and matches by Euclidean distance, but the paper still documents a real sensitivity of fairness metrics to sampling. read the letter →

arxiv 2504.17066 v2 pith:FPZTJFTS submitted 2025-04-23 cs.LG cs.CYcs.SEstat.ML

Kewen Peng , Yicheng Yang , Hao Zhuo This is my paper · ORCID
classification cs.LGcs.CYcs.SEstat.ML
keywords SoftwarefairnesstestingPropensityscorematchingBiasmitigationGroupDecisionthresholdcalibrationmetricsPost-processing
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Random train/test splits can make a biased model look fairer or less fair than it actually is, because the test set inherits the same skewed distribution as the training data. This paper argues that fairness testing should first ask which test samples are comparable across protected groups, and it proposes FairMatch, a post-processing method that uses propensity score matching to answer that question. FairMatch splits the test set into matched pairs of similar examples from opposite protected groups, where the model is already unbiased, and unmatched samples, where most of the discrimination sits; it then recalibrates decision thresholds separately for each group on the unmatched part. Across six tabular benchmarks, the matched subsets show drastically reduced fairness gaps while the unmatched subsets concentrate the bias, and FairMatch's threshold adjustment yields fairness-performance trade-offs at least as good as existing pre-, in-, and post-processing baselines. The paper's broader claim is that reported fairness metrics are highly sensitive to sampling, so fairness studies that ignore propensity may have been evaluating models on the wrong test data.

What carries the argument

The central object is the pair-forming step that the paper calls propensity score matching. Conceptually, a propensity score is the probability that an individual falls into one protected group rather than the other given observable features; in the implementation, a learner is fitted to the training labels, scores the test set, and forms control/treatment pairs by joining each test point to its $k$-nearest Euclidean neighbours with the opposite protected attribute. That pairing splits the test set into a matched subset (treated as already fair) and an unmatched subset (treated as at risk of discrimination). The mitigation step then searches over additive decision-threshold offsets $\theta_1$ and $\theta_2$ for the privileged and unprivileged groups, maximizing the $p$-value of a two-sample $t$-test comparing the shifted score distributions while minimizing $|\theta_1| + |\theta_2|$; the paper also introduces a fairness area-under-the-curve view in which the proportion of matched samples is swept to show how measured fairness drifts with the test composition.

What would settle it

Build a synthetic binary-classification dataset in which an unobserved variable $U$ influences both protected-group membership and the favorable outcome, train a classifier, and run FairMatch's matching on the remaining features. If, after conditioning on $U$, the matched pairs from opposite protected groups still show substantially different favorable-outcome rates, then the matched subset is not actually unbiased and the central claim fails.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a model's fairness is not a property of the model alone; it is a property of the model evaluated on a particular mix of comparable and non-comparable test samples. Using propensity score matching, FairMatch locates test samples that can be paired, through nearest neighbors, with an opposite-protected-group counterpart, and finds that these matched samples display drastically reduced fairness gaps. The remaining unmatched samples are where bias concentrates, and recalibrating subgroup decision thresholds on just those samples—keeping the threshold shift small and demanding that the two groups' score distributions become statistically indistinguishable—mitigates most measured bias. FairMatch reports that this selective strategy matches or beats existing pre-, in-, and post-processing baselines on fairness-performance trade-offs, and it recommends presenting fairness not as a single number but as a projection curve that shows how the reported fairness changes as the proportion of matched samples changes.

Load-bearing premise

The method assumes that whether two people can be paired as similar, using the data the model actually sees, tells us everything about whether the model is being fair to them; if the real reasons for unequal treatment lie in information the data does not contain, the 'unbiased' matched group may still be biased.

Editorial extensions

If this is right

  • Fairness audits should report the matchable fraction of the test set alongside any fairness score, because the same model can be measured fair or unfair depending on that fraction.
  • Debiasing can be selective: matched samples are left alone and only unmatched samples receive threshold recalibration, so the performance cost of fixing bias is spent where bias actually exists.
  • Fairness metrics are far more sensitive to sampling strategy than performance metrics are, so method comparisons made on a single random test split should be re-examined.
  • Because FairMatch is post-processing and only adjusts decision thresholds, it can be applied to an already-deployed binary classifier without retraining the original model.
  • The proposed projection curve gives practitioners a way to reason about how fairness would look under future distribution drift instead of trusting one test-set number.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A direct stress test of the mechanism is to refit the propensity model to the protected attribute rather than the outcome label and check whether the matched subset still shows near-zero disparity; this separates the conceptual propensity claim from the specific implementation choice.
  • The same matched/unmatched split could be applied to individual fairness or to regression and ranking outcomes; the paper's argument is restricted to group fairness on binary tabular data, so extending it would show whether the 'matched samples are already fair' conclusion is metric-independent.
  • If match rate is reported as a quality signal, datasets with very low match rates would be flagged as structurally hard to debias by thresholding alone, implying that pre-processing or new data collection is needed before post-processing can work.
  • The threshold search maximizes a $t$-test $p$-value while minimizing threshold shifts; a practical monitoring step not discussed in the paper would be to track accuracy on the unmatched subset after deployment, since large threshold shifts could trade away calibration even when group fairness improves.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 4 minor

