Introduction

Abstract

AbstractAlong with the wave of informatization technology, a booming era of artificial intelligence (AI) has emerged. In specific, with the proliferation of wireless communication, immense volumes of data are generated by mega-scale terminal devices instead of traditional cloud datacenters. According to the prediction by Ericsson (IoT connections outlook: NB-IoT and CAT-M technologies will account for close to 45 percent of cellular IoT connections in 2024. [Online]. Available: https://www.ericsson.com/en/mobilityreport/reports/june-2019/iot-connections-outlook ) and international data corporation (IDC) (Ericsson, Cisco Annual Internet Report (2018–2023). er[Online]. Available: https://www.cisco.com/c/en/us/solutions/collateral/executive-perspectives/annual-internet-report/white-paper-c11-741490.pdf ), internet of things (IoT) devices will generate 45% of the 40 zettabytes (ZB) global Internet data in 2024, while there will be 5.3 billion total Internet users and 29.3 billion networked devices by 2023. Nevertheless, global devices transferring extremely vast data to cloud datacenters will demand high bandwidth and powerful computational resources (Heintz et al. (Optimizing grouped aggregation in geo-distributed streaming analytics, in Proceedings of the 24th International Symposium on High-Performance Parallel and Distributed Computing, HPDC, ed. by T. Kielmann, D. Hildebrand, M. Taufer (2015), pp. 133–144)), thus creating a bottleneck on the restricted network transmission capabilities, computing power of computing infrastructures, strict delay requirements, etc. Edge intelligence (EI), as a complementary processing architecture by combining edge computing (EC) (Shi et al. (IEEE Int Things J 3(5):637–646, 2016)) and AI, pushes the AI frontier from the cloud to the network edge to open the path for low latency and critical-computation (Li et al. (Edge intelligence: On-demand deep learning model co-inference with device-edge synergy, in Proceedings of the 2018 Workshop on Mobile Edge Communications (MECOMM@SIGCOMM) (2018), pp. 31–36) ).