Learning Linear Dynamical Systems: Improved Rates and the Role of Regularization
Maryam Fazel, (University of Washington, Seattle)
We consider the problem of learning linear dynamical systems from input-output data, or system identification, given limited output samples. Learning the system dynamics is often the basis of associated control or policy decision problems in tasks varying from linear-quadratic control to deep reinforcement learning. Recent literature has provided finite-sample statistical analysis for simple least-squares regression models applied to this problem. When a low-order model is desired, adding a Hankel nuclear norm regularization term to the least squares problem has been helpful in practice (e.g., simplifies model selection, is less sensitive to hyperparameter tuning). In this talk, we discuss new theoretical analysis and insights for the regularized scheme.
Maryam Fazel is the Moorthy Family Professor of Electrical and Computer Engineering at the University of Washington, with adjunct appointments in Computer Science and Engineering, Mathematics, and Statistics. Maryam received her MS and PhD from Stanford University, her BS from Sharif University of Technology in Iran, and was a postdoctoral scholar at Caltech before joining UW. She is a recipient of the NSF Career Award, UWEE Outstanding Teaching Award, UAI conference Best Student Paper Award (with her student), and coauthored a paper selected as a Fast-Breaking paper by Science Watch (2011). She serves as the Associate Chair for Research for the UW ECE department, and directs the Institute for Foundations of Data Science (IFDS), a multi-site, collaborative NSF TRIPODS Institute. She is an associate editor of the SIAM journal on Optimization and the SIAM journal on Mathematics of Data Science. Her current research interests are in the area of optimization in machine learning and control.