Zachary Ulissi is an Assistant Professor of Chemical Engineering at Carnegie Mellon University. He works on the development and application of high-throughput computational methods in catalysis, machine learning models to predict their properties, and active learning methods to guide these systems. Applications include energy materials, CO2 utilization, fuel cell development, and additive manufacturing. He has been recognized nationally for his work including the 3M Non-Tenured Faculty Award and the AIChE 35-under-35 award among others. He has long been part of the national lab HPC community having done his PhD as a DOE CSGF fellow at MIT.
Qualitative Choices in Representations for Molecules, Materials, and Surfaces
Abstract: Applying machine learning methods to atomistic systems starts with a choice on how to represent the system. The number of ways to represent molecules or materials in machine learning model development has exploded in the past few years and depends on the type of system, number of chemical elements, necessary properties, among many other factors. The choice of representation can determine the overall accuracy for a given task as well as how the accuracy scales with increasing dataset sizes. I will review the major classes of representations for common structures starting with small molecules, then discussing bulk materials, inorganic surfaces, and nanoparticles. I will also discuss the common benchmark datasets used to compare these representations, and efforts to automatically try and select the best representation for a given system. Finally, I will show how these choices can become complicated for challenges like catalysis, which blend challenges of the small molecule and bulk material worlds.