We are developing machine learning algorithms to accelerate the discovery and optimization of advanced materials, as part of the Materials Genome Initiative (MGI, https://www.whitehouse.gov/mgi). These new algorithms will form part of a data analysis system that integrates machine learning, solid state physics, experimental materials properties databases, theory-based databases, and the theory of measurement instruments (e.g. X-ray diffractometers) to provide high throughput analysis of materials data.
We are primarily interested in the high throughput analysis of experimental data measured on combinatorial "libraries," both offline and on-the-fly (e.g., real-time, during the measurement experiment), to provide live guidance to improve data collection for the experimentalist. We are particularly interested in hyperspectral methods (algorithms for analyzing spectra collected over 2D coordinates) to analyze X-ray diffraction and Raman spectra data collected from combinatorial libraries. Machine learning methods include, but are not limited to, constraint programming, Bayesian methods, sparse kernel machines, graphical models, and latent variable analysis.
Some examples of materials classes of interest for this project are photovoltaic, thermoelectric, thermochromic, multiferroic, magnetic, and superconductive materials.
Kusne AG, et al: High Throughput Determination of Structural Phase Diagram and Constituent Phases using GRENDEL. Nanotechnology (26)44: 444002, 2015
Mueller T, Kusne AG, Ramprasad R: Machine Learning in Materials Science: Recent Progress and Emerging Applications. Reviews in Computational Chemistry, June 2015
Materials Genome Initiative; Machine learning; Combinatorial library; Informatics; High-throughput; Composition spread; Hyperspectral data Analysis; Data mining; Functional materials;