Summary. This paper proposes FairMatch, a post-processing method for fairness evaluation and mitigation that is claimed to be based on propensity score matching (PSM). The authors argue that the standard random train/test split yields misleading fairness metrics and that PSM can identify the subset of test data on which a model is unbiased, then calibrate decision thresholds for the remaining data. The method consists of two algorithms: Algorithm 1 selects "matched" samples via nearest-neighbor Euclidean-distance pairing across protected groups, and Algorithm 2 searches for subgroup-specific threshold offsets that maximize a t-test p-value between score distributions. The evaluation uses six tabular datasets, four fairness metrics, four performance metrics, and several baselines (EGR, FAX-AI, FairMask, MAAT) with logistic regression and gradient boosting, reporting that FairMatch achieves comparable or better fairness-performance trade-offs.

Significance. If the central claim held, the paper would make a useful contribution: it addresses a real question—whether test-set composition changes reported fairness—and it ships an open repository, uses six datasets, repeats trials 20 times, and applies non-parametric significance and effect-size tests. The f-AUC projection-curve idea is also suggestive. However, the strength of these contributions is conditional on the method actually implementing propensity score matching. As detailed below, the central identification step is not PSM, the unbiased-subset result in Table 4 is essentially built into the matching construction, and the mitigation evaluation is in-sample. These problems undermine both abstract claims (a) and (b).

major comments (5)
  1. [§3.2, Algorithm 1] Algorithm 1 does not implement propensity score matching. It calls PropensityModel.fit(Xtrain, ytrain), so the fitted scores estimate P(y | x), not the probability of belonging to the protected group P(PA | x). Moreover, the variable ps computed from PropensityModel.predict(Xtest) is never used in the matching loop; pairs are selected by Euclidean distance in feature space and filtered by opposite protected attribute. In the Rosenbaum-Rubin framework, PSM requires matching on the estimated probability of treatment and checking covariate balance; neither is present here. Consequently the Abstract and §6 claims that FairMatch can "precisely locate" the unbiased subset and mitigate bias on the rest are not supported by the algorithm as written.
  2. [§3.2, Algorithm 1; Table 3] The output of Algorithm 1 is inconsistent with the paper's description of a matched subset. In the loop, row is only processed when row.PA == 1; the privileged row is added to matched and its opposite-protected neighbor is added to not matched. Thus "matched" contains only privileged samples, while the unprivileged neighbors are placed in "not matched". This makes the "PSMsampled" rows in Tables 3 and 4 undefined as the paired subset, and the phrase "not matched" is also misused because those unprivileged neighbors were in fact paired. The pseudocode also references neighbors.get(row) and neighbors.append(Xtest, k_neighbors) without defining a keyed structure, so the matching loop cannot be executed as written.
  3. [§4.2, Eq. (1), Eq. (2), Algorithm 2] The threshold-search method is not connected to the decision thresholds it claims to adjust. Algorithm 2 adds θ1 to the privileged propensity scores and subtracts θ2 from the unprivileged propensity scores, then runs a t-test, rather than applying offsets to the classifier's decision rule; §4.2 does not specify how θ1 and θ2 translate to thresholds on model outputs. In addition, Eq. (1) and Eq. (2) define a two-objective search, but Algorithm 2 uses a greedy update that requires both a larger p-value and a smaller |θ1|+|θ2| than the previous candidate, with no analysis showing convergence to a point on the intended trade-off frontier. The mitigation results in Tables 7-8 are therefore not grounded in the stated method.
  4. [§6 RQ2, Table 4] The claim that "PSM sampling" locates an unbiased subset is circular. Pairing points that are close in Euclidean feature space and have opposite protected attributes will, for a smooth classifier, tend to have similar predicted scores simply because score functions are locally smooth; the matched subset therefore has balanced predictions by construction. The -1.00 changes in SPD and DI for the PSMsampled rows in Table 4 are exactly what this construction predicts and do not establish that the model is unbiased on those samples in any causal sense. No covariate balance check or discussion of unobserved confounders is reported, despite such checks being required for propensity-score claims.
  5. [§5.4 RQ4 / §6 RQ4, Algorithm 2] The mitigation evaluation is in-sample. Algorithm 2 selects θ1 and θ2 using the same test-set propensity scores on which Tables 7 and 8 report the fairness and performance metrics, and no held-out or cross-validation split is described for the threshold search. The RQ4 comparisons therefore include selection on the test set and may overstate the method's out-of-sample gains. A separate validation set or nested resampling procedure is needed before the "on-par or superior" claim is credible.
minor comments (4)
  1. [Table 5] The header "Accuray" is a typo for "Accuracy".
  2. [§2.1] The sentence "These are are many such examples" contains a duplicated word, and the reference to "V olkswagen" has an erroneous space.
  3. [Table 2] The DI definition uses mismatched brackets, "P (Y = 1|PA = 0]/P [Y = 1|PA = 1)", and the SPD wording should be aligned with the difference formula shown.
  4. [§3.2, Figure 3] Figure 3 is described as showing the proposed sampling schema, but the figure does not clarify how the matched subset is used by the later threshold-search step; a concrete pipeline diagram would improve readability.

