Towards Intelligent Mobile Edging: High Efficiency, Low Cost, and Good Decision



Journal Title

Journal ISSN

Volume Title



In recent years, with the continuous prosperity of the mobile Internet industry and the Internet of Things, the advance of widely used mobile terminals promotes the integration of mobile networks and Artificial Intelligence (AI), two of the most disruptive technologies world has seen nowadays. While each technology has spawned a large number of applications to facilitate our lives, the combination of mobile networks and AI, i.e., intelligent mobile edging, is going to be genuinely transformative. One perspective of combining these two is to exploit mobile networks to better support AI (Wireless for AI), where federated learning (FL) over mobile devices can greatly extend the scale of AI functions and preserve data privacy. Another perspective of the combination of AI and mobile networks is to tackle the challenges of wireless communication with AI strengths (AI for Wireless). Specifically, AI techniques can represent hard-to-model wireless problems and find feasible solutions with low computational complexity. The last point is that the combination can enable various smart mobile applications and services (AI & Wireless for applications). Though intelligent mobile edging has infiltrated many areas due to its advantages, several critical challenges still limit the efficient implementation of intelligent mobile edging, among which efficiency, cost, and performance are major considerations of this dissertation.

Huge energy/time consumption is one of the most significant obstacles restricting the development of AI functions on resource-constrained mobile devices. Besides, the delay-sensitive property of intelligent mobile applications also puts forward higher requirements for the efficiency of AI methodologies and communication service decision-making. Therefore, given these challenges, the objectives of this dissertation are to develop high efficiency, low cost, and good decision intelligent mobile edging methodologies from the three perspectives mentioned above through a combination of theoretical, simulation, and experimental studies. Specifically, this dissertation firstly endeavors to develop a series of efficient FL over mobile devices approaches, where computing and communication resources are well balanced to reduce the total cost during training; and then focuses on making good decisions and improving the performance of implementing AI functions on wireless networks and intelligent wireless services.



Federated learning, Reinforcement learning, Wireless Communications


Portions of this document appear in: Shi, Dian, Liang Li, Tomoaki Ohtsuki, Miao Pan, Zhu Han, and H. Vincent Poor. "Make smart decisions faster: Deciding D2D resource allocation via Stackelberg game guided multi-agent deep reinforcement learning." IEEE Transactions on Mobile Computing 21, no. 12 (2021): 4426-4438; and in: Shi, Dian, Jiahao Ding, Sai Mounika Errapotu, Hao Yue, Wenjun Xu, Xiangwei Zhou, and Miao Pan. "Deep q-network based route scheduling for transportation network company vehicles." In 2018 IEEE Global Communications Conference (GLOBECOM), pp. 1-7. IEEE, 2018; and in: Shi, Dian, Jixiang Lu, Jie Wang, Lixin Li, Kaikai Liu, and Miao Pan. "No One Left Behind: Avoid Hot Car Deaths via WiFi Detection." In ICC 2020-2020 IEEE International Conference on Communications (ICC), pp. 1-6. IEEE, 2020.