Best Effort Voting Power Control for Byzantine-resilient Federated Learning Over the Air
Best Effort Voting Power Control for Byzantine-resilient Federated Learning Over the Air
复制标题
DOI：
10.1109/iccworkshops53468.2022.9814495
复制
发表时间：
2022-05
期刊：
影响因子：
--
通讯作者：
Xin Fan;Yue Wang;Yan Huo;Zhi Tian
中科院分区：
文献类型：
--
作者：
Xin Fan;Yue Wang;Yan Huo;Zhi Tian
基于模拟聚合的空中联合学习（FLOA）在边缘计算范例中提供了高通信效率和隐私提供。当所有边缘设备（工作者）通过共同共享的时频资源同时将其本地更新上传到参数服务器（PS）时，PS只能获得平均更新而不是单个本地更新。因此，这种并发传输和聚合方案减少了通信的延迟和成本，但使FLOA容易受到拜占庭攻击。针对拜占庭弹性FLOA的设计，本文从分析现有FLOA文献中广泛采用的信道反转（CI）功率控制机制入手。我们的理论分析表明，虽然CI可以实现良好的学习性能，在非攻击的情况下，它不能很好地工作，有限的防御能力，拜占庭攻击。然后，我们提出了一种新的方案，称为最佳努力投票（BEV）的功率控制策略，结合随机梯度下降（SGD）。我们提出的BEV-SGD提高了FLOA对拜占庭攻击的鲁棒性，允许所有工作者以最大发射功率发送本地更新。在最强攻击情形下，分别给出了基于CI和基于BEV的FLOA算法的期望收敛速度。比较表明，我们的BEV优于其对应的CI在更好的收敛行为，这是由实验模拟验证。
Analog aggregation based federated learning over the air (FLOA) provides high communication efficiency and privacy provisioning in edge computing paradigm. When all edge devices (workers) simultaneously upload their local updates to the parameter server (PS) through the commonly shared time-frequency resources, the PS can only obtains the averaged update rather than the individual local ones. As a result, such a concurrent transmission and aggregation scheme reduces the latency and costs of communication but makes FLOA vulnerable to Byzantine attacks. For the design of Byzantine-resilient FLOA, this paper starts from analyzing the channel inversion (CI) power control mechanism that is widely used in existing FLOA literature. Our theoretical analysis indicates that although CI can achieve good learning performance in the non-attacking scenarios, it fails to work well with limited defensive capability to Byzantine attacks. Then, we propose a novel scheme called the best effort voting (BEV) power control policy, integrated with stochastic gradient descent (SGD). Our proposed BEV-SGD improves the robustness of FLOA to Byzantine attacks, by allowing all the workers to send their local updates at their maximum transmit power. Under the strongest-attacking circumstance, we derive the expected convergence rates of FLOA with CI and BEV, respectively. The comparison reveals that our BEV outperforms its counterpart with CI in terms of better convergence behavior, which is verified by experimental simulations.