Circularity Check

2 steps flagged · score 8.0 of 10

The 'unbiased subset' is defined by Euclidean nearest-neighbor pairing, not by propensity scores, and the threshold calibration is fitted to the same test set that is later used to report fairness, so both headline claims reduce to construction and in-sample fit.

  1. self definitional [Algorithm 1 (Section 3.2); Section 4.2, 'Matched' definition; Abstract claim (a); Table 4]
    "PropensityModel . f it(Xtrain,ytrain) ps← PropensityModel .predict(Xtest ) ... d← dist(xtest ,xtrain) // Calculate Euclidean distance ... if neighbor.PA̸= row.PA then matched← matched∪ row"

    The 'unbiased subset' is defined by the matching rule, not discovered. The predicted ps is computed but never used in the matching loop; pairs are Euclidean nearest neighbors with opposite protected attribute. For a smooth classifier, close feature vectors receive similar predictions, so the matched set's protected-group score distributions are close by construction. Section 4.2 then asserts 'Matched: Pairs of data points that can be matched by propensity are those not suffering from model discrimination,' and the Abstract claims (a) 'precisely locate subsets of the test data where the model is unbiased.' That is a self-definitional reduction: the subset is unbiased because it was constructed from similar prediction pairs.

  2. fitted input called prediction [Section 4.2, Algorithm 2; Section 6, RQ4, Tables 7-8]
    "we iteratively search for the optimal thresholds corresponding to each subgroup based on the protected attribute. ... Maximize p = P(|T| >|t′|) where t′ = ( ¯X + θ1)− ( ¯Y + θ2)q s2 X nX + s2 Y nY ... Minimize|θ1| +|θ2|"

    Algorithm 2 selects θ1 and θ2 by maximizing the t-test p-value between the two protected groups' score distributions on the test partition, and RQ4 then reports fairness metrics computed on that same test set (Tables 7 and 8). The threshold adjustment is therefore fitted to the very data used to measure the outcome; the reported 'superior fairness-performance trade-offs' are in-sample optimization results, not predictions on new data.

full rationale

The paper's headline claims are (a) precise location of unbiased test subsets and (b) significant bias mitigation on the rest. Claim (a) is built from Algorithm 1: samples are paired by Euclidean distance in feature space under opposite protected attributes, and the same paired subset is then declared 'not suffering from model discrimination' in Section 4.2. Since no propensity score is used in the pairing (the computed ps is never referenced after prediction), and since smooth classifiers assign similar scores to nearby feature vectors, the matched subset has similar outcome-score distributions across protected groups by construction; the Table 4 result that SPD and DI drop by exactly 100% on PSM-sampled rows is the signature of this construction, not an empirical discovery. Claim (b) is an in-sample fit: Algorithm 2 tunes thresholds to maximize a balance test on the test partition, and Tables 7-8 evaluate FairMatch on that same partition, so the threshold choice and the reported fairness numbers are not independent. The paper cites Rosenbaum-Rubin for the PSM framework, but its own Algorithm 1 does not implement that framework, so the citation cannot independently validate the implementation. Consequently the central derivation reduces to the algorithm's own construction and to in-sample threshold fitting; this is circularity of the self-definitional and fitted-input-called-prediction kinds, meriting a score of 8 rather than a lower score, with no evidence of a separately validated external benchmark for the key claims.

