Top row (l to r): Seyyedali Hosseinalipour (Ali Alipour), assistant professor in Electrical Engineering at University at Buffalo, Sheikh Shams Azam, PhD student in ECE at Purdue, Christopher G. Brinton, Elmore Rising Star Assistant Professor of ECE at Purdue, and Nicolò Michelusi, associate professor in Electrical, Computer, and Energy Engineering at Arizona State University. Bottom row (l to r): Vaneet Aggarwal, professor of Industrial Engineering at Purdue, David J. Love, Nick Trbovich Professor of ECE at Purdue, and Huaiyu Dai, professor of ECE at North Carolina State University.

 

A team of researchers from Purdue University and other institutions has won the 2024 IEEE Communications Society William R. Bennett Prize. This award is an annually given recognition of an outstanding original paper published in IEEE/ACM Transactions on Networking or the IEEE Transactions on Network and Service Management over the past three calendar years.

The paper, titled “Multi-Stage Hybrid Federated Learning Over Large-Scale D2D-Enabled Fog Networks,” was published in IEEE/ACM Transactions on Networking in August 2022, and has since received close to 90 citations on Google scholar. It makes significant contributions to federated learning, where edge devices train machine learning models that are frequently synchronized over a network. Whereas federated learning has traditionally focused on a “star” setup where devices are directly connected to an aggregation server, this paper considered the more realistic case where a multi-stage hierarchy of computing nodes separates them from the server. Furthermore, it considers that the devices may exchange models over device-to-device (D2D) communications in 5G-and-beyond wireless. The paper develops mathematical theory and algorithms for optimizing how devices collaborate within and between layers of the hierarchy to learn efficiently.

The paper’s authors are:

• Seyyedali Hosseinalipour (Ali Alipour), formerly a postdoctoral researcher in the Elmore Family School of Electrical and Computer Engineering (ECE) at Purdue, advised by Christopher G. Brinton and Mung Chiang. Alipour is now an assistant professor in Electrical Engineering at University at Buffalo.
• Sheikh Shams Azam, a PhD student in ECE at Purdue, advised by Brinton. Azam is currently on leave from his studies as part of Apple’s AI/ML Residency Program.
• Christopher G. Brinton, the Elmore Rising Star Assistant Professor of ECE at Purdue.
• Nicolò Michelusi, associate professor in Electrical, Computer, and Energy Engineering at Arizona State University, and formerly assistant professor of ECE at Purdue.
• Vaneet Aggarwal, professor of Industrial Engineering at Purdue.
• David J. Love, the Nick Trbovich Professor of ECE at Purdue.
• Purdue-ece-researchers-part-of-team-aw professor of ECE at North Carolina State University, and Alipour’s PhD advisor.

The IEEE Communications Society solicits annual nominations for the William R. Bennett Prize from guest editors and readers of the publication. Mung Chiang, President and Roscoe H. George Distinguished Professor of ECE at Purdue, provided the nomination for this paper. The IEEE Communications Society Awards Committee deliberates the final selection among the nominations.

“On behalf of the authors, we are incredibly honored to receive the prestigious IEEE Communications Society William R. Bennett Prize,” Brinton said. “Our hope is that this work will continue to inspire research on expanding the frontier of intelligent model training and management over next generation network systems.”