Quantitative Auditing of AI Fairness with Differentially Private Synthetic Data

Bow-Yaw Wang; Chih-Cheng Rex Yuan

REVIEW 4 major objections 6 minor 65 references

Quantitative Auditing of AI Fairness with Differentially Private Synthetic Data

T0 review · 4 major / 6 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 Differentially private synthetic data can stand in for real data in AI fairness audits, keeping average metric error below 0.1.

desk verdict A useful empirical validation of a DP-synthetic-data auditing workflow, but the headline claim that fairness is 'generally preserved' is not supported by the paper's own tables. read the letter →

arxiv 2504.21634 v1 pith:MGNFIDTJ submitted 2025-04-30 cs.CY cs.AIcs.LG

Chih-Cheng Rex Yuan , Bow-Yaw Wang This is my paper · ORCID

classification cs.CYcs.AIcs.LG

keywords fairnessauditingdifferentialprivacysyntheticdataMarkovrandomfieldRényimetricsconditionaluseaccuracyequalityequalizedodds

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

The paper tries to establish that an auditor can replace a sensitive real dataset with a differentially private synthetic copy, discard the original, and still reach trustworthy conclusions about an AI system's fairness. It proposes a workflow in which the auditor generates the synthetic data immediately upon receiving the real data and keeps only the synthetic version for all future audits. Across the Adult, COMPAS, and Diabetes datasets and several models, the average absolute difference between fairness measures computed on real and synthetic data stays below 0.1. This would let third-party fairness audits shed their main privacy liability: auditors would no longer need to be custodians of raw sensitive records.

What carries the argument

The engine is the select-measure-generate method for differentially private synthetic data, built around a Markov random field—a graphical model whose nodes are attributes and whose edges are dependencies preserved in the synthetic output. The selection step computes pairwise mutual information between all attributes, takes a maximum spanning tree over those weights, and protects every 1-way marginal plus the 2-way marginals on the tree's edges; by default the edge between the protected attribute and the ground truth is not selected. The measure step adds Gaussian noise calibrated to Rényi differential privacy, splitting the privacy budget equally between the 1-way marginals and the selected cliques. The generate step samples new rows from the fitted Markov random field. This construction lets the synthetic data inherit the joint statistics that demographic parity, equalized odds, and accuracy equality depend on, while severing the record-level link to real individuals.

What would settle it

Run the same pipeline on a dataset in which the protected attribute influences the label only through a three-way interaction that no single tree edge captures; if the average difference between synthetic and real fairness scores then exceeds 0.1 for demographic parity or equalized odds, the claimed approximation would be shown not to generalize. A cheaper check is to rerun the authors' notebooks on a dataset with more than a dozen attributes and compare the error distribution.

Watch

Extended reading notes

Core claim

The central finding is that fairness-relevant statistics of a dataset survive differentially private synthesis well enough for auditing. Using a select-measure-generate pipeline, the authors fit a Markov random field to noise-perturbed marginals selected by a maximum-spanning-tree heuristic, generate 100 synthetic datasets per (dataset, model) pair, and compute six fairness measures on each. Averaged over those runs, the absolute difference between synthetic and original fairness scores is below 0.1 for every dataset and model tested. The paper also identifies a structural exception: sufficiency measures such as conditional use accuracy equality condition on the model's prediction, which is not present in the original data and therefore is not captured by the marginals; those measures show larger errors, up to roughly 0.32 in one Adult experiment. The authors conclude that the synthetic-data workflow remains valid for auditing and that a known edge between the protected attribute and the ground truth can be added to reduce the sufficiency-measure error.

Load-bearing premise

The argument rests on the assumption that the one-way counts and pairwise links chosen by the maximum-spanning-tree heuristic preserve how the protected attribute, the true label, and the model's predictions depend on one another; the paper's own results show this assumption breaks for sufficiency metrics, whose errors reach about 0.32 when the protected-attribute/ground-truth link is not encoded.

Editorial extensions

If this is right

Auditors can adopt a generate-then-discard protocol: synthesize on receipt of the data, erase the originals, and keep the synthetic copy for repeated audits of future models.
Independence and separation metrics—demographic parity, equalized odds, and overall accuracy equality—are reproduced closely enough on the tested datasets that a fairness violation visible in real data should remain visible in synthetic data.
Sufficiency metrics such as conditional use accuracy equality should be interpreted cautiously in this workflow, since their errors can exceed the 0.1 average by a wide margin.
The framework requires no dataset-specific tuning beyond choosing the protected attribute, so it transfers to a new dataset and model without re-engineering.
With the Gaussian mechanism under Rényi differential privacy at a fixed budget, the auditor can state a formal bound on what the released synthetic data reveals about any individual.

