Optimization in Complex Systems: Delayed Feedback and Scarce Communication Resources
Modern engineering technologies (such as federated learning) often require solving various optimization problems in complex systems, where the environments are time-varying and the decision makers have limited communication resources, e.g., energy and bandwidth. The optimization algorithms should be able to adapt to fast dynamics, work with imperfect information, save resources, and preserve the data privacy of users. In this seminar, I will present some of my recent works on online/decentralized optimization addressing these challenges in complex systems. First, I will study the impact of delayed feedback information on the performance of constrained online convex optimization algorithms. Performance upper bounds are established in terms of the feedback delays, and conditions for sublinear regret and constraint violations are given. Further, I will reduce the communication overhead of decentralized online optimization by taking an event-triggering approach, in which neighboring agents communicate infrequently. A tradeoff between communication cost and optimization performance can be observed. Additionally, when the communication bandwidth is scarce, I will develop a quantization-based algorithm to solve multi-agent optimization problems with nonlinear pairwise constraints. The effect of quantization accuracy on optimization performance is demonstrated. Finally, several potential directions for future research will be discussed.
Xuanyu Cao received the B.E. degree in electrical engineering from Shanghai Jiao Tong University, in 2013, and the M.S. and Ph.D. degrees in electrical engineering from the University of Maryland, College Park, in 2016 and 2017, respectively. From August 2017 to May 2019, he was a Postdoctoral Research Associate with the Department of Electrical Engineering, Princeton University. Since June 2019, he has been a Postdoctoral Research Associate with the Coordinated Science Lab, University of Illinois at Urbana-Champaign. His research lies in the optimization theory and algorithms, game theory and network economics, statistical/adaptive signal processing, and their applications in real-world systems and networks such as communication networks, power grids, data centers, and social networks. He is a Senior Member of IEEE and an Editor for the IEEE Transactions on Vehicular Technology.