paper-with-me

홈 › Papers

Mitigating Disparate Impact of Differentially Private Learning through Bounded Adaptive Clipping

2025-06-02 · Linzh Zhao, Aki Rehn, Mikko A. Heikkilä, Razane Tajeddine, Antti Honkela

Differential privacy (DP) has become an essential framework for privacy-preserving machine learning. Existing DP learning methods, however, often have disparate impacts on model predictions, e.g., for minority groups. Gradient clipping, which is often used in DP learning, can suppress larger gradients from challenging samples. We show that this problem is amplified by adaptive clipping, which will often shrink the clipping bound to tiny values to match a well-fitting majority, while significantly reducing the accuracy for others. We propose bounded adaptive clipping, which introduces a tunable lower bound to prevent excessive gradient suppression. Our method improves the accuracy of the worst-performing class on average over 10 percentage points on skewed MNIST and Fashion MNIST compared to the unbounded adaptive clipping, and over 5 percentage points over constant clipping.

📄 PDF Abstract BibTeX arXiv:2506.01396

Code (0)

등록된 구현이 없습니다.

Tasks

Privacy Preserving

Similar Papers 제목 키워드 기반

Private and Fair Machine Learning: Revisiting the Disparate Impact of Differentially Private SGD

2025-10-02 · Lea Demelius, Dominik Kowald, Simone Kopeinik, Roman Kern 외 arxiv

Differential privacy (DP) is a prominent method for protecting information about individuals during data analysis. Training neural networks with differentially private stochastic gradient descent (DPSGD) influences the m…

Removing Disparate Impact of Differentially Private Stochastic Gradient Descent on Model Accuracy

2020-03-08 · Depeng Xu, Wei Du, Xintao Wu

When we enforce differential privacy in machine learning, the utility-privacy trade-off is different w.r.t. each group. Gradient clipping and random noise addition disproportionately affect underrepresented and complex c…

Post-processing of Differentially Private Data: A Fairness Perspective

2022-01-24 · Keyu Zhu, Ferdinando Fioretto, Pascal Van Hentenryck

Post-processing immunity is a fundamental property of differential privacy: it enables arbitrary data-independent transformations to differentially private outputs without affecting their privacy guarantees. Post-process…

Fairness

Neither Private Nor Fair: Impact of Data Imbalance on Utility and Fairness in Differential Privacy

2020-09-10 · Tom Farrand, FatemehSadat Mireshghallah, Sahib Singh, Andrew Trask

Deployment of deep learning in different fields and industries is growing day by day due to its performance, which relies on the availability of data and compute. Data is often crowd-sourced and contains sensitive inform…

Fairness

Disparate Impact in Differential Privacy from Gradient Misalignment

2022-06-15 · Maria S. Esipova, Atiyeh Ashari Ghomi, Yaqiao Luo, Jesse C. Cresswell

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 …

Fairness