On Dynamic Noise Influence in Differential Private Learning

Junyuan Hong, Zhangyang Wang, Jiayu Zhou

Keywords: privacy, private learning, dynamic policy

Abstract: Protecting privacy in learning while maintaining the model performance has become increasingly critical in many applications that involve sensitive data. Private Gradient Descent (PGD) is a commonly used private learning framework, which adds noise according to the Differential Privacy protocol.Recent studies show that dynamic privacy schedules of decreasing noise magnitudes can improve loss at the final iteration, and yet theoretical understandings of the effectiveness of such schedules and their connections to optimization algorithms remain limited. In this paper, we provide comprehensive analysis of noise influence in dynamic privacy schedules to answer these critical questions. We first present a dynamic noise schedule minimizing the utility upper bound of PGD, and show how the noise influence from each optimization step collectively impacts utility of the final model. Our study also reveals how impacts from dynamic noise influence change when momentum is used. We empirically show the connection exists for general non-convex losses, and the influence is greatly impacted by the loss curvature.

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

One-sentence Summary: Improve utility upper bound for differential private learning by dynamic noise influence

Reviewed Version (pdf): https://openreview.net/references/pdf?id=WpIRxj10MP