面向边缘智能的空中计算

 

 

曹晓雯^1,2,莫小鹏^1,2, 许杰¹

摘要：边缘智能模型训练中的无线通信开销已成为系统性能瓶颈，空中计算是解决该问题的重要技术。利用无线多址接入信道的信号叠加特性，空中计算技术能够在多终端无线信号传输的同时，对参数汇总计算，从而实现通信计算一体化设计，降低无线通信开销，提高边缘智能系统性能。通过实例介绍了空中计算的基本原理及其在边缘智能中的应用，并展望了未来研究方向。


Abstract: The wireless communication overheads in the training of edge intelligence models have become the bottleneck of system performance. Over-the-air computation has emerged as a promising solution to address this issue. By exploiting the signal superposition property of wireless multiple access channels, over-the-air computation implements the aggregation of model parameters in a swift manner, during the concurrent transmission of multiple terminal devices. Via such an integrated communication and computation design, this technique can significantly reduce the wireless communication overhead and improve the AI training performance. Through case analysis, the basic principles of over-the-air computation and its application in edge intelligence are first introduced, and then the future research directions are presented. 

Keywords: edge intelligence; over-the-air computation; distributed machine learning; multiple access channel