Active Machine Learning to Guide Discovery of Non-Equilibrium Plasma Interactions with Complex Interfaces

Active learning (AL) is the branch of machine learning concerned with systematically querying samples from an experimental system (or a computational model) to train a data-driven model that maps (experimental) design parameters to process performance measures. AL has emerged as a useful tool for guiding high-throughput experiments and expensive computations in a variety of science and engineering fields. In this talk, we will discuss the promise of constrained and multi-objective Bayesian optimization methods for AL-guided exploration of the multivariable and highly nonlinear parameter space of non-equilibrium plasmas (NEPs) in a systematic and resource-efficient manner. We will demonstrate how AL approaches can pave the way for automated and “optimal” exploration of the parameter space of NEPs, towards establishing insights into the complex behavior of the plasma when interacting with interfaces.