Reading between the lines

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

Inference: because the 0.1-bound result is empirical on three tabular datasets, a natural next test is high-dimensional or text-derived data, where the maximum-spanning-tree marginal set will omit more of the joint structure that fairness metrics may depend on.
Inference: the paper's diagnosis of sufficiency-metric errors implies an inexpensive remedy an auditor could adopt without changing the pipeline—include the protected-attribute/ground-truth clique explicitly, at the cost of a small slice of the privacy budget.
Inference: the privacy guarantee protects individuals in the training data, but the published fairness scores themselves are aggregate statistics; inference attacks on those statistics remain outside what this framework claims to prevent.

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.

Desk Editor's Note

read the letter

Quick take: this is a concrete empirical study of an operational workflow—generate DP synthetic data, discard the raw data, and compute fairness metrics on the synthetic data—but the paper's central claim that synthetic data "generally preserves" fairness is overstated. The average-difference criterion hides per-metric errors that would flip an audit conclusion.

What is actually new: the tripartite framework (data provider, model maker, auditor) is clearly described, and the idea of discarding the real data immediately after synthesis is a defensible operational proposal. The paper reuses Private-PGM (the NIST 2018 winner) and the authors' own fairness checker, both appropriate choices. Running 100 synthetic datasets per model and averaging is good practice. The intuition about Conditional Use Accuracy Equality being hardest to approximate—because it conditions on Y_hat in both numerator and denominator—is reasonable and partially supported.

The soft spots are real. In Table 5, Demographic Parity on Adult-RF differs by 0.18; in Table 6, Conditional Use Accuracy Equality (TP) differs by 0.32. Those are not small. The paper's own Section 7.1 admits that adding the (S,Y) edge improves accuracy, but the results of that experiment are not reported. There are no confidence intervals or standard deviations across the 100 runs, and no code link is given despite the claim that the notebooks are public. The stress-test point is correct: the maximum-spanning-tree marginal selection does not guarantee preservation of P(X|S), so even demographic parity can fail badly. The boxed-variable analysis in Section 7 is too weak to explain why demographic parity fails.

The citation pattern is fine. The authors cite the existing work that already raised the fairness-of-DP-synthetic-data question [13, 23, 61] and position their contribution as an evaluation rather than a new method. Self-citation of their own fairness checker is acceptable.

The paper deserves a serious referee. It is an honest empirical validation of a niche but growing practice, and the flaws are correctable: report per-metric errors with uncertainty, include the (S,Y) experiment, release code with a hash, and soften the conclusion to name the metrics and settings where synthetic auditing is reliable. As is, it would need heavy revision; as a case study with those fixes, it would be a solid contribution to the DP-synthetic-data and fairness-auditing community.

Recommendation: send to peer review, but require the revisions above before acceptance.

Referee Report

4 major / 6 minor

Summary. The paper proposes a framework for fairness auditing of AI systems without retaining real data: an auditor generates differentially private synthetic data from the real data using the Private-PGM tool (winner of the 2018 NIST challenge), discards the real data, and evaluates an AI model with a fairness checker using standard metrics (demographic parity, equalized odds, conditional use accuracy equality, overall accuracy equality). The empirical evaluation compares fairness metrics computed on original versus synthetic data for Adult, COMPAS, and Diabetes datasets across several classifiers. The paper reports that the average absolute difference across six metrics is below 0.1 for every dataset/model combination and concludes that synthetic data generally preserves fairness properties well enough for meaningful auditing, while acknowledging in Section 7.2 that Conditional Use Accuracy Equality is harder to approximate.