Assumptions & free parameters 4 free parameters · 5 assumptions · 1 invented entities

The central method rests on several untested assumptions and parameters chosen to make results look good. The propensity model is fitted to the outcome, the matching uses feature distance, and the threshold offsets are tuned on the test set itself.

free parameters (4)
  • k (number of KNN neighbors) = not stated
    Algorithm 1 uses k-nearest neighbors to find opposite-PA matches; k is not specified and will strongly affect matching.
  • PropensityModel type = logistic regression
    The paper states several candidate models were tried and a smooth one selected (Section 7), making the model choice an ad hoc fit to make the method work.
  • θ1, θ2 (decision threshold offsets) = values per dataset, not listed
    These are fitted on the test set via Algorithm 2 to maximize a t-test p-value, an in-sample optimization.
  • Threshold search step = unknown
    Algorithm 2 searches pools 0:100 for θ1 and θ2 but does not specify the step size or stopping rule.
assumptions (5)
  • domain assumption Biased or imbalanced training data is one of the major reasons for unfairness in machine learning models.
    Stated in Section 1 as a premise for the whole approach.
  • domain assumption The testing set shares a likelihood of bias similar to the training set.
    Stated in Section 1; justifies applying propensity adjustment to the test set.
  • domain assumption Treatment (protected attribute) is completely random conditional on specified variables (unconfoundedness).
    Invoked in RQ3 Section 5.4; untestable and likely violated for race/sex.
  • ad hoc to paper The future deployment distribution is represented by the propensity-matched subset.
    Figure 1 caption states 'ensure that the test set reflects the ideal data distribution in the future'; no evidence supports this.
  • ad hoc to paper A t-test p-value comparing score distributions is a valid criterion for distributional balance.
    Algorithm 2 maximizes p-value, but t-tests only compare means and are sample-size dependent.
invented entities (1)
  • f-AUC (fairness area under the curve)
    purpose: Summarizes how fairness scores vary with the proportion of propensity-matchable samples.
    A new aggregate metric introduced in Section 4.1 with no external validation or benchmark.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Whence Is A Model Fair? Fixing Fairness Bugs via Propensity Score Matching." pith.science (2026). https://pith.science/paper/FPZTJFTS

@misc{pith2026250417066,
  author       = {Pith},
  title        = {Pith review of: Whence Is A Model Fair? Fixing Fairness Bugs via Propensity Score Matching},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FPZTJFTS}},
  note         = {Machine review of arXiv:2504.17066}
}
read the original abstract

Fairness-aware learning aims to mitigate discrimination against specific protected social groups (e.g., those categorized by gender, ethnicity, age) while minimizing predictive performance loss. Despite efforts to improve fairness in machine learning, prior studies have shown that many models remain unfair when measured against various fairness metrics. In this paper, we examine whether the way training and testing data are sampled affects the reliability of reported fairness metrics. Since training and test sets are often randomly sampled from the same population, bias present in the training data may still exist in the test data, potentially skewing fairness assessments. To address this, we propose FairMatch, a post-processing method that applies propensity score matching to evaluate and mitigate bias. FairMatch identifies control and treatment pairs with similar propensity scores in the test set and adjusts decision thresholds for different subgroups accordingly. For samples that cannot be matched, we perform probabilistic calibration using fairness-aware loss functions. Experimental results demonstrate that our approach can (a) precisely locate subsets of the test data where the model is unbiased, and (b) significantly reduce bias on the remaining data. Overall, propensity score matching offers a principled way to improve both fairness evaluation and mitigation, without sacrificing predictive performance.

Figures

Figures reproduced from arXiv: 2504.17066 by the authors.

Figure 1
Figure 1. Fairness testing proposed in this paper. To evaluate the effectiveness of any de-biasing model, we want to [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Curves drawn from three datasets. The gray error area indicates the stan￾dard deviation as we use different random seeds for sub-sampling. The x-axis repre￾sents the percentage of unmatched testing samples used when calculating the fairness scores. The y-axis represents the fairness metric (DI in this case), where lower scores indicate better fairness. 13 [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Propensity score matching selects samples that represent the future distribution of class labels and the [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Examples derived from three datasets. The distribution of propensity scores is drawn among the matched/unmatched samples. 21 [PITH_FULL_IMAGE:figures/full_fig_p021_4.png]
Figure 5
Figure 5. Figure 5: RQ3 result: The comparison of performance-fairness trade-offs before and after applying PSM. Each arrow represents a set of comparisons conducted on the datasets listed in [PITH_FULL_IMAGE:figures/full_fig_p025_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 45 canonical work pages

  1. [1]

    Chakraborty, T

    J. Chakraborty, T. Xia, F. M. Fahid, T. Menzies, Software engineering for fairness: A case study with hyperparameter optimization (2019). arXiv: 1905.05786

    arXiv 2019

  2. [2]

