James Anderson Receives NSF CAREER Award
James Anderson, assistant professor of electrical engineering, has received a National Science Foundation (NSF) CAREER award for his work on developing randomized algorithms for the analysis and control of large-scale cyber-physical systems, such as connected autonomous cars, smart power grids, and even the internet.
“My group is focused on a fundamental problem in scientific computing: how to design scalable algorithms capable of producing optimal solutions in as close to real-time as possible,” said Anderson, who works at the interface of engineering and computational mathematics.“Our goal is to use feedback control to provide robust, reliable, and secure autonomy to large-scale networked systems, from swarms of drones flying in formation to smart city infrastructure.”
Modeling and designing controllers for large-scale systems is technically and computationally challenging at best and can, in the worse case, be impossible. Distributed control attempts to alleviate the computational problem by moving from one large centralized decision maker to multiple smaller decision makers spread throughout the network. From a theoretical perspective, distributed control is unfortunately a considerably more difficult problem. Despite recent progress in distributed control, there is a large gap between the complexity and magnitude of the systems that engineers can control and the complexity of systems that they want to control.
Speed and accuracy are often competing goals. While researchers can increase accuracy by supplying more data, they frequently find that they have more data than they can possibly compute with in any reasonable amount of time. “Luckily for us, real data is often noisy and/or incomplete, and we exploit this,” noted Anderson, who is also a member of the Data Science Institute. “Randomized algorithms enable us to navigate the trade-off between accuracy and computational tractability without having to use all the data.”
The five-year, $500,000 NSF grant will support Anderson’s work of creating randomized algorithms and embedding them into workflows for modeling, analysis, control, and optimization of large networked-systems. For years, researchers have been focusing on high-precision solutions and as a result have been beaten by the curse of dimensionality. Taking a different approach, Anderson will embrace the messiness of real data and will formulate “approximate” solutions using randomization, at a fraction of the cost of high-precision solutions. Correctly designed, these approximate solutions provide comparable levels of performance and robustness, while allowing us to work with systems of an unprecedented size.
“Our work will bridge the gap between theory, scalable computation, and data privacy and thus help make autonomous systems that we depend upon safe and reliable,” Anderson added. “We expect our results will enable the widespread adoption of distributed control in the aerospace, robotics, automotive, and energy industries.”