Significance. If the central claim were established, this would be a practically valuable contribution: it would let third-party auditors avoid custody of sensitive data while retaining quantitative fairness evaluation, with a formal DP guarantee and an open-source implementation. The paper has real strengths: the fairness measures are standard definitions, the comparison against real-data metrics is an external benchmark rather than a circular evaluation, the synthesis pipeline is the well-known NIST-winning Private-PGM, and the code is publicly available. However, the reported results do not support the headline claim as stated; the average-difference criterion hides large per-metric errors that can reverse an audit verdict, so the significance hinges on a substantial reframing or additional analysis.

major comments (4)

[Section 7.1, Tables 5-7] The criterion 'the average of the differences ... is below 0.1' masks errors that are decisive for audit outcomes. In Table 5 (Adult, random forest), Demographic Parity is 0.1933 on the original data and 0.0114 on synthetic data, a difference of 0.1818; under a common epsilon=0.1 threshold the original data would be flagged as unfair while the synthetic data would not. Table 6 (Adult, logistic regression) shows a Conditional Use Accuracy Equality (TP) difference of 0.3154, and Table 7 (Adult, KNN) shows Demographic Parity and Equalized Odds (TP) differences of 0.1675 and 0.1557. These are not small perturbations; they are large enough to flip the conclusion of a fairness audit, so 'generally preserves fairness with a reasonable degree of accuracy' (Section 7.1) is not supported by the paper's own tables.
[Section 7.2, Tables 5-9] The proposed explanation that Conditional Use Accuracy Equality should be hardest because the predicted outcome appears in both numerator and denominator is contradicted by the data: Demographic Parity and Equalized Odds also show large errors (e.g., Table 7 differences of 0.1675 and 0.1557; Table 5 difference of 0.1818 for Demographic Parity). The more fundamental issue is structural: the model predictions Yhat are not part of the data at synthesis time, so the synthetic data does not directly preserve P(Yhat | S). The default maximum-spanning-tree selection in Section 5.1 retains pairwise marginals only, and Section 7.1 admits that adding the (S,Y) edge would improve accuracy; the paper should analyze this dependence rather than attributing the failure to the fairness measure's algebraic form.
[Section 6, paragraph on 100 runs] The paper averages fairness scores over 100 synthetic datasets but reports no standard deviation, range, or confidence interval. For a quantitative audit, an average error below 0.1 is not sufficient: the auditor needs to know how much a single synthetic dataset can deviate from the real-data value, because the proposed workflow uses one generated dataset after discarding the real data. Without this information, the framework's reliability for a single audit is not established.
[Section 5.1 and Section 7.1, added (S,Y) edge] The statement that adding the (S,Y) edge 'would have been more accurate' is not quantified anywhere; no table, figure, or magnitude accompanies it. Since the framework discards the real data immediately after synthesis, the auditor cannot generally know which higher-order edges are needed, so the default configuration is what matters. The paper should either validate the default workflow with decision-focused error bounds or explicitly restrict the framework's applicability to fairness measures that are robust to the default marginals.

minor comments (6)

[Section 3.1] The phrase 'predicated result' should be 'predicted result'.
[Section 5.1] The RDP guarantee statement 'this procedure satisfies (α, α/2σ^2)-RDP' appears to omit the sensitivity Δ_f that appears in Theorem 3.3; please reconcile the notation.
[Section 6] The dataset descriptions are imprecise (e.g., Adult 'about 30000 individuals' and COMPAS 'about 7000'); specify exact versions, preprocessing, and train/test splits to ensure reproducibility.
[Tables 5-9] Consider adding standard deviations or ranges across the 100 synthetic datasets, and provide a threshold context (e.g., epsilon=0.1) when interpreting the size of the differences.
[Section 7.1] The phrase 'by common sense or in an alternative assumption' is informal for a research paper; specify the exact procedure for adding the (S,Y) edge.
[Section 8] The statement that 'biases present in the original data are captured and reflected in the synthetic counterpart' is stronger than the experimental results support; qualify it with the exceptions shown in Tables 5-9.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central claim is benchmarked against fairness metrics computed on the original real data, an external ground truth.

