Machine Learning for Materials Design
From Santa Fe Institute Events Wiki
|Working Group Navigation|
organized by David Wolpert (Los Alamos National Laboratory and SFI External Professor), Frank Alexander and Turab Lookman (Los Alamos National Laboratory)
The participants in the working group will help create a roadmap for developing a robust and accessible knowledge discovery process for materials research using methods from informatics and data driven modeling. This will include identifying areas of research required in the computer science / applied mathematics community that can help to address pressing needs in accelerating the discovery of new materials with targeted properties. The group brings together experts in machine learning (ML), statistics, computational physics and materials science who will first examine the current state of the art in which off-the-shelf methods have so far been used in a manual and heuristic manner. They will subsequently study how to cast the problem within a consistent machine learning and optimization framework involving, for example, a probabilistic approach. This will facilitate using robust methods developed, for example, in the biosciences literature to deal with small data sets in terms of distributions encoding prior knowledge which can then make posterior predictions with confidence levels. The same applies to the method of Probability Collectives developed in the ML field.
Materials Informatics (MI) is an emerging area that has gained prominence at the national level (as the Materials Genome / Advanced Manufacturing Initiative) as well as with funding agencies, such as the DOE and NSF. However, unlike the mature field of bioinformatics, MI is a nascent area where few information-theoretic tools have been formulated or applied. Hence, this is an excellent opportunity for SFI to take the initiative by supporting this working group as they distill and evaluate the appropriateness of methodologies from computers science/systems engineering as well as mathematics to the problem of accelerating materials discovery using heterogeneous data from multiple measurement sources and theoretical calculations.