    Chakraborty, S

    J. Chakraborty, S. Majumder, T. Menzies, Bias in machine learning soft- ware: Why? how? what to do?, in: Proceedings of the 29th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering, ESEC/FSE 2021, Associa- tion for Computing Machinery, New York, NY , USA, 2021, p. 429–440. doi:10.1145/3468...

    arXiv 2021

  3. [3]

    Z. Chen, J. M. Zhang, F. Sarro, M. Harman, Maat: a novel ensemble ap- proach to addressing fairness and performance bugs for machine learning software, in: Proceedings of the 30th ACM Joint European Software Engi- neering Conference and Symposium on the Foundations of Software Engi- neering, 2022, pp. 1122–1134

    work page 2022

  4. [4]

    K. Peng, J. Chakraborty, T. Menzies, Fairmask: Better fairness via model- based rebalancing of protected attributes, IEEE Transactions on Software Engineering (2022)

    work page 2022

  5. [5]

    P. R. Rosenbaum, D. B. Rubin, The central role of the propensity score in observational studies for causal effects, Biometrika 70 (1) (1983) 41–55

    work page 1983

  6. [6]

    A. F. Cruz, P. Saleiro, C. Bel ´em, C. Soares, P. Bizarro, Promoting fairness through hyperparameter optimization, in: 2021 IEEE International Confer- ence on Data Mining (ICDM), IEEE, 2021, pp. 1036–1041

    work page 2021

  7. [7]

    URL http://mlr.cs.umass.edu/ml/datasets/Adult

    Uci:adult data set (1994). URL http://mlr.cs.umass.edu/ml/datasets/Adult

    work page 1994

  8. [8]

    URL https://github.com/propublica/compas-analysis

    propublica/compas-analysis (2015). URL https://github.com/propublica/compas-analysis

    work page 2015

Show all 66 references
  1. [9]

    Buolamwini, T

    J. Buolamwini, T. Gebru, Gender shades: Intersectional accuracy disparities in commercial gender classification, in: Conference on fairness, account- ability and transparency, 2018, pp. 77–91. 29

    2018

  2. [10]

    Garvie, The perpetual line-up: Unregulated police face recognition in America, Georgetown Law, Center on Privacy & Technology, 2016

    C. Garvie, The perpetual line-up: Unregulated police face recognition in America, Georgetown Law, Center on Privacy & Technology, 2016

    2016

  3. [11]

    Caliskan, J

    A. Caliskan, J. J. Bryson, A. Narayanan, Semantics derived automati- cally from language corpora contain human-like biases, Science 356 (6334) (2017) 183–186

    2017

  4. [12]

    Rudin, Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead, Nature Machine In- telligence 1 (5) (2019) 206–215

    C. Rudin, Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead, Nature Machine In- telligence 1 (5) (2019) 206–215. doi:10.1038/s42256-019-0048-x . URL https://doi.org/10.1038/s42256-019-0048-x

    2019 doi

  5. [13]

    S. U. Noble, Algorithms of oppression. How search engines reinforce racism, New York University Press, New York, 2018. URL http://algorithmsofoppression.com/

    2018

  6. [14]

    Johnson, T

    B. Johnson, T. Menzies, Unfairness is everywhere. so what to do? an inter- view with jeanna matthews, IEEE Software (2023)

    2023

  7. [15]

    Majumder, J

    S. Majumder, J. Chakraborty, G. R. Bai, K. T. Stolee, T. Menzies, Fair enough: Searching for sufficient measures of fairness, ACM Trans. Softw. Eng. Methodol.Just Accepted (mar 2023). doi:10.1145/3585006. URL https://doi.org/10.1145/3585006

    2023 doi

  8. [16]

    URL https://archive.ics.uci.edu/ml/datasets/Statlog+ (German+Credit+Data)

    Uci:statlog (german credit data) data set (2000). URL https://archive.ics.uci.edu/ml/datasets/Statlog+ (German+Credit+Data)

    2000

  9. [17]

    URL https://www.kaggle.com/c/bank-marketing-uci

    Bank marketing uci (2017). URL https://www.kaggle.com/c/bank-marketing-uci

    2017

  10. [18]

    URL https://archive.ics.uci.edu/ml/datasets/Heart+Disease

    Uci:heart disease data set (2001). URL https://archive.ics.uci.edu/ml/datasets/Heart+Disease

    2001

  11. [19]

    URL https://meps.ahrq.gov/mepsweb/

    Medical expenditure panel survey (2015). URL https://meps.ahrq.gov/mepsweb/

    2015

  12. [20]

    Canellas, Defending ieee software standards in federal criminal court, Computer 54 (6) (2021) 14–23

    M. Canellas, Defending ieee software standards in federal criminal court, Computer 54 (6) (2021) 14–23. doi:10.1109/MC.2020.3038630. 30

