mlsanityData to 14 Oct 2026

A Fairness Analysis on Private Aggregation of Teacher Ensembles

arXiv 2109.08630 ↗First posted 17 Sep 2021cs.LG cs.AI cs.CRcited by 11

Not on the map. Why?

Abstract (arXiv)

The Private Aggregation of Teacher Ensembles (PATE) is an important private machine learning framework. It combines multiple learning models used as teachers for a student model that learns to predict an output chosen by noisy voting among the teachers. The resulting model satisfies differential privacy and has been shown effective in learning high-quality private models in semisupervised settings or when one wishes to protect the data labels. This paper asks whether this privacy-preserving framework introduces or exacerbates bias and unfairness and shows that PATE can introduce accuracy disparity among individuals and groups of individuals. The paper analyzes which algorithmic and data properties are responsible for the disproportionate impacts, why these aspects are affecting different groups disproportionately, and proposes guidelines to mitigate these effects. The proposed approach is evaluated on several datasets and settings.

Key facts. Cited by 11 papers in the map (data through 14 Oct 2026). Builds on 16 of its 31 references that are in the map. First posted on arXiv 17 Sep 2021.

Builds on

16 of 31 references.

References · 16

  1. 2015Distillation as a Defense to Adversarial Perturbations against Deep Neural Networkscited by 678
  2. 2017Renyi Differential Privacycited by 650
  3. 2019Differential Privacy Has Disparate Impact on Model Accuracycited by 316
  4. 2018Scalable Private Learning with PATEcited by 269
  5. 2020Fairness without Demographics through Adversarially Reweighted Learningcited by 251
  6. 2018Differentially Private Empirical Risk Minimization Revisited: Faster and More Generalcited by 128
  7. 2018Differentially Private Fair Learningcited by 76
  8. 2019Training individually fair ML models with Sensitive Subspace Robustnesscited by 73
  9. 2017Efficient Private ERM for Smooth Objectivescited by 61
  10. 2020Neither Private Nor Fair: Impact of Data Imbalance on Utility and Fairness in Differential Privacycited by 61
  11. 2019Fair Decision Making using Privacy-Protected Datacited by 58
  12. 2020Differentially Private and Fair Deep Learning: A Lagrangian Dual Approachcited by 53
4 more
  1. 2020Fair Learning with Private Demographic Datacited by 46
  2. 2021DP-SGD vs PATE: Which Has Less Disparate Impact on Model Accuracy?cited by 33
  3. 2021Decision Making with Differential Privacy under a Fairness Lenscited by 32
  4. 2021AI-SARAH: Adaptive and Implicit Stochastic Recursive Gradient Methodscited by 15

Built on by

Cited by year

Citing papers per year: 2021: 2; 2022: 3; 2023: 5; 2024: 0; 2025: 1.

11 papers.

  1. 2021Robin Hood and Matthew Effects: Differential Privacy Has Disparate Impact on Synthetic Datacited by 48
  2. 2022Differential Privacy and Fairness in Decisions and Learning Tasks: A Surveycited by 47
  3. 2023Learning with Impartiality to Walk on the Pareto Frontier of Fairness, Privacy, and Utilitycited by 10
  4. 2022On the Alignment of Group Fairness with Attribute Privacycited by 9
  5. 2023An Empirical Analysis of Fairness Notions under Differential Privacycited by 9
  6. 2022SF-PATE: Scalable, Fair, and Private Aggregation of Teacher Ensemblescited by 8
  7. 2021DP-SGD vs PATE: Which Has Less Disparate Impact on GANs?cited by 7
  8. 2023SoK: Unintended Interactions among Machine Learning Defenses and Riskscited by 6
  9. 2023On the Fairness Impacts of Private Ensembles Modelscited by 4
  10. 2023SoK: Taming the Triangle -- On the Interplays between Fairness, Interpretability and Privacy in Machine Learningcited by 4
  11. 2025PRISM: Diversifying Dataset Distillation by Decoupling Architectural Priorscited by 0