FLORAS: Differentially Private Wireless Federated Learning Using Orthogonal Sequences
FLORAS: Differentially Private Wireless Federated Learning Using Orthogonal Sequences
复制标题
DOI：
10.1109/icc45041.2023.10278611
复制
发表时间：
2023-05
期刊：
影响因子：
--
通讯作者：
Xizixiang Wei;Tianhao Wang;Ruiquan Huang;Cong Shen;Jing Yang;H. Poor;Charles L. Brown
中科院分区：
文献类型：
--
作者：
Xizixiang Wei;Tianhao Wang;Ruiquan Huang;Cong Shen;Jing Yang;H. Poor;Charles L. Brown
我们为无线联邦学习（FL）系统提出了一种新颖的私有上行链路空中计算（AirComp）方法，称为 FLORAS。从通信设计的角度来看，FLORAS 通过利用正交序列的特性，消除了发射机处的信道状态信息 (CSIT) 的需求。从隐私角度来看，我们证明 FLORAS 可以提供纯差分隐私（DP）保证，并明确地将可实现的 $\epsilon$-DP 级别描述为 FLORAS 参数配置的函数。推导了一种新颖的 FL 收敛界限，它与纯 DP 保证相结合，允许在收敛速度和 DP 保证水平之间进行平滑权衡。基于真实世界数据集的实验不仅证实了理论发现，而且还实证证明了 FLORAS 相对于最先进的 AirComp 方法的通信和隐私优势。
We propose a novel private-preserving uplink over-the-air computation (AirComp) method, termed FLORAS, for wireless federated learning (FL) systems. From the communication design perspective, FLORAS eliminates the requirement of channel state information at the transmitters (CSIT) by leveraging the properties of orthogonal sequences. From the privacy perspective, we prove that FLORAS can offer pure differential privacy (DP) guarantee, and explicitly characterize the achievable $\epsilon$-DP level as a function of the FLORAS parameter configuration. A novel FL convergence bound is derived which, combined with the pure DP guarantee, allows for a smooth tradeoff between convergence rate and DP guarantee levels. Experiments based on real-world datasets not only corroborate the theoretical findings but also empirically demonstrate the communication and privacy advantages of FLORAS over state-of-the-art AirComp methods.