full rationale

The paper's central claim is an empirical comparison: fairness measures computed on differentially private synthetic data are compared against the same measures computed on the original real datasets (Tables 5-9), and the reported 'average difference below 0.1' is a description of that external benchmark, not a quantity derived from the method's assumptions. The synthesis engine is the NIST-winning Private-PGM implementation (refs 35, 39), which is not self-authored, and the fairness measures are standard definitions given in Table 2. The only self-citation is reference [64], used for the fairness checker and the tripartite auditing framing; the checker merely evaluates standard parity, separation, and sufficiency quantities, and its authorship does not force any numerical outcome. No fitted parameter is relabeled as a prediction: model predictions Y-hat come from externally trained Kaggle models, and fairness values on synthetic data are computed rather than derived from the selected marginals by construction. Concerns about per-metric errors, the maximum-spanning-tree marginal selection, or the difficulty of preserving P(S,Y-hat) are correctness and utility risks, not circularity. The derivation chain is therefore self-contained with respect to its benchmark, and no circular step is exhibited.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central claim depends on two hand-set choices: a privacy budget of epsilon=1 and the set of marginals used for synthesis. No privacy-utility sweep is shown, and the main experiments deliberately omit the (S,Y) edge that fairness metrics disproportionately depend on, with an unreported follow-up that adds it. The synthesis relies on standard RDP results and on the assumption that MST-selected marginals suffice for fairness preservation. No invented entities are introduced.

free parameters (2)

Privacy budget epsilon = 1 (half on 1-way marginals, half on cliques)
Chosen by hand for all experiments; no sweep of epsilon is reported, so the claim that the framework balances privacy and utility is not parameterized.
MRF clique set (marginals) = All 1-way marginals plus MST-selected 2-way cliques; (S,Y) edge excluded in main results
The choice of which marginals to preserve determines how well fairness metrics are reproduced. The main experiments deliberately exclude the (S,Y) edge, which is the exact relationship many fairness metrics depend on; an unreported follow-up adds it.

assumptions (3)

standard math Standard definitions of Rényi differential privacy and the Gaussian mechanism with (alpha, alpha/(2 sigma^2))-RDP guarantee from [39, 41].
Invoked in Section 3.2 and Section 5.1 to claim the synthesis procedure is DP; no new privacy proof is given.
domain assumption The select-measure-generate pipeline with marginals selected by mutual information maximum spanning tree preserves enough joint structure to compute fairness metrics.
This is the core modeling assumption underlying the framework. Section 5.1 describes MST selection without (S,Y); Section 7 shows the assumption fails for sufficiency metrics, and the paper only partially remediates it with an unreported (S,Y) experiment.
domain assumption Kaggle-sourced models are representative AI systems whose fairness can be audited via the fairness checker.
Section 6 states models are 'extracted from various authors from Kaggle' and finetuned; no details on training or validation are provided, and the models are not released.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Quantitative Auditing of AI Fairness with Differentially Private Synthetic Data." pith.science (2026). https://pith.science/paper/MGNFIDTJ

@misc{pith2026250421634,
  author       = {Pith},
  title        = {Pith review of: Quantitative Auditing of AI Fairness with Differentially Private Synthetic Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MGNFIDTJ}},
  note         = {Machine review of arXiv:2504.21634}
}

read the original abstract

Fairness auditing of AI systems can identify and quantify biases. However, traditional auditing using real-world data raises security and privacy concerns. It exposes auditors to security risks as they become custodians of sensitive information and targets for cyberattacks. Privacy risks arise even without direct breaches, as data analyses can inadvertently expose confidential information. To address these, we propose a framework that leverages differentially private synthetic data to audit the fairness of AI systems. By applying privacy-preserving mechanisms, it generates synthetic data that mirrors the statistical properties of the original dataset while ensuring privacy. This method balances the goal of rigorous fairness auditing and the need for strong privacy protections. Through experiments on real datasets like Adult, COMPAS, and Diabetes, we compare fairness metrics of synthetic and real data. By analyzing the alignment and discrepancies between these metrics, we assess the capacity of synthetic data to preserve the fairness properties of real data. Our results demonstrate the framework's ability to enable meaningful fairness evaluations while safeguarding sensitive information, proving its applicability across critical and sensitive domains.

Figures

Figures reproduced from arXiv: 2504.21634 by the authors.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

65 extracted references · 41 canonical work pages

[1]

Christian Arnold and Marcel Neunhoeffer. 2020. Really Useful Synthetic Data–A Framework to Evaluate the Quality of Differentially Private Synthetic Data. arXiv preprint arXiv:2004.07740 (2020)

arXiv 2020
[2]

Prashant Banerjee. 2019. EDA, Logistic Regression, PCA. https://www.kaggle.com/code/prashant111/eda-logistic-regression-pca Accessed: 2024-11-10

work page 2019
[3]

