Guiding Experiments in Materials Science Using Information Driven Approaches Based on Data as well as Theory

Abstract:

My aim is to show how data, as well as physics based models, can be utilized in conjunction with data science to guide materials discovery. For the former, with the continuing development of self-driving, autonomous laboratories, I will provide a perspective on the merits of active learning approaches, such as Bayesian Global Optimization and Reinforcement Learning, for accelerated materials discovery with application to solid-solid phase change alloys in the NiTi based family. For the latter, in contrast to purely data-driven approaches to construct surrogate models, I will show how a physics based model, using Landau theory, may be utilized to guide data science. Applications I will discuss here include the search for superelastic shape memory alloys with a wide temperature window (~ 300C) and BaTiO3 based ferroelectric ceramics with multiple properties.

Bio:

Turab Lookman obtained his B.Sc., Ph.D. degrees from Kings College, University of London, and held appointments at Western University, London, Ontario, and the University of Toronto in Canada until 1998. He was then at the Theoretical Division at Los Alamos National Laboratory till 2019, where he was elected Fellow of the American Physical Society (APS) in 2012 and a Laboratory Fellow in 2018. His interests and expertise lie in hard and soft condensed matter physics, materials science, applied mathematics and computational methods. His work on information directed approaches to materials discovery started in 2012 when he led an LDRD-DR funded by DOE to investigate how ML tools could be applied to accelerate materials discovery. Since 2022 he has been dividing his time between the US and China.

(https://scholar.google.com/citations?user=wRp-044AAAAJ&hl=en)