DP-SGD vs PATE: Which Has Less Disparate Impact on Model Accuracy?
This work addresses fairness issues in privacy-preserving machine learning for researchers and practitioners, though it is incremental as it compares existing methods rather than introducing new ones.
The paper compared the fairness impacts of two differentially private deep learning methods, DP-SGD and PATE, finding that both have disparate effects on model accuracy across sub-groups, but PATE's impact is less severe than DP-SGD's, with specific reductions in accuracy drops for under-represented groups.
Recent advances in differentially private deep learning have demonstrated that application of differential privacy, specifically the DP-SGD algorithm, has a disparate impact on different sub-groups in the population, which leads to a significantly high drop-in model utility for sub-populations that are under-represented (minorities), compared to well-represented ones. In this work, we aim to compare PATE, another mechanism for training deep learning models using differential privacy, with DP-SGD in terms of fairness. We show that PATE does have a disparate impact too, however, it is much less severe than DP-SGD. We draw insights from this observation on what might be promising directions in achieving better fairness-privacy trade-offs.