mlsanityData to 14 Oct 2026

Disparate Impact in Differential Privacy from Gradient Misalignment

arXiv 2206.07737 ↗First posted 15 Jun 2022cs.LG cs.AI cs.CRcited by 31

Not on the map. Why?

Abstract (arXiv)

As machine learning becomes more widespread throughout society, aspects including data privacy and fairness must be carefully considered, and are crucial for deployment in highly regulated industries. Unfortunately, the application of privacy enhancing technologies can worsen unfair tendencies in models. In particular, one of the most widely used techniques for private model training, differentially private stochastic gradient descent (DPSGD), frequently intensifies disparate impact on groups within data. In this work we study the fine-grained causes of unfairness in DPSGD and identify gradient misalignment due to inequitable gradient clipping as the most significant source. This observation leads us to a new method for reducing unfairness by preventing gradient misalignment in DPSGD.

Key facts. Cited by 31 papers in the map (data through 14 Oct 2026). Builds on 24 of its 44 references that are in the map. First posted on arXiv 15 Jun 2022.

Builds on

24 of 44 references.

References · 24

  1. 2016Communication-Efficient Learning of Deep Networks from Decentralized Datacited by 7,669
  2. 2014Deep Learning Face Attributes in the Wildcited by 3,693
  3. 2016Deep Learning with Differential Privacycited by 2,728
  4. 2019A Survey on Bias and Fairness in Machine Learningcited by 1,917
  5. 2018The Secret Sharer: Evaluating and Testing Unintended Memorization in Neural Networkscited by 864
  6. 2017Renyi Differential Privacycited by 650
  7. 2019Differential Privacy Has Disparate Impact on Model Accuracycited by 316
  8. 2019Differentially Private Learning with Adaptive Clippingcited by 220
  9. 2020Auditing Differentially Private Machine Learning: How Private is Private SGD?cited by 218
  10. 2019R'enyi Differential Privacy of the Sampled Gaussian Mechanismcited by 204
  11. 2021Adversary Instantiation: Lower Bounds for Differentially Private Machine Learningcited by 199
  12. 2021A survey on datasets for fairness-aware machine learningcited by 157
12 more
  1. 2020Understanding Gradient Clipping in Private SGD: A Geometric Perspectivecited by 142
  2. 2020On the Privacy Risks of Algorithmic Fairnesscited by 82
  3. 2019AdaCliP: Adaptive Clipping for Private SGDcited by 77
  4. 2018Differentially Private Fair Learningcited by 76
  5. 2019Differentially Private Model Publishing for Deep Learningcited by 71
  6. 2020Neither Private Nor Fair: Impact of Data Imbalance on Utility and Fairness in Differential Privacycited by 61
  7. 2019Fair Decision Making using Privacy-Protected Datacited by 58
  8. 2021Decentralized Federated Learning through Proxy Model Sharingcited by 57
  9. 2020Differentially Private and Fair Deep Learning: A Lagrangian Dual Approachcited by 53
  10. 2020Fair Learning with Private Demographic Datacited by 46
  11. 2021CaPC Learning: Confidential and Private Collaborative Learningcited by 33
  12. 2019Privacy Enhanced Multimodal Neural Representations for Emotion Recognitioncited by 22

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2022: 2; 2023: 5; 2024: 10; 2025: 11; 2026: 3.

31 papers.

Sort:

By area:

  1. 2023Privacy and Fairness in Federated Learning: on the Perspective of Trade-offcited by 26
  2. 2023Fair Machine Unlearning: Data Removal while Mitigating Disparitiescited by 26
  3. 2022Differentially Private Learning with Per-Sample Adaptive Clippingcited by 21
  4. 2024Conformal Prediction Sets Can Cause Disparate Impactcited by 14
  5. 2023Augment then Smooth: Reconciling Differential Privacy with Certified Robustnesscited by 6
  6. 2023SoK: Unintended Interactions among Machine Learning Defenses and Riskscited by 6
  7. 2025Trustworthy AI Must Account for Interactionscited by 5
  8. 2022A Closer Look at the Calibration of Differentially Private Learnerscited by 4
  9. 2023Toward the Tradeoffs between Privacy, Fairness and Utility in Federated Learningcited by 4
  10. 2024Differentially Private Clustered Federated Learningcited by 4
  11. 2024PUFFLE: Balancing Privacy, Utility, and Fairness in Federated Learningcited by 4
  12. 2024Regulation Games for Trustworthy Machine Learningcited by 3
19 more
  1. 2024Differentially Private Fair Binary Classificationscited by 3
  2. 2024PFGuard: A Generative Framework with Privacy and Fairness Safeguardscited by 3
  3. 2025SoK: What Makes Private Learning Unfair?cited by 3
  4. 2024Differentially Private Prototypes for Imbalanced Transfer Learningcited by 2
  5. 2024Combining Machine Learning Defenses without Conflictscited by 2
  6. 2025Learning with Differentially Private (Sliced) Wasserstein Gradientscited by 2
  7. 2025Mitigating Disparate Impact of Differentially Private Learning through Bounded Adaptive Clippingcited by 2
  8. 2024Enhancing DP-SGD through Non-monotonous Adaptive Scaling Gradient Weightcited by 1
  9. 2025Private Rate-Constrained Optimization with Applications to Fair Learningcited by 1
  10. 2025SoftAdaClip: A Smooth Clipping Strategy for Fair and Private Model Trainingcited by 1
  11. 2025Private and Fair Machine Learning: Revisiting the Disparate Impact of Differentially Private SGDcited by 1
  12. 2025On Optimal Hyperparameters for Differentially Private Deep Transfer Learningcited by 1
  13. 2026Differential Privacy in Two-Layer Networks: How DP-SGD Harms Fairness and Robustnesscited by 1
  14. 2024On the Impact of Output Perturbation on Fairness in Binary Linear Classificationcited by 0
  15. 2025Steps Adaptive Decay DPSGD: Enhancing Performance on Imbalanced Datasets with Differential Privacy with HAM10000cited by 0
  16. 2025On the Fairness of Privacy Protection: Measuring and Mitigating the Disparity of Group Privacy Risks for Differentially Private Machine Learningcited by 0
  17. 2025FedPF: Accurate Target Privacy Preserving Federated Learning Balancing Fairness and Utilitycited by 0
  18. 2026Federated Learning at the Forefront of Fairness: A Multifaceted Perspectivecited by 0
  19. 2026Adaptive Sampling and Clipping for Private Worst-Case Group Optimizationcited by 0