Disparate Impact in Differential Privacy from Gradient Misalignment

Maria S. Esipova, Atiyeh Ashari Ghomi, Yaqiao Luo, Jesse C Cresswell

Keywords: Differential privacy, fairness, privacy

TL;DR: DPSGD can have unfair outcomes on protected groups because of direction errors caused by per-sample gradient clipping, but unfairness can be dramatically reduced with a global clipping technique.

Abstract: 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.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Social Aspects of Machine Learning (eg, AI safety, fairness, privacy, interpretability, human-AI interaction, ethics)

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:2206.07737/code)