Federated Learning with Bayesian Differential Privacy

Aleksei Triastcyn, Boi Faltings

2019 · 199 citations · 36 references

DOIFull text

Open access

Concepts

TL;DR

The study aims to strengthen federated learning with formal privacy guarantees by applying Bayesian differential privacy to achieve sharper privacy loss bounds. The authors adapt Bayesian privacy accounting to the federated setting and propose multiple improvements for efficient privacy budgeting at client and instance levels. Experiments on image classification, including a medical task, show that the approach attains tighter privacy budgets (ε < 1 per client, ε < 0.1 per instance), improves model accuracy, and reduces communication rounds compared to state‑of‑the‑art differential privacy methods.

Abstract

We consider the problem of reinforcing federated learning with formal privacy guarantees. We propose to employ Bayesian differential privacy, a relaxation of differential privacy for similarly distributed data, to provide sharper privacy loss bounds. We adapt the Bayesian privacy accounting method to the federated setting and suggest multiple improvements for more efficient privacy budgeting at different levels. Our experiments show significant advantage over the state-of-the-art differential privacy bounds for federated learning on image classification tasks, including a medical application, bringing the privacy budget below ε = 1 at the client level, and below ε = 0.1 at the instance level. Lower amounts of noise also benefit the model accuracy and reduce the number of communication rounds.

References

36