|About this Abstract
||2016 TMS Annual Meeting & Exhibition
||Computational Materials Discovery and Optimization: From 2D to Bulk Materials
||A General-Purpose Toolkit for Predicting the Properties of Materials using Machine Learning
||Logan Ward, Amar Krishna, Rosanne Liu, Vinay Hegde, Ankit Agrawal, Alok Choudhary, Chris Wolverton
|On-Site Speaker (Planned)
Machine-learning-based models offer the ability to rapidly predict properties that are difficult (or at least expensive) to compute using conventional, physics-based computational tools. In general, these approaches work by using machine learning (ML) to infer functional relationships between a material property and attributes derived from known information about a material using a large dataset of measurements. However clear in concept, designing the attributes has proven to be less-than-straightforward – as evidenced by the variety of techniques available in the literature. In this work, we present two general-purpose strategies for creating ML models: one if only the composition a material is known, and another if the crystal structure is also known. We will show that these strategies are capable of predicting many different material properties and, in some cases, are more accurate than other methods in the literature.
||Planned: A print-only volume