TITLE: Augmented Distributed Optimization: From Algorithms to Applications
ABSTRACT: In the past ten years, we have seen significant growth in the development of distributed algorithms to optimize a network utility (usually the sum of local cost functions) over a peer-to-peer network. These algorithms make use of local network resources, such as computing power and message-passing, to allow agents to determine an optimal common solution. Examples include distributed sensing, resource allocation and machine learning. In this talk, I will introduce some recently developed “augmented” distributed algorithms that converge at faster convergence rates than existing methods while still being able to reach the exact optimum. In fact, I will show that these proposed algorithms have comparable performance with their centralized counterparts even for directed and time-varying networks, which has never been achieved before. I will also talk about some of their potential applications in sensor network, smart grid and wind farm systems, respectively. Some possible future research topics will be also envisioned in the end of the talk.
BIO: Jinming Xu is currently working as a postdoctoral scholar at Purdue University. He was a research fellow of the EXQUITUS center in the School of Electrical and Electronic Engineering at Nanyang Technological University (NTU), Singapore, from 2016 to 2017, and in the Ira A. Fulton Schools of Engineering, Arizona State University, from 2017 to 2018, respectively. He received his BSc in mechanical engineering from Shandong University in 2009 and his PhD in electrical engineering from NTU in 2016. His current research interests are primarily in distributed optimization and control, and its applications in large-scale signal processing, machine learning and networked dynamical systems.