I. Introduction
With the increasing popularity of Internet of Things, massive smart devices are connected to the Internet, which generate unprecedented amounts of data [1] needed to be processed rapidly. Edge computing is proposed to provide computing services proximity to data sources, with advantages such as low latency and low communication overheads. In recent years, more and more artificial intelligence (AI) applications based on edge computing are developed to provide diverse and high-quality services at the network edge. The confluence of edge computing and AI gives birth to an emerging technology field namely edge intelligence. Recently, edge intelligence has attracted great attentions from both academia and industry [2]–[4]. Edge intelligence has also been widely incorporated into the 6G development blueprints of many research institutions and enterprises.