DCL Lecture Series: Andrew Lamperski - Some New Results of Modeling and Optimization from Streaming Data

Decision and Control Laboratory Lecture Series

Coordinated Science Laboratory

 

Some New Results of Modeling and Optimization from Streaming Data

Dr. Andrew Lamperski

Assistant Professor, Department of Electrical and Computer Engineering

University of Minnesota

 

Wednesday, November 13, 2019

3:00pm – 4:00pm

CSL Auditorium (B02)

__________________________________________________________________________________________________________________________________________________

Abstract:

This talk will describe results on system identification, adaptive least-squares, and online learning from streaming data.

First we will describe results on non-asymptotic system identification of closed-loop systems. Recently, there have been several works that have provided non-asymptotic bounds on identification error. However, these bounds are restricted to open-loop identification, and cannot be used in applications with feedback, such as network identification. We describe new non-asymptotic prediction error bounds for a closed-loop subspace identification method.

Next we will discuss results on recursive least-squares and related problems. Recursive least-squares algorithms use forgetting factors to adapt to non-stationary data.   We show how to rigorously achieve strong adaptive performance, under a metric known as dynamic regret, for a class of forgetting factor algorithms which includes recursive least-squares.

Finally, we will briefly describe applications of online learning to the tuning of neural stimulators. We will describe results on restoring lower-body function from a clinical trial on spinal cord injury patients. Theoretical foundations from online learning will be sketched.

Bio: Andrew Lamperski

received the B.S. degree in biomedical engineering and mathematics in 2004 from the Johns Hopkins University, Baltimore, MD, and the Ph.D. degree in control and dynamical systems in 2011 from the California Institute of Technology, Pasadena. He held postdoctoral positions in control and dynamical systems at the California Institute of Technology from 2011 - 2012 and in mechanical engineering at The Johns Hopkins University in 2012. From 2012 - 2014, did postdoctoral work in the Department of Engineering, University of Cambridge, on a scholarship from the Whitaker International Program. In 2014, he joined the Department of Electrical and Computer Engineering, University of Minnesota as an Assistant Professor. His research interests include optimal control and machine learning, with applications to neuroscience and robotics.