Joel Paulson Laboratory for Sustainability, Energy, Environment, Process Engineering
About
Education
- Ph.D., MIT, 2016
- M.S. CEP, MIT, 2013
- B.S., University of Texas at Austin, 2011
_____
Joel Paulson received his PhD in 2016 from the Massachusetts Institute of Technology (MIT), where he won an NSF Graduate Research Fellowship and multiple awards for research and outstanding teaching and mentoring.
His advisors were Professors Richard Braatz and Michael Strano. After completing his PhD, Dr. Paulson held a postdoctoral research position in Professor Ali Mesbah's group at the University of California, Berkeley.
While at UC Berkeley, he was a finalist for the 2017 International Federation of Automatic Control (IFAC) Conference Best Paper Award.
Professor Paulson has published several book chapters and over a dozen articles in such peer-reviewed journals as ACS Nano, Journal of Physical Chemistry Letters, Organic Process Research & Development, Journal of Process Control, and International Journal of Robust and Nonlinear Control.
KEY DISTINCTIONS
- NSF CAREER Award, 2023
- Winner, Application Paper Prize at IFAC World Congress, 2020
- Finalist for Young Author Prize at IFAC World Congress, 2017 National Science Foundation Graduate Research Fellowship, 2011-2016
- Director’s Student Presentation Award Finalist, AIChE CAST division, 2016
- School of Engineering Graduate Student Award for Extraordinary Teaching and Mentoring, MIT, 2015 (only 1 chosen per year across all departments)
- Outstanding Graduate Teaching Assistant Award, Chemical Engineering, MIT, 2015 Goodwin Medal Nominee, MIT, 2015
- Robert T. Haslam Fellowship, MIT, 2015
- Best Poster Presentation Award, Chemical Engineering, UT Austin, 2011 National Merit Scholarship, 2008
- Semifinalist, Siemens Research Competition, 2007
University of Texas at Austin:
- David H. Koch (1962) Fellow 2011 Best Poster Presentation Award, 2011. (Winner from over 50 posters.)
- Semifinalist, Siemens Research Competition, 2007
- TAMS Research Scholarship, 2007
- Bobby Bragan Scholarship, 2004
Research
LEARNING EFFICIENT REPRESENTATIONS OF MODEL PREDICTIVE CONTROL LAWS USING DEEP LEARNING
Innovation: Joel Paulson created a method for executing advanced control algorithms at a rate faster than the millisecond scale. Inspired from recent advances in machine learning and control theory, the method uses deep learning to efficiently approximate the behavior of complex optimization-based control schemes, while still providing strong theoretical guarantees in the presence of constraints and uncertainty
Impact: The proposed approach provides a path toward achieving high-performance in emerging safety-critical control applications including those that have traditionally been treated as “too challenging” to solve using state-of-the-art control methods.
Examples include the control of certain biomedical systems, unmanned vehicles, quadcopters, and humanoid robots. The method was recently demonstrated experimentally on an atmospheric pressure plasma device that can be used for biomedical purposes such as combating antibiotic-resistant bacteria, shrinking cancerous tumors, and accelerating the healing rate in chronic wounds.
- J.A. Paulson and A. Mesbah. Approximate closed-loop robust model predictive control with guaranteed stability and constraint satisfaction. IEEE Control Systems Letters. 4:719-724, 2020.
- A.D. Bonzanini, J.A. Paulson, D.B. Graves, and A. Mesbah. Safe dose delivery in fast sampling atmospheric plasmas using projection-based approximate economic MPC. IFAC World Congress, 2020.
FAST UNCERTAINTY PROPAGATION AND PARAMETER ESTIMATION IN COMPUTATIONALLY INTENSIVE GENOME-SCALE BIOLOGICAL MODELS USING MACHINE LEARNING
Innovation:
i. Construction and validation of mathematical models is biological systems involving genome-scale molecular networks is a very challenging problem. The task of uncertainty quantification (UQ) represents: (i) calibrating the model with experimental data and (ii) propagating uncertainties through the model to characterize the quality of the model predictions. Although many methods for UQ have been developed, the majority of them are intractable on experiment-to-evaluate computational models. I developed a novel metamodeling approach that can vastly accelerate UQ methods for dynamic genome-scale biological system models in the presence of high-throughput experimental data.
Potential Impact/Benefit to Society:
i. Modeling is known to have a big impact on process understanding and optimization; however, unless the accuracy of the model is rigorously understood, then the impact of the model is limited. Thus, the developed method helps to answer this question of accuracy. In particular, most process models are functions of several unknown parameters (such as heat transfer coefficients or rate constants) that must be estimated from data. Once this has been done using the proposed approach, we can easily decide if more experiments are required or even which experiments to perform in the future to gain a better understanding of the process.
ii. We applied the method to infer extracellular kinetic parameters in a batch fermentation reactor consisting of diauxic growth on E. coli on a glucose/xylose mixed media. To the best of our knowledge, due to the complexity of the model, this problem had been unable to be solved in the literature using standard methods. Our novel metamodel enabled this problem to be solved a factor of more than 800 times faster and provided significant physical insights that had previously been unknown (such as the reported data set was insufficient for uniquely estimating all parameters).
______
- J.A. Paulson, M. Martin-Casas, and A. Mesbah. Fast uncertainty quantification for dynamic flux balance analysis using non-smooth polynomial chaos expansions. PLOS Computational Biology, 15, e1007308, 2019.