Materials Learning Software

Active Learning for Accelerated Design of Layered Materials

Hetero-structures made from vertically stacked monolayers of transition metal dichalcogenides hold great potential for thermoelectric devices of the future. Discovery of the optimal layered material for specific applications necessitates the estimation of key material properties, however, screening of properties via brute force ab initio calculations of the entire material structure space exceeds the limits of current computing resources.

We have applied active learning to discover, with high probability, which structure, in a vast structure search space, is optimal with respect to a desired property using minimal computation. Specifically, Bayesian Optimization is used to predict the optimal n-layered TMDC hetero-structure for thermoelectric devices.

In our Active Learning scheme, a small, randomly selected set of structures is computed and used as an initial training data set. Then, the Bayesian optimization process suggests the next structure to compute which will most likely exhibit a desired property (exploitation), or which will allow us to learn the most about the structure space (exploration). The structure is automatically generated and sent to the Materials Project Database for computation. After its computation, the structure is added to the initial training data set and the cycle begins again, ending after some user-defined threshold is reached. With high probability, a structure holding a desired property can be found from a large search space, performing minimal amount of expensive computation. The figure below shows a schematic of the workflow.

For more information and to view the code, check out our Github Repository .

Bassman, L., Rajak, P., et al., npj Computational Materials 4, 74 (2018).
Bassman, L., Rajak, P., et al., MRS Advances 3, 397-402 (2018).

ML Defect Classification Tool

  • Machine Learning Defect Classification Tool performs phase and defect classification, and predict optimal layered materials with desired functionality such as thermoelectricity using active learning.
Repository

https://github.com/USCCACS/ML-defect-analysis