    2021

  13. [21]

    J. Matthews, How should we regulate ai? practical strategies for regulation and risk management from the ieee1012 stan- dard for system, software, and hardware verification and valida- tion.https://ieeeusa.org/assets/public-policy/committees/ aipc/How-Should-We-Regulate-AI.pdf (2023)

    2023

  14. [22]

    Green, The flaws of policies requiring human oversight of government algorithms, Computer Law & Security Review 45 (2022) 105681

    B. Green, The flaws of policies requiring human oversight of government algorithms, Computer Law & Security Review 45 (2022) 105681

    2022

  15. [23]

    Tizpaz-Niari, A

    S. Tizpaz-Niari, A. Kumar, G. Tan, A. Trivedi, Fairness-aware configu- ration of machine learning libraries, in: Proceedings of the 44th Inter- national Conference on Software Engineering, ICSE ’22, Association for Computing Machinery, New York, NY , USA, 2022, p. 909–920. doi...

    2022

  16. [24]

    Y . Li, L. Meng, L. Chen, L. Yu, D. Wu, Y . Zhou, B. Xu, Training data de- bugging for the fairness of machine learning software, in: Proceedings of the 44th International Conference on Software Engineering, ICSE ’22, Associ- ation for Computing Machinery, New York, NY , USA, ...

    2022

  17. [25]

    T. Li, Y . Cao, J. Zhang, S. Zhao, Y . Huang, A. Liu, Q. Guo, Y . Liu, Run- ner: Responsible unfair neuron repair for enhancing deep neural network fairness, in: Proceedings of the 46th IEEE/ACM International Conference on Software Engineering, 2024, pp. 1–13

    2024

  18. [26]

    Gohar, S

    U. Gohar, S. Biswas, H. Rajan, Towards understanding fairness and its com- position in ensemble machine learning, in: 2023 IEEE/ACM 45th Interna- tional Conference on Software Engineering (ICSE), 2023, pp. 1533–1545. doi:10.1109/ICSE48619.2023.00133

    2023

  19. [27]

    M. Hort, J. M. Zhang, F. Sarro, M. Harman, Search-based automatic repair for fairness and accuracy in decision-making software, Empirical Software Engineering 29 (1) (2024) 36

    2024

  20. [28]

    Y . Xiao, J. M. Zhang, Y . Liu, M. R. Mousavi, S. Liu, D. Xue, Mirrorfair: Fixing fairness bugs in machine learning software via counterfactual predic- tions, Proc. ACM Softw. Eng. 1 (FSE) (jul 2024). doi:10.1145/3660801. URL https://doi.org/10.1145/3660801 31

    2024 doi

  21. [29]

    A. Hoag, J. E. Kostas, B. C. Da Silva, P. S. Thomas, Y . Brun, Seldonian toolkit: Building software with safe and fair machine learning, in: 2023 IEEE/ACM 45th International Conference on Software Engineering: Com- panion Proceedings (ICSE-Companion), IEEE, 2023, pp. 107–111

    2023

  22. [30]

    Ferrara, G

    C. Ferrara, G. Sellitto, F. Ferrucci, F. Palomba, A. De Lucia, Fairness-aware machine learning engineering: how far are we?, Empirical software engi- neering 29 (1) (2024) 9

    2024

  23. [31]

    Z. Chen, J. M. Zhang, M. Hort, M. Harman, F. Sarro, Fairness testing: A comprehensive survey and analysis of trends, ACM Trans. Softw. Eng. Methodol. 33 (5) (jun 2024). doi:10.1145/3652155. URL https://doi.org/10.1145/3652155

    2024 doi

  24. [32]

    J. M. Zhang, M. Harman, ”ignorance and prejudice” in software fairness, in: 2021 IEEE/ACM 43rd International Conference on Software Engineering (ICSE), 2021, pp. 1436–1447. doi:10.1109/ICSE43902.2021.00129

    2021

  25. [33]

    Chakraborty, S

    J. Chakraborty, S. Majumder, Z. Yu, T. Menzies, Fairway: A way to build fair ml software, in: Proceedings of the 28th ACM Joint Meeting on Euro- pean Software Engineering Conference and Symposium on the Foundations of Software Engineering, ESEC/FSE 2020, Association for Comput...

