Koopman Operator Theory Based Machine Learning of Dynamical Systems, Igor Mezic

ISS Informal Systems Seminar
Koopman Operator Theory Based Machine Learning of Dynamical Systems
Igor Mezic – University of California, Santa Barbara, United States
Apr 21, 2023

Many approaches to machine learning have struggled with applications that possess complex process dynamics. In contrast, human intelligence is adapted, and - - arguably - built to deal with complex dynamics. The current theory holds that human brain achieves that by constantly rebuilding a model of the world based on the feedback it receives. I will describe an approach to machine learning of dynamical systems based on Koopman Operator Theory (KOT) that also produces generative, predictive, context-aware models. The approach is adaptable to (feedback) control applications. KOT has deep mathematical roots and I will discuss its basic tenets. I will also present computational methods that enable lean computation. A number of examples will be discussed, including use in fluid dynamics, power grid dynamics, network security, soft robotics, and game dynamics. Acknowledgement: Support from ARO, AFOSR, DARPA and ONR is gratefully acknowledged.