Products

MAST-SEY

MAST-SEY is an open-source Monte Carlo code capable of predicting secondary electron emission using input data generated entirely from first principle (density functional theory) calculations.

Resources for Machine Learning in Materials

Software
- MAterials Simulation Toolkit – Machine Learning (MAST-ML) is an automated tool for setting up, executing, and managing output machine learning tasks in materials science.
  - MAST-ML documentation | Source (MAST-ML)
Educational materials
- Modules and course plans for introductions to machine learning for materials scientists.
- ML curriculum
- One week “Introduction to Machine Learning for Materials Science” Lab
- 5-minute machine learning prediction activity
- “An Introduction to Machine Learning for Materials Science: A Basic Workflow for Predicting Materials Properties“, nanoHUB online presentation by Benjamin Afflerbach
- “The Materials Simulation Toolkit for Machine Learning (MAST-ML): Automating Development and Evaluation of Machine Learning Models for Materials Property Prediction“, nanoHUB online presentation by Ryan Jacobs
Databases: Material’s property data sets and descriptors for easy exploration of machine learning.
- ML materials database | Source (MAST-ML)
Best practices: Guidelines for publishing papers with machine learning models
- Repository on Materials Standards | Specific checklist for machine learning papers

Web apps and materials for learning about optimization (high-school level)

These materials were created by Marc Brousseau, a teacher at Middleton high school in Madison, WI, working with graduate student Ben Afflerbach in the Morgan group. This work was done as part of a Research Experience for Teachers (RET) program (https://mrsec.wisc.edu/ret/) at UW Madison run by the Materials Research Science and Engineering Center (MRSEC), funded under grant number DMR-1720415.

- - Optimization of Area of a Box: https://www.geogebra.org/classic/zqeafevg
  - Local Extrema with Two Independent Variables: https://www.geogebra.org/classic/waj5hm97
  - Presentation: https://docs.google.com/presentation/d/1L7QNNfpbY7L6Xd3qmN9B3E-bm0-y8TsTwHHeayC06j8/edit?usp=sharing
  - Handout: https://docs.google.com/document/d/1oWJQPpjzGAGIsjV0md-gpJr5bmO1M0yTHMGPjQBRYG8/edit?usp=sharing

StructOpt app

The StructOpt app provides a general genetic algorithm based optimizer in python targeted at identifying stable atomic structures.

MAterials Simulation Toolkit (MAST)

MAST is an automated workflow manager and post-processing tool that focuses on diffusion and defect workflows using density functional theory. It interfaces primarily with the Vienna Ab-initio Simulation Package (VASP).

Download code | MAST manual | Source (MAST) DOIs for MAST versions: dx.doi.org/10.5281/zenodo.11916; dx.doi.org/10.5281/zenodo.11917

Diffusion Data App

Diffusion Data App is a database of over 250 (and growing) impurity diffusion coefficients calculated with ab initio density functional theory. Includes hosts Mg, Al, Cu, Ni, Pd, Pt, and W.

MaterialsHUB

MaterialsHUB is an online hub of applications for computational materials science research and education related to defects and diffusion.

AtomTouch

AtomTouch is a 3D interactive mobile platform (e.g., phone, tablet) molecular simulation software tool for educational purposes.

UW links: Project page, Online App, Source code on Github.
IOS App – “AtomTouch” on Google play and iTunes
Tool and educational material on BrainPOP

Material Application Domain Machine Learning (MADML)

MADML is software package to develop and assess regression models from tabular data. The developed models provide property predictions, property prediction uncertainty, and flags whether a model’s prediction comes from the model’s domain.

Transfernet

Transfernet is a software package for training neural networks developed with PyTorch and then transferring trained weights to another models (i.e., transfer learning). The transfer can be from one neural network weights to another or using the output from a hidden layer as features for other models (e.g., Random Forest, LASSO, SVR, etc.).

Multilearn

Multilearn is a software package for training neural networks developed with PyTorch that can learn on multiple data sets concurrently (i.e., multi-task learning). The data sets can have different number of observations and features.