Exploring the Unfairness of DP-SGD Across Settings
Not on the map. Why?
Abstract (arXiv)
End users and regulators require private and fair artificial intelligence models, but previous work suggests these objectives may be at odds. We use the CivilComments to evaluate the impact of applying the em de facto standard approach to privacy, DP-SGD, across several fairness metrics. We evaluate three implementations of DP-SGD: for dimensionality reduction (PCA), linear classification (logistic regression), and robust deep learning (Group-DRO). We establish a negative, logarithmic correlation between privacy and fairness in the case of linear classification and robust deep learning. DP-SGD had no significant impact on fairness for PCA, but upon inspection, also did not seem to lead to private representations.
Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Builds on 20 of its 34 references that are in the map. First posted on arXiv 24 Feb 2022.
Builds on
20 of 34 references.
References · 20
- 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
- 2019RoBERTa: A Robustly Optimized BERT Pretraining Approachcited by 13,266
- 2019DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lightercited by 4,243
- 2016Deep Learning with Differential Privacycited by 2,682
- 2020WILDS: A Benchmark of in-the-Wild Distribution Shiftscited by 1,366
- 2019Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalizationcited by 1,232
- 2017Renyi Differential Privacycited by 643
- 2019Nuanced Metrics for Measuring Unintended Bias with Real Data for Text Classificationcited by 396
- 2018Fairness Without Demographics in Repeated Loss Minimizationcited by 363
- 2020BERTweet: A pre-trained language model for English Tweetscited by 346
- 2019Differential Privacy Has Disparate Impact on Model Accuracycited by 316
- 2018Empirical Risk Minimization under Fairness Constraintscited by 226
8 more
- 2019Fairness risk measurescited by 85
- 2018A General Approach to Adding Differential Privacy to Iterative Training Procedurescited by 84
- 2020On the Privacy Risks of Algorithmic Fairnesscited by 82
- 2020Neither Private Nor Fair: Impact of Data Imbalance on Utility and Fairness in Differential Privacycited by 61
- 2019Diffprivlib: The IBM Differential Privacy Librarycited by 51
- 2021Decision Making with Differential Privacy under a Fairness Lenscited by 32
- 2021DP-SGD vs PATE: Which Has Less Disparate Impact on Model Accuracy?cited by 32
- 2021Assessing Fairness in Classification Parity of Machine Learning Models in Healthcarecited by 4
Built on by
Citing papers per year: 2022: 1; 2023: 1; 2024: 0; 2025: 1.
3 papers.