Argonne National Laboratory

Seminar Series

Date Title

March 28, 2017

2:00 pm

Bldg. 440, A105-106

"Using machine learning to guide materials design and discovery", Max Hutchinson, Citrine Informatics.  Host:  Maria Chan

 

Machine learning (ML) encompasses a diverse set of techniques for constructing quantitative models given examples of a phenomenon.  In materials informatics, the primary challenge is data scarcity: the space of materials is very high dimensional, and experiments, including those in silico, are relatively expensive.  It is therefore especially important to focus experiments on regions of interest and high information density.  This talk presents a general experimental design strategy that uses the predictions of machine learning models to select subsequent experimental targets.  This technique is benchmarked on a search for high ZT thermoelectric materials, and it accelerates the rate at which extreme values are found by nearly 3x.  This technique can be easily applied to other materials design and discovery problems, and the talk will conclude with a demo of optimal experimental design using the Citrination platform.

March 29, 2017

3:00 pm

Bldg. 440, A105-106

“Machine Learning for Accelerating Materials Design", Tarak Patra, Theory and Modeling Group, Nanoscience and Technology Division, Argonne National Laboratory