Postdoc Mingzhe Chen from SRIBD receives 2020 IEEE ICC Best Paper Award

Created: Wednesday, 27 May 2020

Postdoc Mingzhe Chen from SRIBD receives 2020 IEEE ICC Best Paper Award

Mingzhe Chen from the joint-post-doctoral fellowship program of Shenzhen Research Institute of Big Data and Princeton University has won the Best Paper Award of 2020 IEEE International Conference on Communications (ICC). H. Vincent Poor, Walid Saad and Shuguang Cui are the co-authors of the paper titled “Convergence Time Minimization of Federated Learning over Wireless Networks”.

 

The IEEE International Conference on Communications (ICC) is one of the IEEE Communications Society’s two important conferences held in every midyear. ICC is dedicated to driving innovation in nearly every aspect of communications. Each year, around 3,000 researchers submit their proposals for paper presentations and program sessions to be held. After extensive peer review, the best of the proposals are selected for the conference program. The Best Paper award for the workshop will be decided by all Executive Committee/Advisory Board Members.

 

The IEEE International Conference on Communications

 

This paper focused on the minimization of convergence time of federated learning (FL) over wireless networks. Due to the limited resource blocks (RBs) in a wireless network, only a subset of users can be selected and transmit their local FL model parameters to the BS at each learning step. To overcome this problem, a probabilistic user selection scheme is proposed using which the BS will connect to the users, whose local FL models have large effects on its global FL model, with high probabilities. Furthermore, artificial neural networks (ANNs) are used to estimate the local FL models of the users that are not allocated any RBs for local FL model transmission to further reduce the FL convergence time.

 

Two major issues came out when Mingzhe Chen and co-authors were working on this project. The first one was how to train ANNs to estimate the local FL models of users without increasing the users’ energy consumption. A probabilistic user selection scheme is found to solve this problem and it also enables the FL algorithm reaches the optimal convergence value. In addition, the proposed probabilistic user selection scheme can provide the training data samples for training ANNs. Secondly, the team came across with the problem of how to define the output. At first, the team directly used the future local FL models of a user as the output of an ANN, which results in a bad prediction accuracy. Finally, they decided to use the difference between the local FL models of any two devices. The paper stands out among the papers submitted to IEEE ICC and receives the Best Paper Award.

 

Mingzhe Chen、H. Vincent Poor、 Walid Saad and Shuguang Cui

 

Mingzhe Chen received the Ph. D. degree from the Beijing University of Posts and Telecommunications, Beijing, China, in 2019. He is currently a postdoctoral fellow at the Shenzhen Research Institute of Big Data and also at the Electrical Engineering Department of Princeton University. He was an exemplary reviewer for IEEE Transactions on Wireless Communications in 2018 and IEEE Transactions on Communications in 2018 and 2019. He served as a Co-Chair for 2020 ICC Workshop on Edge Machine Learning for 5G Mobile Networks and Beyond. He will serve as a Co-Chair for 2020 GLOBECOM Workshop on Edge Learning over 5G Networks and Beyond and serve as a Guest Editor for the IEEE Journal on Selected Areas in Communications (JSAC) Special Issue on Distributed Learning over Wireless Edge Networks. His research interests include machine learning, virtual reality, unmanned aerial vehicles, game theory, wireless networks, and caching.