arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2109.09142cs.DC
收藏

具有差分隐私的分散式无线联邦学习

Decentralized Wireless Federated Learning with Differential Privacy

Shuzhen Chen, Dongxiao Yu, Yifei Zou, Jiguo Yu, Xiuzhen Cheng

摘要:

本文研究了无线物联网中的分散式联邦学习算法。传统的联邦学习参数服务器架构存在容错能力低、通信开销大、私有数据不可访问等问题。为了解决这些问题,我们提出了一种分布式无线联合学习算法DWFL。该算法工作在一个系统中,工人组织在一个对等和无服务器的方式,和工人交换他们的隐私保护数据的模拟传输方案通过无线信道并行。通过严格的分析,我们表明,DWFL满足$(ε,δ)$ -差分隐私和隐私预算每个工人规模为$\mathcal{O}(\frac{1}{\sqrt{N}})$,在正交传输方法中的恒定预算相比。此外,DWFL的收敛速度与具有中央参数服务器的最知名的集中式算法的$\mathcal{O}(\sqrt{\frac{1}{TN}})$相同。大量的实验表明,我们的算法DWFL也表现良好,在真实的设置。

英文摘要:

This paper studies decentralized federated learning algorithms in wireless IoT networks. The traditional parameter server architecture for federated learning faces some problems such as low fault tolerance, large communication overhead and inaccessibility of private data. To solve these problems, we propose a Decentralized-Wireless-Federated-Learning algorithm called DWFL. The algorithm works in a system where the workers are organized in a peer-to-peer and server-less manner, and the workers exchange their privacy preserving data with the analog transmission scheme over wireless channels in parallel. With rigorous analysis, we show that DWFL satisfies $(ε,δ)$-differential privacy and the privacy budget per worker scales as $\mathcal{O}(\frac{1}{\sqrt{N}})$, in contrast with the constant budget in the orthogonal transmission approach. Furthermore, DWFL converges at the same rate of $\mathcal{O}(\sqrt{\frac{1}{TN}})$ as the best known centralized algorithm with a central parameter server. Extensive experiments demonstrate that our algorithm DWFL also performs well in real settings.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