Jason Hattrick-Simpers
1 Department of Materials Science and Engineering University of Toronto
Natural Resources Canada CanmetMATERIALS
EXTENDED ABSTRACT: Artificial intelligence and high-throughput experimentation are increasingly being combined to generate self driving laboratories, scientific robots tasked with autonomously exploring the space of possible materials. These tools have already been demonstrated to generate lOx - lOOx accelerations in the discovery of new materials and recent studies have even begun to display their potential in helping elucidate mechanism. What is more exciting is that there exists an opportunity to use the Al's developed by these tools as a mirror to help scientists understand how our own biases, assumptions, and errors influence the data, labels, and values we use to train our models. Here I will touch on some of our recent work in this field. The first point of discussion will fbcus on how "ground truth" labels can be more subjective than we expect and demonstrate how one can use label uncertainty to balance model performance with expert uncertainty. We will then discuss how dataset bias can cause materials datasets to violate the assumptions of IID leading to disastrously poor predictions for truly new materials and discuss an active learning approach we have recently developed to identify and recommend out of distribution data points for mvestigation. The final point of discussion will be focused on how statistical tools can be used to extract more information than we might have thought from the data we already have. A particular example will be of how a full 3D reconstruction of a membrane can be generated from a single 2D SEM cross section.
Jason Hattrick-Simpers is a Professor at the Department of Materials Science and Engineering, University of Toronto and a Research Scientist at CanmetMATERIALS. He graduated with a B.S. in Mathematics and a B.S. in Physics from Rowan University and a Ph.D. in Materials Science and Engineering from the University of Maryland. Prof Hattrick-Simpers's research interests fbcus on the use of Al and experimental automation to discover new functional alloys and oxides that can survive in extreme environments and materials for energy conversion and storage. Specific topics of interest to the group include corrosion resistant ultra-hard alloys, oxides, nitrides, and carbides; thermoelectric materials for heat to energy conversion; novel metals for hydrogen fueling stations; and oxides for CO2 conversion.
Prior to joining UofT Prof. Hattrick-Simpers was a staff scientist at the National Institute
of Standards and Technology (NIST) in Gaithersburg, MD where he co-developed tools for discovering novel corrosion resistance of alloys, developed active learning approaches to guide thin film and additive manufacturing alloy studies, and developed tools and best practices to enable trust in Al within the materials science community. He has published over 80 papers and given more than 50 invited seminars and talks. He was an associate editor of ACS Combinatorial Science from 2017 - 2020 and is part of the organizing committee for the International Workshop on Combinatorial Materials Science and Technology.