mlsanityData to 14 Oct 2026

Privacy Vulnerabilities in Marginals-based Synthetic Data

arXiv 2410.05506 ↗First posted 7 Oct 2024cs.CR cs.LGcited by 10

Not on the map. Why?

Abstract (arXiv)

When acting as a privacy-enhancing technology, synthetic data generation (SDG) aims to maintain a resemblance to the real data while excluding personally-identifiable information. Many SDG algorithms provide robust differential privacy (DP) guarantees to this end. However, we show that the strongest class of SDG algorithms--those that preserve textitmarginal probabilities, or similar statistics, from the underlying data--leak information about individuals that can be recovered more efficiently than previously understood. We demonstrate this by presenting a novel membership inference attack, MAMA-MIA, and evaluate it against three seminal DP SDG algorithms: MST, PrivBayes, and Private-GSD. MAMA-MIA leverages knowledge of which SDG algorithm was used, allowing it to learn information about the hidden data more accurately, and orders-of-magnitude faster, than other leading attacks. We use MAMA-MIA to lend insight into existing SDG vulnerabilities. Our approach went on to win the first SNAKE (SaNitization Algorithm under attacK ... varepsilon) competition.

Key facts. Cited by 10 papers in the map (data through 14 Oct 2026). Builds on 27 of its 63 references that are in the map. First posted on arXiv 7 Oct 2024.

Builds on

27 of 63 references.

References · 27

  1. 2016Membership Inference Attacks against Machine Learning Modelscited by 2,281
  2. 2021Membership Inference Attacks From First Principlescited by 878
  3. 2022TabDDPM: Modelling Tabular Data with Diffusion Modelscited by 377
  4. 2016From Softmax to Sparsemax: A Sparse Model of Attention and Multi-Label Classificationcited by 341
  5. 2015A note on the evaluation of generative modelscited by 292
  6. 2018Differentially Private Generative Adversarial Networkcited by 235
  7. 2023Mixed-Type Tabular Data Synthesis with Score-based Diffusion in Latent Spacecited by 186
  8. 2015On Wasserstein Two Sample Testing and Related Families of Nonparametric Testscited by 166
  9. 2019Reconstruction and Membership Inference Attacks against Generative Modelscited by 122
  10. 2020DP-CGAN: Differentially Private Synthetic Data and Label Generationcited by 118
  11. 2020Synthetic Data -- Anonymisation Groundhog Daycited by 108
  12. 2021Winning the NIST Contest: A scalable and general approach to differentially private synthetic datacited by 102
15 more
  1. 2021Benchmarking Differentially Private Synthetic Data Generation Algorithmscited by 81
  2. 2022Synthetic Data -- what, why and how?cited by 67
  3. 2023Membership Inference Attacks against Synthetic Data through Overfitting Detectioncited by 66
  4. 2019Block Neural Autoregressive Flowcited by 62
  5. 2022TAPAS: a Toolbox for Adversarial Privacy Auditing of Synthetic Datacited by 57
  6. 2021Differentially Private Query Release Through Adaptive Projectioncited by 52
  7. 2023SoK: Privacy-Preserving Data Synthesiscited by 38
  8. 2022Private Synthetic Data for Multitask Learning and Marginal Queriescited by 35
  9. 2023Achilles' Heels: Vulnerable Record Identification in Synthetic Data Publishingcited by 28
  10. 2023Generating Private Synthetic Data with Genetic Algorithmscited by 18
  11. 2023Synthetic is all you need: removing the auxiliary data assumption for membership inference attacks against synthetic datacited by 18
  12. 2024Systematic Assessment of Tabular Data Synthesiscited by 18
  13. 2024The Elusive Pursuit of Reproducing PATE-GAN: Benchmarking, Auditing, Debuggingcited by 13
  14. 2023Assessment of Differentially Private Synthetic Data for Utility and Fairness in End-to-End Machine Learning Pipelines for Tabular Datacited by 8
  15. 2023Generating tabular datasets under differential privacycited by 6

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 5; 2026: 5.

10 papers.

  1. 2025The DCR Delusion: Measuring the Privacy Risk of Synthetic Datacited by 11
  2. 2025Synthetic Tabular Data: Methods, Attacks and Defensescited by 9
  3. 2025The Importance of Being Discrete: Measuring the Impact of Discretization in End-to-End Differentially Private Synthetic Datacited by 5
  4. 2025Privacy Auditing Synthetic Data Release through Local Likelihood Attackscited by 5
  5. 2025TAMIS: Tailored Membership Inference Attacks on Synthetic Datacited by 1
  6. 2026FHAIM: Fully Homomorphic AIM For Private Synthetic Data Generationcited by 1
  7. 2026ReTabSyn: Realistic Tabular Data Synthesis via Reinforcement Learningcited by 1
  8. 2026Risk-Equalized Differentially Private Synthetic Data: Protecting Outliers by Controlling Record-Level Influencecited by 0
  9. 2026SoK: Reconstruction Attacks on Synthetic Tabular Data (Insights from Winning the NIST CRC)cited by 0
  10. 2026Position: Privacy Is a Claim, Not a Property of Synthetic Datacited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024High Epsilon Synthetic Data Vulnerabilities in MST and PrivBayes
  2. 2024"What do you want from theory alone?" Experimenting with Tight Auditing of Differentially Private Synthetic Data Generation
  3. 2026ReMIA: a Powerful and Efficient Alternative to Membership Inference Attacks against Synthetic Data Generators
  4. 2025Understanding the Impact of Data Domain Extraction on Synthetic Data Privacy
  5. 2025Synth-MIA: A Testbed for Auditing Privacy Leakage in Tabular Data Synthesis
  6. 2024SafeSynthDP: Leveraging Large Language Models for Privacy-Preserving Synthetic Data Generation Using Differential Privacy
  7. 2024Lost in the Averages: A New Specific Setup to Evaluate Membership Inference Attacks Against Machine Learning Models
  8. 2025When Privacy Isn't Synthetic: Hidden Data Leakage in Generative AI Models