DP-RBAdaBound: A differentially private randomized block-coordinate adaptive gradient algorithm for training deep neural networks

作者：

Highlights：

• Our algorithm solves the problems of training deep learning models.

• Our algorithm achieves the regret bound of O(T) for convex loss functions.

• Our algorithm preserves ϵ-differential privacy.

摘要

•Our algorithm solves the problems of training deep learning models.•Our algorithm achieves the regret bound of O(T) for convex loss functions.•Our algorithm preserves ϵ-differential privacy.

论文关键词：Adaptive gradient methods,Deep learning models,Differential privacy,Randomized block-coordinate

论文评审过程：Received 25 June 2021, Revised 10 August 2022, Accepted 13 August 2022, Available online 20 August 2022, Version of Record 29 August 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.118574