| About this Abstract |
| Meeting |
MS&T23: Materials Science & Technology
|
| Symposium
|
2023 Undergraduate Student Poster Contest
|
| Presentation Title |
Machine Learning Model Explainability for the Development of High-entropy Alloys |
| Author(s) |
David Flores, Wesley Reinhart, Arindam Debnath |
| On-Site Speaker (Planned) |
David Flores |
| Abstract Scope |
High Entropy Alloys (HEAs) are compositionally complex materials with unique mechanical properties. However, their complexity also makes them difficult to design, prompting use of data-driven surrogate models to elucidate their composition-processing-property relationships. While accurate, the inner workings of these ‘black-box’ models are incomprehensible to humans in a meaningful scientific context.
In this work, we develop a set of design principles for these relationships using automated “explainable AI” techniques. The workflow consists of training and validating surrogate models, and performing and processing various feature importance calculations to distill a set of design rules catered to domain scientists. This process of knowledge distillation will generate new understanding of the “black-box”, and of HEAs.
These methods will provide further insight into what information surrogate models learn, supporting ongoing studies in multi-task and transfer learning. From a broader perspective, this knowledge will facilitate interdisciplinary collaboration between machine learning researchers and domain experts. |