Most modern applications in artificial intelligence (AI) rely on a centralized data center to perform the training and inference over large datasets. Distributed (or federated) learning techniques allow the learning to be moved from large data centers to distributed edge entities, where users have more control of both their own data and computational resources. The local entities can collaborate to perform complex learning or inference tasks, but the frequent message exchanges required may cause a significant increase in the wireless traffic, which must be properly addressed. In this talk, I will discuss the impact of distributed learning over wireless networks and introduce two of our recent works in this direction. By focusing on cellular networks, we first examine the optimization of wireless resources for federated learning between local users and the base-station, and introduce a new knowledge caching framework to facilitate both training and access of machine learning models by these users. The model caching strategy must take into consideration the training efficiency, channel conditions, and the user preference. Then, for internet-of-things or wireless sensor networks, we propose a novel in-network learning framework, where low-cost sensor devices (each hosting only a small neural network) may collaborate to form a deep neural network over wireless multihop links. We utilize over-the-air computation to improve the communication efficiency, and network reconfiguration to adapt to varying sensor deployments and inference tasks. Through these works, we show that the new traffic type generated by distributed learning requires new wireless designs and considerations that are different from conventional voice and data traffic.

Y.-W. Peter Hong received his B.S. degree from National Taiwan University in 1999, and his Ph.D. degree from Cornell University in 2005, both in electrical engineering. He is a Professor of the Institute of Communications Engineering and Department of Electrical Engineering at National Tsing Hua University, Hsinchu, Taiwan, and Director of the LiteOn-NTHU Joint Research Center. His research interests include AI/ML in wireless communications, signal processing for sensor networks, UAV communications, distributed learning and optimization, and physical layer secrecy. Dr. Hong currently serves as Senior Area Editor of IEEE Transactions on Signal Processing. He is also a Distinguished Lecturer of IEEE Communications Society (2022-2023) and the Vice Director of the IEEE ComSoc Asia-Pacific Board (2022-2023). He received several awards for his research contributions, including the 2010 IEEE ComSoc Asia-Pacific Outstanding Young Researcher Award, 2011 Y. Z. Hsu Scientific Paper Award, 2011 National Science Council Wu Ta-You Memorial Award, 2012 CIEE Outstanding Young Electrical Engineer Award, 2018 Ministry of Science and Technology Outstanding Research Award, and 2022 CIEE Outstanding Electrical Engineering Professor Award.