The main focus of our research is to enable learning, analysis, and predictive control of complex chemical and biomolecular systems via developing suitable computational methods and application-relevant control theory. Our group is particularly interested in studying systems with probabilistic uncertainty (i.e., stochastic systems). Our research leverages the interplay between theory, computation, and application.
Machine Learning and Predictive Control for Non-Equilibrium Plasma Treatment of Complex Surfaces
Non-equilibrium plasmas for surface functionalization of bio-materials and plasma medicine.
Lack of mechanistic insights into interactions of non-equilibrium plasmas (NEPs) with complex surfaces (e.g., in plasma medicine, plasma catalysis, or plasma bioprocessing), poses a major challenge in reliable and effective NEP processing. For example, therapeutic applications of NEPs in plasma medicine demand selective and high-efficacy treatment of complex biological substrates, while ensuring safe and reproducible treatment. This research investigates the application of machine learning and predictive control for characterizing and controlling the underpinning mechanisms of plasma-interfacial reactions with complex surfaces. The research is conducted in collaboration with the Graves Lab.
Metamodeling and Fast Uncertainty Quantification of Complex Chemical and Biological Systems
Accurate and fast quantification of model uncertainties is crucial when using models of complex chemical and biomolecular systems for analysis and decision-support tasks.
The prediction of the behavior of chemical and biological systems can be subject to various sources of uncertainty including unknown model parameters, unknown model structure, and experimental uncertainty such as measurement error. Accurate quantification of these uncertainties, as well as their impact on the quality of model predictions, is vital when utilizing these models for systems analysis or for decision-support and optimization tasks such as parameter estimation and optimal experiment design. This research focuses on the two major problems in uncertainty quantification: (1) Forward uncertainty propagation for optimal Bayesian experiment design and hypothesis testing, and (2) Inverse uncertainty quantification using Bayesian inference and optimization-based estimation methods. This research also investigates the use of metamodeling techniques to enable fast uncertainty quantification of complex systems.
Learning-based Control of Colloidal Self-Assembly Systems
Exquisite control of self-assembly systems enables manufacturing highly-ordered materials with unique properties.
Self-assembly (SA) is the process by which discrete components (e.g., nanoparticles in solution) spontaneously organize into an ordered structure. The spontaneous self-organization central to SA enables "bottom-up" materials synthesis, which allows for creating highly ordered, three-dimensional crystalline structures with unique optical, electrical, or mechanical properties. However, SA is an inherently stochastic process prone to kinetic arrest. This means that during assembly structures often become “stuck” in kinetically favorable local minima, and desired, high-valued structures corresponding to global free energy minima may not be created on a time-scale appropriate for cost-effective manufacturing. Leveraging recent advances in learning-based control and reinforcement learning, the overarching objective of this research is to develop a computational framework for optimal feedback control of SA systems that can systematically modulate the systems' inherently stochastic and nonlinear dynamics for reproducible manufacturing of advanced materials.