    2020

  26. [34]

    Zhang, J

    P. Zhang, J. Wang, J. Sun, G. Dong, X. Wang, X. Wang, J. S. Dong, T. Dai, White-box fairness testing through adversarial sampling, in: Proceedings of the ACM/IEEE 42nd International Conference on Software Engineering, 2020, pp. 949–960

    2020

  27. [35]

    J. Wang, Y . Yang, S. Wang, J. Hu, Q. Wang, Context- and fairness-aware in- process crowdworker recommendation, ACM Trans. Softw. Eng. Methodol. 31 (3) (mar 2022). doi:10.1145/3487571. URL https://doi.org/10.1145/3487571

    2022 doi

  28. [36]

    Z. Chen, J. M. Zhang, F. Sarro, M. Harman, A comprehensive empirical study of bias mitigation methods for machine learning classifiers, ACM Trans. Softw. Eng. Methodol. 32 (4) (may 2023). doi:10.1145/3583561. URL https://doi.org/10.1145/3583561 32

    2023 doi

  29. [37]

    J. Yang, J. Jiang, Z. Sun, J. Chen, A large-scale empirical study on improv- ing the fairness of image classification models, in: Proceedings of the 33rd ACM SIGSOFT International Symposium on Software Testing and Analy- sis, 2024, pp. 210–222

    2024

  30. [38]

    M. Hort, J. M. Zhang, F. Sarro, M. Harman, Fairea: A model behaviour mu- tation approach to benchmarking bias mitigation methods, in: Proceedings of the 29th ACM joint meeting on European software engineering confer- ence and symposium on the foundations of software engineerin...

    2021

  31. [39]

    URL https://www.whitehouse.gov/ostp/ai-bill-of-rights/

    White House Office of Science and Technology Policy, Blueprint for an ai bill of rights: Making automated systems work for the american people, accessed: September 21, 2024 (October 2022). URL https://www.whitehouse.gov/ostp/ai-bill-of-rights/

    2024

  32. [40]

    C. o. t. E. U. European Parliament, Regulation (eu) 2024/1689 of the euro- pean parliament and of the council of 13 june 2024 laying down harmonised rules on artificial intelligence and amending regulations (ec) no 300/2008, (eu) no 167/2013, (eu) no 168/2013, (eu) 2018/858, (...

    2024

  33. [41]

    Chopra, K

    R. Chopra, K. Clarke, B. Boynton, D. McCain, C. E. Lhamon, M. Fontes Rainer, S. S. Wadhia, S. Nanda, C. A. Burrows, L. M. Khan, Joint statement on enforcement of civil rights, fair competition, consumer protec- tion, and equal opportunity laws in automated systems, Joint State...

    2024

  34. [42]

    Kamiran, T

    F. Kamiran, T. Calders, Data preprocessing techniques for classification without discrimination, Knowledge and Information Systems 33 (1) (2012) 1–33

    2012

  35. [43]

    Feldman, S

    M. Feldman, S. A. Friedler, J. Moeller, C. Scheidegger, S. Venkatasubrama- nian, Certifying and removing disparate impact, in: proceedings of the 21th ACM SIGKDD international conference on knowledge discovery and data mining, 2015, pp. 259–268. 33

    2015

  36. [44]

    Biswas, H

    S. Biswas, H. Rajan, Do the machine learning models on a crowd sourced platform exhibit bias? an empirical study on model fairness, Proceedings of the 28th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering (N...

    2020

  37. [45]

    Kamishima, S

    T. Kamishima, S. Akaho, H. Asoh, J. Sakuma, Fairness-aware classifier with prejudice remover regularizer, in: P. A. Flach, T. De Bie, N. Cristianini (Eds.), Machine Learning and Knowledge Discovery in Databases, Springer Berlin Heidelberg, Berlin, Heidelberg, 2012, pp. 35–50

    2012

  38. [46]

    Agarwal, A

    A. Agarwal, A. Beygelzimer, M. Dud ´ık, J. Langford, H. Wallach, A reduc- tions approach to fair classification, in: International conference on machine learning, PMLR, 2018, pp. 60–69

    2018

  39. [47]

    Kamiran, S

    F. Kamiran, S. Mansha, A. Karim, X. Zhang, Exploiting reject option in classification for social discrimination control, Inf. Sci. (2018). doi:10. 1016/j.ins.2017.09.064

    2018

  40. [48]

    P. A. Grabowicz, N. Perello, A. Mishra, Marrying fairness and explainability in supervised learning, in: Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency, 2022, pp. 1905–1916

    2022

  41. [49]

    Giguere, B

    S. Giguere, B. Metevier, Y . Brun, P. S. Thomas, S. Niekum, B. C. da Silva, Fairness guarantees under demographic shift, in: International Conference on Learning Representations, 2022. URL https://openreview.net/forum?id=wbPObLm6ueA

    2022

  42. [50]

    B. An, Z. Che, M. Ding, F. Huang, Transferring fairness under distribution shifts via fair consistency regularization, Advances in Neural Information Processing Systems 35 (2022) 32582–32597

    2022

  43. [51]

    Y . Roh, K. Lee, S. E. Whang, C. Suh, Improving fair training under corre- lation shifts, in: A. Krause, E. Brunskill, K. Cho, B. Engelhardt, S. Sabato, J. Scarlett (Eds.), Proceedings of the 40th International Conference on Ma- chine Learning, V ol. 202 of Proceedings of Mach...

