面向高效通信边缘学习网络的通信计算一体化设计

 

 

朱光旭,李航

摘要：面向边缘学习网络，探讨了一种新型的基于空中计算的模型聚合方案，并对其中的关键使能技术展开论述。该方案利用无线多址信道的波形叠加特性将通信与计算在空中无缝融合，能够突破现有的通信-计算分离设计框架的局限性，从而大大提高频谱利用率并降低通信延时，并缓解了制约联邦式边缘学习大规模扩展的通信时延问题。
关键词：边缘智能；联邦式边缘学习；计算；多址接入


Abstract: A new model aggregation scheme based on over-the-air computing is discussed, and the key enabling technologies are discussed. The proposed solution can achieve the desired model aggregation over the air via exploiting the wave-superposition property of multi-access channels, seamlessly integrating communication and computation. Therefore, it can break through the limitations of the classic design principle of decoupling communication and computation, greatly improve the spectrum efficiency and reduce the communication delay, and alleviate the bottleneck of communication delay which restricts the large-scale expansion of federated edge learning. 

Keywords: edge intelligence; federated edge learning; computation; multiple access