Uncertainty Quantification in Computational Predictive Science

Computational predictive models are standard tools for analysis of complex systems. However, due to variabilities on their parameters and, possible wrong assumptions made on their conception, they are uncertain with respect to the real system. The first source of uncertainty is inherent to measurements limitations, material variabilities, etc. Meanwhile, the second type is essentially due to lack of knowledge about the underlying governing laws. Uncertainty Quantification (UQ) is a multi-disciplinary area that deals with quantitative characterization and reduction of uncertainties in applications, which is extremely necessary to give robustness to computational forecasts. This short course covers the basic UQ vocabulary; basic notions on probability and statistics; some approaches to model uncertainties; the main uncertainty propagation techniques. Computer activities are developed in parallel to theoretical expositions, as a way to give a hands-on tone to the course.