Differentially-Private Distributed Optimization with Guaranteed Optimality
Differentially-Private Distributed Optimization with Guaranteed Optimality
复制标题
DOI：
10.1109/cdc49753.2023.10383285
复制
发表时间：
2023-12
期刊：
影响因子：
--
通讯作者：
Yongqiang Wang;A. Nedić
中科院分区：
文献类型：
--
作者：
Yongqiang Wang;A. Nedić
隐私保护在分布式优化和学习中越来越受到关注。由于差分隐私正在成为一个事实上的隐私保护标准，最近出现的结果集成差分隐私与分布式优化。然而，为了确保差分隐私（有限的累积隐私预算），所有现有的方法都必须牺牲可证明的收敛到最优解。在本文中，我们提出了一个差分隐私分布式优化算法，可以确保，第一次，$\N $差分隐私和最优性，即使在无限的时间范围内。数值仿真结果验证了该方法的有效性。
Privacy protection is gaining increased attention in distributed optimization and learning. As differential privacy is becoming a de facto standard for privacy preservation, recently results have emerged integrating differential privacy with distributed optimization. However, to ensure differential privacy (with a finite cumulative privacy budget), all existing approaches have to sacrifice provable convergence to the optimal solution. In this paper, we propose a differentially-private distributed optimization algorithm that can ensure, for the first time, both $\epsilon$ -differential privacy and optimality, even on the infinite time horizon. Numerical simulation results confirm the effectiveness of the proposed approach.