After completing her undergraduate degree in chemical engineering at the University of Minnesota, Alison Cozad joined the research group of Professor Nick Sahinidis as a doctoral student in the Department of Chemical Engineering at Carnegie Mellon University in the fall of 2009. The main goal of her project was to develop systematic optimization techniques for simulation-based optimization. To address this problem, Alison began studying a central problem in modeling, namely that of learning an algebraic model from data obtained from simulations or experiments. She developed a systematic methodology that uses a small number of simulations or experiments to learn models that are as accurate and as simple as possible. The simplicity of these models is a unique distinguishing feature of Alison’s work in contrast to prior work in machine learning and design of experiments.
Alison’s methodology has been implemented in the Automated Learning of Algebraic Models for Optimization (ALAMO) software that has been crucial in the development of surrogate models for the Department of Energy National Energy Technology Laboratory’s (NETL) Carbon Capture Simulation Initiative (CCSI) project. Following completion of the basic ALAMO methodology, Alison proceeded to develop the first systematic methodology for simultaneous data- and science-driven discoveries. ALAMO relies on simulation (or experimentation) to collect data and build simple algebraic models that describe the data. However, in many cases, the quality of surrogate models can be improved by relying on partial theoretical understanding of the system at hand. To capitalize on science-based understanding of a system during surrogate model building, Alison proposed the use of constraint regression in combination with a practical approach to solve the underlying semi-infinite optimization problems. This development opens up the road for a vast array of applications. For instance, ensuring the non-negativity of a modeled geometric length or pressure drop, enforcing a sum-to-one constraint on modeled chemical compositions, and ensuring that derivative bounds obey thermodynamic principles are all practical applications of Alison’s constrained regression approach to model building.
Alison’s doctoral research has also led to a stunning contribution to a field in machine learning known as symbolic regression, the objective of which is to identify a nonlinear function that discovers relationships hidden in data. All previously known symbolic regression techniques search for such a function via genetic (biomimetic) operations that offer no guarantee of quality of the final solution. Alison was able to produce the first deterministic (mixed-integer programming nonlinear optimization) model that is guarantee to find the best possible solution for symbolic regression problems.
Alison’s doctoral work was groundbreaking and highly innovative. In recognition, the Department of Chemical Engineering at Carnegie Mellon University has recognized Alison with the 2014 Ken Meyer Award. The Ken Meyer Award is presented every year to a senior doctoral student who has demonstrated excellence in graduate research in chemical engineering. The faculty bases their selection of the student for this award on research quality, productivity, recognition, and impact.
The CCSI Team congratulates Dr. Cozad on her accomplishments and contributions to our project. We wish her the best for the future.