About this Abstract |
Meeting |
2nd World Congress on High Entropy Alloys (HEA 2021)
|
Symposium
|
2nd World Congress on High Entropy Alloys (HEA 2021)
|
Presentation Title |
Building Data-driven Models with Noisy Input Features-
Application: Strength Prediction of High Entropy Alloys |
Author(s) |
Sharmila Karumuri, Zachary McClure, Ilias Bilionis, Alejandro Strachan, Michael Titus |
On-Site Speaker (Planned) |
Sharmila Karumuri |
Abstract Scope |
Building models for predicting various properties of interest using descriptor information obtained from experiments is common practice. However, the issue is that some of the descriptor information coming from these experiments could be noisy leading to input uncertainty. Using these noisy inputs with traditional regression techniques, e.g., Gaussian process regression, is likely to lead to poor performance. To overcome this issue, we propose a hierarchical Bayesian approach which denoises the inputs prior to connecting them to the output quantity of interest. We demonstrate the problem and our proposed approach by carrying out a comparative study using noisy hardness information vs using denoised hardness information to predict the strength of high entropy alloys. |
Proceedings Inclusion? |
Undecided |