| About this Abstract |
| Meeting |
MS&T24: Materials Science & Technology
|
| Symposium
|
Frontiers of Machine Learning on Materials Discovery
|
| Presentation Title |
Machine Learning Materials Properties with Accurate Predictions, Uncertainty Estimates, Domain Guidance, and Persistent Online Accessibility |
| Author(s) |
Ryan Jacobs, Lane Enrique Schultz, Paul Voyles, Dane Morgan |
| On-Site Speaker (Planned) |
Ryan Jacobs |
| Abstract Scope |
This work addresses a critical need in the materials science community, which is the availability of persistent, easily accessible and useable machine learning models of materials properties that provide the user with predictions, uncertainties on those predictions (i.e., error bars), and guidance on domain of applicability to inform the user whether the model is reliable. We develop random forest models for 33 materials properties spanning an array of data sources (computational and experimental) and property types (electrical, mechanical, thermodynamic, etc.). All models have calibrated ensemble error bars and domain of applicability guidance enabled by kernel-density-estimate-based feature distance measures. All models are hosted on the Garden-AI infrastructure, providing an easy-to-use, persistent interface for model dissemination callable with only a few lines of python code. We demonstrate the power of our approach by using our models in a fully ML-based materials discovery exercise to search for stable, highly active perovskite catalyst materials. |