Using Machine Learning to Determine Structure-Property Relations in Materials

Description

The Material Measurement Lab at NIST employs machine learning (ML) for the prediction and discovery of  materials characteristics. Our applications of machine learning accelerate materials research as well as help the community learn about ML capabilities and gain confidence in similar applications. In addition to the projects below, we use machine learning for data extraction and uncertainty predictions. Many of these projects are part of NIST's contribution to the federal multi-agency Materials Genome Initiative. 

Projects:

• JARVIS-ML  (Materials properties prediction using machine learning) - JARVIS is a repository designed to automate materials discovery using classical force-field, density functional theory, machine learning calculations and experiments. This project provides machine learning prediction tools trained on JARVIS-DFT data. Some of the ML-predictions focus on energetics, heat of formation, GGA/METAGGA bandgaps, bulk and shear modulus. 
• Deducing Prior Deformation from Simple Mechanical Analysis
• High-Performance Crystal Plasticity by Machine Learning-Interpolation
• Physically Informed Neural Network (PINN) Potentials - Machine learning interatomic potentials
  • Gaussian process regression
  • Interpolating moving least squares
  • Kernel ridge regression
  • Compressed sensing
  • Artificial neural network potentials
  • Physically informed neural network potentials (Mishin, 2017)
  • Contact james.hickman [at] nist.gov (James Hickman) and yuri.mishin [at] nist.gov (Yuri Mishin)
• Teaching Liquid State Theory to an Artificial Neural Network (ANN)
• Using Machine Learning to Predict DNA Sequences for CNT Fractionation (Learn more)
• Materials Design Toolkit

 

Created March 25, 2019, Updated October 30, 2026