Loading Events

Control Seminar

Accelerating Control Algorithms with Randomized Linear Algebra

James AndersonAssociate ProfessorColumbia University
WHERE:
1311 EECS BuildingMap
SHARE:

Abstract: Learning a dynamical system model from input-output data is a fundamental component of the control synthesis pipeline. Recently there has been a large effort focusing on how to characterize the sample complexity of learning algorithms applied to various control problems. These results unfortunately do not relate to the computational overhead required to solve the problem. As we will show, even for moderately sized systems, numerical routines which we take for granted, such as the singular value decomposition (SVD) are off limits. In this talk, I will demonstrate how an approximate SVD computed via randomization can produce system models with accuracy that is close to the deterministic optimal solution, but with a dramatic speedup in computation time. In particular, I will describe two problem instances where randomized algorithms for numerical linear algebra can accelerate classical system identification algorithms. I will then describe our recent work on how the same techniques can be adapted to provide approximate but efficient projections onto the positive-semidefinite cone, and how this can be embedded into existing SDP solvers to improve performance.

Bio: James Anderson is an associate professor in the Department of Electrical Engineering at Columbia University, where he is also a member of the Data Science Institute. From 2016 to 2019 he was a senior postdoctoral scholar in the Department of Computing+Mathematical Sciences at the California Institute of Technology. Prior to Caltech, he held a Junior Research Fellowship at St John’s College at the University Oxford and was also affiliated with the Department of Engineering Science. He received his DPhil (PhD) from Oxford in 2012 and the BSc and MSc degrees from the University of Reading in 2005 and 2006 respectively. His research interests include foundations of control, optimization, and learning with applications to cyber-physical systems.

*** This Event will take place in a hybrid format. The location for in-person attendance will be room 1311 EECS. Attendance will also be available via Zoom.

Join Zoom Meeting: https://umich.zoom.us/j/96731875637

Meeting ID: 967 3187 5637

Passcode: XXXXXX (Will be sent via e-mail to attendees)

Zoom Passcode information is also available upon request to Kristi Rieger([email protected])

Faculty Host

Lisa LiAssistant ProfessorElectrical Engineering and Computer Science