Boaz Barak, Kamalika Chaudhuri, Cynthia Dwork, Satyen Kale, Frank McSherry, and Kunal Talwar. 2007. Privacy, accuracy, and consistency too: a holistic solution to contingency table release. In Proceedings of the twenty-sixth ACM SIGMOD-SIGACT-SIGART symposium on Principles of database systems. 273–282

work page 2007
[4]

Solon Barocas, Moritz Hardt, and Arvind Narayanan. 2023. Fairness and machine learning: Limitations and opportunities . MIT press

2023
[5]

Joachim Baumann, Alessandro Castelnovo, Riccardo Crupi, Nicole Inverardi, and Daniele Regoli. 2023. Bias on demand: a modelling framework that generates synthetic data with bias. In Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency . 1002–1013

work page 2023
[6]

Barry Becker and Ronny Kohavi. 1996. Adult. UCI Machine Learning Repository. DOI: https://doi.org/10.24432/C5XW20

doi:10.24432/c5xw20 1996
[7]

Rachel KE Bellamy, Kuntal Dey, Michael Hind, Samuel C Hoffman, Stephanie Houde, Kalapriya Kannan, Pranay Lohia, Jacquelyn Martino, Sameep Mehta, Aleksandra Mojsilović, et al. 2019. AI Fairness 360: An extensible toolkit for detecting and mitigating algorithmic bias. IBM Journal of Research and Development 63, 4/5 (2019), 4–1

2019
[8]

Richard Berk, Hoda Heidari, Shahin Jabbari, Michael Kearns, and Aaron Roth. 2021. Fairness in criminal justice risk assessments: The state of the art. Sociological Methods & Research 50, 1 (2021), 3–44

2021

Show all 65 references

[9]

Sarah Bird, Miro Dudík, Richard Edgar, Brandon Horn, Roman Lutz, Vanessa Milan, Mehrnoosh Sameki, Hanna Wallach, and Kathleen Walker. 2020. Fairlearn: A toolkit for assessing and improving fairness in AI. Microsoft, Tech. Rep. MSR-TR-2020-32 (2020)

2020
[10]

Claire McKay Bowen and Fang Liu. 2020. Comparative study of differentially private data synthesis methods. Statist. Sci. 35, 2 (2020), 280–307. Manuscript submitted to ACM Quantitative Auditing of AI Fairness with Differentially Private Synthetic Data 15

2020
[11]

Claire McKay Bowen and Joshua Snoke. 2019. Comparative study of differentially private synthetic data algorithms from the NIST PSCR differential privacy synthetic data challenge. arXiv preprint arXiv:1911.12704 (2019)

2019 arXiv
[12]

Isaac Byrne. 2017. Income Prediction (84.369% Accuracy). https://www.kaggle.com/code/ipbyrne/income-prediction-84-369-accuracy Accessed: 2024-11-10

2017
[13]

Victoria Cheng, Vinith M Suriyakumar, Natalie Dullerud, Shalmali Joshi, and Marzyeh Ghassemi. 2021. Can you fake it until you make it? impacts of differentially private synthetic data on downstream classification fairness. In Proceedings of the 2021 ACM Conference on Fairness,...

2021
[14]

Alexandra Chouldechova. 2017. Fair prediction with disparate impact: A study of bias in recidivism prediction instruments. Big data 5, 2 (2017), 153–163

2017
[15]

Sam Corbett-Davies, Emma Pierson, Avi Feller, Sharad Goel, and Aziz Huq. 2017. Algorithmic decision making and the cost of fairness. InProceedings of the 23rd acm sigkdd international conference on knowledge discovery and data mining . 797–806

2017
[16]

Damien Desfontaines and Balázs Pejó. 2019. Sok: differential privacies. arXiv preprint arXiv:1906.01337 (2019)

2019 arXiv
[17]

Jinshuo Dong, Aaron Roth, and Weijie J Su. 2022. Gaussian differential privacy.Journal of the Royal Statistical Society: Series B (Statistical Methodology) 84, 1 (2022), 3–37

2022
[18]

Cynthia Dwork, Frank McSherry, Kobbi Nissim, and Adam Smith. 2006. Calibrating noise to sensitivity in private data analysis. In Theory of Cryptography: Third Theory of Cryptography Conference, TCC 2006, New York, NY, USA, March 4-7, 2006. Proceedings 3 . Springer, 265–284

2006
[19]

