Enoch Yeung (UC Santa Barbara (UCSB), USA)

Title: Data-Driven State Criticality and Observability with Koopman
Operator Methods

Abstract: I will present two major results to show the use of data-
driven Koopman methods to identify critical states and observable
subspaces to solve problems in synthetic biology. In the first, I
present the use of dynamic mode decomposition (DMD) to model the
transcriptome-wide response of a root-isolate bacterium to a novel
chemical compound. By solving an observability maximization problem
for the DMD model, we find a panel of biomarkers that can act as
effective biosensors for the compound, even in field studies with the
bacterium. This study establishes a new precedent for reasoning about
state criticality and system observability, even without prior
knowledge of a network model. Second, I present new theoretical
results that show how Koopman methods can be used to evaluate
criticality of states to optimize performance of a nonlinear system.
Historically, this problem is solved using either direct sensitivity
analysis on a known model or by generating local function
distributions that span the nonlinear observable subspace of a system.
In the absence of a known model, I present a new Koopman-based method
for estimating the observable subspace of a nonlinear system purely
from data. Our results provide a route for data-driven discovery of
critical states that affect an output-based performance measure.

The seminar will take place in Room S08 at the Faculty of Sciences.