Tuesday, March 17, 2020, 1:00–2:00PM
Building 1, Room 1107
Tuesday, March 17, 2020, 3:00–4:00PM
VTC to Building 101, Lecture Room C
Host: Andrew Dienstfrey
Abstract: In the fields of specialty chemical and materials manufacture, the ability to respond quickly to customer need leads directly to substantial competitive advantage. Discovering new materials to balance the tightening constraints of environmental concern and expected performance has historically been a very time and cost intensive process. Applying data-driven approaches to exploration of the materials design space is often challenging for a wide range of reasons: relevant experimental data points are few and expensive to generate; the process history of individual samples is often complex and unique; and large parts of the space are physically inaccessible. The confluence of complexity, rarity and sparsity creates the necessity to incorporate priors and physical understanding in order to maximize the statistical power of one’s corpus, in contrast to many of the approaches adopted in Big Data and Deep Learning.
Citrine is the artificial intelligence platform for materials and chemicals. Our platform’s data resources and AI workflows help organizations hit R&D and manufacturing milestones in less than half the time. We work with the world’s most advanced companies and research groups in: alloys, semiconductors, solar, optoelectronics, thermoelectrics, batteries, composites, polymers, small molecules, 3d printing, formulations and have cut years from the development times of new products in these categories. In this presentation, we will discuss the iterative process we call “sequential learning”1, which leverages families of models to identify the best experiments to perform to reduce model uncertainty and/or identify the highest-performing materials. This will include lessons learned around effective strategies in model validation and uncertainty estimation when presented with real customer data.
1 Ling, J. et al. High-Dimensional Materials and Process Optimization Using Data-Driven Experimental Design with Well-Calibrated Uncertainty Estimates. Integr Mater Manuf Innov 6, 207–217 (2017). https://doi.org/10.1007/s40192-017-0098-z
Bio: Dr. Kenneth Kroenlein is the Scientific Data Architect at Citrine Informatics, acting as the steward for their novel graph-based data model on their materials informatics platform. After graduation from Princeton University with a PhD in Mechanical Engineering in 2007, he joined the Thermodynamics Research Center at the National Institute of Standards and Technology to continue pursuing his interests in optimal recommended values for thermophysical and thermochemical properties for well characterized systems. He joined Citrine in 2019. His research interests cover a broad range of the scientific areas such as phenomenological thermodynamics, computational physics and numerical methods, information management and communication, and machine learning techniques. Dr. Kroenlein serves on board of the International Association of Chemical Thermodynamics (IACT) as secretary, on the organizing committee for the Symposium on Thermophysical Properties and for the International Workshop on Combinatorial Materials Science and Technology, and on the editorial board of the Journal of Physical and Chemical Reference Data.
Note: Visitors from outside NIST must contact Cathy Graham; (301) 975-3800; at least 24 hours in advance.
This talk will be broadcast on-line using BlueJeans. Contact acmdseminar [at] nist.gov (acmdseminar[at]nist[dot]gov) for details.