Cynthia Dwork, Aaron Roth, et al. 2014. The algorithmic foundations of differential privacy. Foundations and Trends® in Theoretical Computer Science 9, 3–4 (2014), 211–407

2014
[20]

Ju Fan, Tongyu Liu, Guoliang Li, Junyou Chen, Yuwei Shen, and Xiaoyong Du. 2020. Relational data synthesis using generative adversarial networks: A design space exploration. arXiv preprint arXiv:2008.12763 (2020)

2020 arXiv
[21]

Vitaly Feldman, Ilya Mironov, Kunal Talwar, and Abhradeep Thakurta. 2018. Privacy amplification by iteration. In2018 IEEE 59th Annual Symposium on Foundations of Computer Science (FOCS) . IEEE, 521–532

2018
[22]

Emilio Ferrara. 2023. Fairness and bias in artificial intelligence: A brief survey of sources, impacts, and mitigation strategies. Sci 6, 1 (2023), 3

2023
[23]

Georgi Ganev, Bristena Oprisanu, and Emiliano De Cristofaro. 2022. Robin hood and matthew effects: Differential privacy has disparate impact on synthetic data. In International Conference on Machine Learning . PMLR, 6944–6959

2022
[24]

Moritz Hardt, Eric Price, and Nati Srebro. 2016. Equality of opportunity in supervised learning. Advances in neural information processing systems 29 (2016)

2016
[25]

Honglu Jiang, Jian Pei, Dongxiao Yu, Jiguo Yu, Bei Gong, and Xiuzhen Cheng. 2021. Applications of differential privacy in social network analysis: A survey. IEEE transactions on knowledge and data engineering 35, 1 (2021), 108–127

2021
[26]

Lan Jiang, Clara Belitz, and Nigel Bosch. 2024. Synthetic Dataset Generation for Fairer Unfairness Research. In Proceedings of the 14th Learning Analytics and Knowledge Conference. 200–209

2024
[27]

Michael Kahn. [n. d.]. Diabetes. UCI Machine Learning Repository. DOI: https://doi.org/10.24432/C5T59G

doi
[28]

Daniel Kifer and Ashwin Machanavajjhala. 2012. A rigorous and customizable framework for privacy. In Proceedings of the 31st ACM SIGMOD- SIGACT-SIGAI symposium on Principles of Database Systems . 77–88

2012
[29]

Jon Kleinberg, Sendhil Mullainathan, and Manish Raghavan. 2016. Inherent trade-offs in the fair determination of risk scores. arXiv preprint arXiv:1609.05807 (2016)

2016 arXiv
[30]

Daphane Koller. 2009. Probabilistic Graphical Models: Principles and Techniques

2009
[31]

Jil Kothari. 2020. Prediction of income with 98% accuracy! https://www.kaggle.com/code/jilkothari/prediction-of-income-with-98-accuracy Accessed: 2024-11-10

2020
[32]

Jeff Larson, Surya Mattu, Lauren Kirchner, and Julia Angwin. 2016. ProPublica Compas Analysis—Data and Analysis for ‘Machine Bias’. https: //github.com/propublica/compas-analysis

2016
[33]

Yingzhen Li and Richard E Turner. 2016. Rényi divergence variational inference. Advances in neural information processing systems 29 (2016)

2016
[34]

Yingzhou Lu, Minjie Shen, Huazheng Wang, Xiao Wang, Capucine van Rechem, Tianfan Fu, and Wenqi Wei. 2023. Machine learning for synthetic data generation: a review. arXiv preprint arXiv:2302.04062 (2023)

2023 arXiv
[35]

Ryan McKenna. 2018. Differential Privacy Synthetic Data Challenge Algorithms. https://github.com/usnistgov/PrivacyEngCollabSpace/tree/master/ tools/de-identification/Differential-Privacy-Synthetic-Data-Challenge-Algorithms/rmckenna. Accessed: 2024-12-05

2018
[36]

Ryan McKenna. 2021. Marginal-based Methods for Differentially Private Synthetic Data. https://www.youtube.com/watch?v=UKzh9QgNRxA. Accessed: 2024-11-26

2021
[37]

Ryan McKenna and Terrance Liu. 2022. A Simple Recipe for Private Synthetic Data Generation. https://differentialprivacy.org/synth-data-1/ Accessed: 2024-11-10

2022
[38]

