mlsanityData to 7 Oct 2026

Mitigating Statistical Bias within Differentially Private Synthetic Data

arXiv 2108.10934 ↗First posted 24 Aug 2021stat.ML cs.CR cs.LGcited by 4

Not on the map. Why?

Abstract (arXiv)

Increasing interest in privacy-preserving machine learning has led to new and evolved approaches for generating private synthetic data from undisclosed real data. However, mechanisms of privacy preservation can significantly reduce the utility of synthetic data, which in turn impacts downstream tasks such as learning predictive models or inference. We propose several re-weighting strategies using privatised likelihood ratios that not only mitigate statistical bias of downstream estimators but also have general applicability to differentially private generative models. Through large-scale empirical evaluation, we show that private importance weighting provides simple and effective privacy-compliant augmentation for general applications of synthetic data.

Key facts. Cited by 4 papers in the map (data through 7 Oct 2026). Builds on 24 of its 55 references that are in the map. First posted on arXiv 24 Aug 2021.

Builds on

24 of 55 references.

References · 24

  1. 2016Deep Learning with Differential Privacycited by 2,682
  2. 2017Real-valued (Medical) Time Series Generation with Recurrent Conditional GANscited by 329
  3. 2018Differentially Private Generative Adversarial Networkcited by 235
  4. 2018Improving the Gaussian Mechanism for Differential Privacy: Analytical Calibration and Optimal Denoisingcited by 216
  5. 2013A General Framework for Updating Belief Distributionscited by 172
  6. 2020DP-CGAN: Differentially Private Synthetic Data and Label Generationcited by 118
  7. 2018Private Selection from Private Candidatescited by 75
  8. 2019Bias Correction of Learned Generative Models using Likelihood-Free Importance Weightingcited by 70
  9. 2015Privacy for Free: Posterior Sampling and Stochastic Gradient Monte Carlocited by 61
  10. 2018Discriminator Rejection Samplingcited by 48
  11. 2015Pareto Smoothed Importance Samplingcited by 42
  12. 2018Metropolis-Hastings Generative Adversarial Networkscited by 40
12 more
  1. 2017Differentially Private Mixture of Generative Neural Networkscited by 39
  2. 2021Leveraging Public Data for Practical Private Query Releasecited by 39
  3. 2020Differentially Private Synthetic Data: Applied Evaluations and Enhancementscited by 36
  4. 2016On the Theory and Practice of Privacy-Preserving Bayesian Data Analysiscited by 34
  5. 2017General Bayesian Updating and the Loss-Likelihood Bootstrapcited by 34
  6. 2018Differentially Private Data Generative Modelscited by 33
  7. 2019Differentially Private Generative Adversarial Networks for Time Series, Continuous, and Discrete Open Datacited by 31
  8. 2020Private Post-GAN Boostingcited by 20
  9. 2013Bayesian Differential Privacy through Posterior Samplingcited by 17
  10. 2019Privacy-preserving parametric inference: a case for robust statisticscited by 17
  11. 2018Differentially Private Releasing via Deep Generative Model (Technical Report)cited by 10
  12. 2020Foundations of Bayesian Learning from Synthetic Datacited by 10

Built on by

Cited by year

Citing papers per year: 2023: 2; 2024: 2.

4 papers.

  1. 2023The Real Deal Behind the Artificial Appeal: Inferential Utility of Tabular Synthetic Datacited by 5
  2. 2024Debiasing Synthetic Data Generated by Deep Generative Modelscited by 2
  3. 2023Differentially Private Statistical Inference through β-Divergence One Posterior Samplingcited by 0
  4. 2024A density ratio framework for evaluating the utility of synthetic datacited by 0