Uncertainty Quantification in Glass Properties Prediction
Friday, July 31, 2015 - 11:30am - 11:45am
ARMS Room 1010
Adama Tandia (Corning Incorporated)
Glass design is a very complex and expensive process. To speed up a fundamental understanding of the relations between glass compositions and properties, modeling & simulation is heavily relied upon. But, because of the lack of physics based models for key properties of multicomponent glasses, we turn to data driven modeling approaches like Neural Networks and Symbolic Regression. Experimental data scarceness and/or irregularity are some of the main reasons why modelers consider uncertainty quantification to address issues related to uncertainty propagation, model parameterization, optimal sampling techniques for data generation/augmenting, and validation.