Ryan McKenna, Gerome Miklau, and Daniel Sheldon. 2021. Private-PGM. https://github.com/journalprivacyconfidentiality/private-pgm-jpc- 778/tree/v2021-10-04-jpc

2021
[39]

Ryan McKenna, Gerome Miklau, and Daniel Sheldon. 2021. Winning the NIST Contest: A scalable and general approach to differentially private synthetic data. arXiv preprint arXiv:2108.04978 (2021)

2021 arXiv
[40]

Ryan McKenna, Daniel Sheldon, and Gerome Miklau. 2019. Graphical-model based estimation and inference for differential privacy. In International Conference on Machine Learning . PMLR, 4435–4444. Manuscript submitted to ACM 16 Chih-Cheng Rex Yuan and Bow-Yaw Wang

2019
[41]

Ilya Mironov. 2017. Rényi differential privacy. In 2017 IEEE 30th computer security foundations symposium (CSF) . IEEE, 263–275

2017
[42]

Kevin P Murphy. 2023. Probabilistic machine learning: Advanced topics . MIT press

2023
[43]

Mohammed Mustafa. 2022. Diabetes Prediction Dataset. https://www.kaggle.com/datasets/iammustafatz/diabetes-prediction-dataset Accessed: 2024-11-26

2022
[44]

National Institute of Standards and Technology. 2018. 2018 Differential Privacy Synthetic Data Challenge. https://www.nist.gov/ctl/pscr/open- innovation-prize-challenges/past-prize-challenges/2018-differential-privacy-synthetic Accessed: 2024-11-10

2018
[45]

Joseph Near and David Darais. 2021. Differentially Private Synthetic Data. https://www.nist.gov/blogs/cybersecurity-insights/differentially-private- synthetic-data. Accessed: 2024-11-26

2021
[46]

Dan Ofer. 2017. COMPAS Dataset. https://www.kaggle.com/datasets/danofer/compass Accessed: 2024-11-26

2017
[47]

Dana Pessach and Erez Shmueli. 2022. A review on fairness in machine learning. ACM Computing Surveys (CSUR) 55, 3 (2022), 1–44

2022
[48]

Trivellore E Raghunathan. 2021. Synthetic data. Annual review of statistics and its application 8, 1 (2021), 129–140

2021
[49]

UC Irvine Machine Learning Repository. 2016. Adult Census Income Dataset. https://www.kaggle.com/datasets/uciml/adult-census-income/data Accessed: 2024-11-26

2016
[50]

Lucas Rosenblatt, Xiaoyan Liu, Samira Pouyanfar, Eduardo de Leon, Anuj Desai, and Joshua Allen. 2020. Differentially private synthetic data: Applied evaluations and enhancements. arXiv preprint arXiv:2011.05537 (2020)

2020 arXiv
[51]

Panjawat Rungtranont. 2023. Diabetes: EDA | Random Forest + HP. https://www.kaggle.com/code/tumpanjawat/diabetes-eda-random-forest-hp Accessed: 2024-11-26

2023
[52]

Pedro Saleiro, Benedict Kuester, Loren Hinkson, Jesse London, Abby Stevens, Ari Anisfeld, Kit T Rodolfa, and Rayid Ghani. 2018. Aequitas: A bias and fairness audit toolkit. arXiv preprint arXiv:1811.05577 (2018)

2018 arXiv
[53]

Innocent Sky. 2024. Logistics Regression COMPAS Recidivism Racial Bias. https://www.kaggle.com/code/innocentsky/logistics-regression-compas- recidivism-racial-bias Accessed: 2024-11-26

2024
[54]

Theresa Stadler, Bristena Oprisanu, and Carmela Troncoso. 2022. Synthetic data–anonymisation groundhog day. In31st USENIX Security Symposium (USENIX Security 22). 1451–1468

2022
[55]

Yuchao Tao, Ryan McKenna, Michael Hay, Ashwin Machanavajjhala, and Gerome Miklau. 2021. Benchmarking differentially private synthetic data generation algorithms. arXiv preprint arXiv:2112.09238 (2021)

2021 arXiv
[56]

Aleksei Triastcyn and Boi Faltings. 2020. Bayesian differential privacy for machine learning. In International Conference on Machine Learning . PMLR, 9583–9592

2020
[57]

