|About this Abstract
||2018 TMS Annual Meeting & Exhibition
||Computational Method and Experimental Approaches for Model Development and Validation, Uncertainty Quantification, and Stochastic Predictions
||The Role of Data Analysis in Uncertainty Quantification: Examples from Materials Science
||Paul Patrone, Andrew Dienstfrey
|On-Site Speaker (Planned)
Scientists must often confront noise in raw data when attempting to extract meaningful information and estimate uncertainties. Oftentimes, however, the task of assessing the extent to which an analysis routine itself affects uncertainties is overlooked. In this talk, I therefore consider three examples where attention to analyis methods can be leveraged to better understand uncertainties in information extracted from datasets common in materials science. In particular, I consider issues pertaining to: (i) time-averaging of simulation data generated by molecular dynamics; (ii) spectral approaches to constructing distribution functions; and (iii) estimation of critical points associated with yield in simulated stress-strain curves. As a key theme of this discussion, I show how the quality of extracted results can be improved by accounting for global constraints underlying data, such as correlations, smoothness, and convexity.
||Planned: Supplemental Proceedings volume