| About this Abstract |
| Meeting |
2024 TMS Annual Meeting & Exhibition
|
| Symposium
|
AI/Data Informatics: Computational Model Development, Verification, Validation, and Uncertainty Quantification
|
| Presentation Title |
Global Uncertainty Reduction Through Efficient Acquisition Function Candidate Selection in Predefined Design Spaces for Predicting NMR Peak Positions |
| Author(s) |
Ramsey Issa, Taylor D. Sparks |
| On-Site Speaker (Planned) |
Ramsey Issa |
| Abstract Scope |
Most materials science tasks involve defined objectives that seek to optimize material properties to enhance performance. For example, when developing high entropy alloys, one might seek to maximize the yield strength or minimize corrosion rate. These tasks are typically carried out in an adaptive design strategy using Bayesian optimization, where the acquisition function selects candidates that will optimize the material property. But how do we select candidates when objectives are not as clearly defined? For example, characterization tools like NMR generate data in the form of spectra, which are not tied to a specific application. As the optimal spectra can vary based on application, an active learning strategy that emphasizes model accuracy based on the most informative candidate, specifically through uncertainty reduction, becomes crucial. Here, we show using the negative integrated posterior variance acquisition function, that candidates can be selected iteratively, minimizing costly DFT simulations to predict NMR peak positions. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, Computational Materials Science & Engineering, Modeling and Simulation |