Jonathan Ullman. 2022. What is Synthetic Data? https://differentialprivacy.org/synth-data-0/ Accessed: 2024-11-10

2022
[58]

Tim van Erven and Peter Harremoës. 2010. Rényi divergence and majorization. In 2010 IEEE International Symposium on Information Theory . IEEE, 1335–1339

2010
[59]

Tim Van Erven and Peter Harremos. 2014. Rényi divergence and Kullback-Leibler divergence. IEEE Transactions on Information Theory 60, 7 (2014), 3797–3820

2014
[60]

Cedric Deslandes Whitney and Justin Norman. 2024. Real Risks of Fake Data: Synthetic Data, Diversity-Washing and Consent Circumvention. In The 2024 ACM Conference on Fairness, Accountability, and Transparency . 1733–1744

2024
[61]

Sierra Wyllie, Ilia Shumailov, and Nicolas Papernot. 2024. Fairness feedback loops: training on synthetic data amplifies bias. In The 2024 ACM Conference on Fairness, Accountability, and Transparency . 2113–2147

2024
[62]

Lei Xu, Maria Skoularidou, Alfredo Cuesta-Infante, and Kalyan Veeramachaneni. 2019. Modeling tabular data using conditional gan. Advances in neural information processing systems 32 (2019)

2019
[63]

Min-Hsuan Yeh, Blossom Metevier, Austin Hoag, and Philip Thomas. 2024. Analyzing the Relationship Between Difference and Ratio-Based Fairness Metrics. In The 2024 ACM Conference on Fairness, Accountability, and Transparency . 518–528

2024
[64]

Chih-Cheng Rex Yuan and Bow-Yaw Wang. 2024. Ensuring Fairness with Transparent Auditing of Quantitative Bias in AI Systems. In2024 Pacific Neighborhood Consortium Annual Conference and Joint Meetings (PNC) . IEEE, 25–32

2024
[65]

Indr ˙e Žliobait˙e. 2017. Measuring discrimination in algorithmic decision making. Data Mining and Knowledge Discovery 31, 4 (2017), 1060–1089. Manuscript submitted to ACM

2017

Pith tools

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

[1] [1]

Christian Arnold and Marcel Neunhoeffer. 2020. Really Useful Synthetic Data–A Framework to Evaluate the Quality of Differentially Private Synthetic Data. arXiv preprint arXiv:2004.07740 (2020)

arXiv 2020

[2] [2]

Prashant Banerjee. 2019. EDA, Logistic Regression, PCA. https://www.kaggle.com/code/prashant111/eda-logistic-regression-pca Accessed: 2024-11-10

work page 2019

[3] [3]

Boaz Barak, Kamalika Chaudhuri, Cynthia Dwork, Satyen Kale, Frank McSherry, and Kunal Talwar. 2007. Privacy, accuracy, and consistency too: a holistic solution to contingency table release. In Proceedings of the twenty-sixth ACM SIGMOD-SIGACT-SIGART symposium on Principles of database systems. 273–282

work page 2007

[4] [4]

Solon Barocas, Moritz Hardt, and Arvind Narayanan. 2023. Fairness and machine learning: Limitations and opportunities . MIT press

2023

[5] [5]

Joachim Baumann, Alessandro Castelnovo, Riccardo Crupi, Nicole Inverardi, and Daniele Regoli. 2023. Bias on demand: a modelling framework that generates synthetic data with bias. In Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency . 1002–1013

work page 2023

[6] [6]

Barry Becker and Ronny Kohavi. 1996. Adult. UCI Machine Learning Repository. DOI: https://doi.org/10.24432/C5XW20

doi:10.24432/c5xw20 1996

[7] [7]

Rachel KE Bellamy, Kuntal Dey, Michael Hind, Samuel C Hoffman, Stephanie Houde, Kalapriya Kannan, Pranay Lohia, Jacquelyn Martino, Sameep Mehta, Aleksandra Mojsilović, et al. 2019. AI Fairness 360: An extensible toolkit for detecting and mitigating algorithmic bias. IBM Journal of Research and Development 63, 4/5 (2019), 4–1

2019

[8] [8]

Richard Berk, Hoda Heidari, Shahin Jabbari, Michael Kearns, and Aaron Roth. 2021. Fairness in criminal justice risk assessments: The state of the art. Sociological Methods & Research 50, 1 (2021), 3–44

2021