    2023

  44. [52]

    Y . Li, X. Wang, Y . Xing, S. Fan, R. Wang, Y . Liu, C. Shi, Graph fairness learning under distribution shifts, in: Proceedings of the ACM Web Confer- ence 2024, WWW ’24, Association for Computing Machinery, New York, NY , USA, 2024, p. 676–684.doi:10.1145/3589334.3645508. URL...

    2024

  45. [53]

    R. K. Bellamy, K. Dey, M. Hind, S. C. Hoffman, S. Houde, K. Kannan, P. Lohia, J. Martino, S. Mehta, A. Mojsilovic, et al., Ai fairness 360: An extensible toolkit for detecting, understanding, and mitigating unwanted al- gorithmic bias, arXiv preprint arXiv:1810.01943 (2018)

    2018 arXiv

  46. [54]

    Menzies, A

    T. Menzies, A. Butcher, A. Marcus, D. Cok, F. Shull, B. Turhan, T. Zimmer- mann, Local versus global lessons for defect prediction and effort estimation, IEEE Transactions on software engineering 39 (6) (2013) 822–834

    2013

  47. [55]

    Menzies, E

    T. Menzies, E. Kocaguneli, B. Turhan, L. Minku, F. Peters, Sharing data and models in software engineering, Morgan Kaufmann, 2014

    2014

  48. [56]

    Chakraborty, K

    J. Chakraborty, K. Peng, T. Menzies, Making fair ml software using trust- worthy explanation, in: Proceedings of the 35th IEEE/ACM International Conference on Automated Software Engineering, ASE ’20, Association for Computing Machinery, New York, NY , USA, 2020, p. 1229–1233. ...

    2020

  49. [57]

    Hardt, E

    M. Hardt, E. Price, N. Srebro, Equality of opportunity in supervised learning (2016). arXiv:1610.02413

    2016 arXiv

  50. [58]

    Pleiss, M

    G. Pleiss, M. Raghavan, F. Wu, J. Kleinberg, K. Q. Weinberger, On fairness and calibration, arXiv preprint arXiv:1709.02012 (2017)

    2017 arXiv

  51. [59]

    K. Peng, C. Kaltenecker, N. Siegmund, S. Apel, T. Menzies, Veer: enhanc- ing the interpretability of model-based optimizations, Empirical Software Engineering 28 (3) (2023) 61

    2023

  52. [60]

    Mittas, L

    N. Mittas, L. Angelis, Ranking and clustering software cost estimation mod- els through a multiple comparisons algorithm, IEEE Transactions on soft- ware engineering 39 (4) (2012) 537–551. 35

    2012

  53. [61]

    T. Xia, R. Krishna, J. Chen, G. Mathew, X. Shen, T. Menzies, Hyperpa- rameter optimization for effort estimation, arXiv preprint arXiv:1805.00336 (2018)

    2018 arXiv

  54. [62]

    Macbeth, E

    G. Macbeth, E. Razumiejczyk, R. D. Ledesma, Cliff’s delta calculator: A non-parametric effect size program for two groups of observations, Univer- sitas Psychologica 10 (2) (2011) 545–555

    2011

  55. [63]

    M. R. Hess, J. D. Kromrey, Robust confidence intervals for effect sizes: A comparative study of cohen’sd and cliff’s delta under non-normality and het- erogeneous variances, in: annual meeting of the American Educational Re- search Association, 2004, pp. 1–30

    2004

  56. [64]

    Calmon, D

    F. Calmon, D. Wei, B. Vinzamuri, K. Natesan Ramamurthy, K. R. Varshney, Optimized pre-processing for discrimination prevention, in: I. Guyon, U. V . Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, R. Garnett (Eds.), Advances in Neural Information Processing Systems...

    2017

  57. [65]

    Galhotra, Y

    S. Galhotra, Y . Brun, A. Meliou, Fairness testing: testing software for dis- crimination, Proceedings of the 2017 11th Joint Meeting on Foundations of Software Engineering - ESEC/FSE 2017 (2017). doi:10.1145/3106237. 3106277. URL http://dx.doi.org/10.1145/3106237.3106277

    2017

  58. [66]

    B. H. Zhang, B. Lemoine, M. Mitchell, Mitigating unwanted biases with adversarial learning (2018). arXiv:1801.07593. 36

    2018 arXiv

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.