Quantifying Uncertainty in Model-Based Prediction

A two-day course by Ryan McClarren

The use of computational simulation and model-based prediction is ubiquitous in the world of engineering.  The question that does not get asked often enough is how much trust can we put in the results of these models.

Answering the question of how certain (or uncertain) predictions based on simulation are is crucial for

• Analysts presenting a results to decision makers
• Regulators reviewing the safety or risk analyses
• Technical managers deciding between different paths
• Computational scientists that want to inform users of uncertainties

This course is an opportunity to learn the theory and practice of giving defensible, quantitative uncertainties for simulations and numerical calculations. The course is jargon-free and only requires a knowledge of mathematics typically covered in undergraduate engineering and physical sciences curricula.

In this two day course, taught entirely by Ryan McClarren, attendees will get time for one-on-one discussion as well as interactive lectures. The outline for the course is

Day 1:

1. The landscape of UQ and Predictive Modeling
2. Modeling uncertainties with probability distributions (including correlated uncertainties)
3. Local Sensitivities via Derivatives
4. Regression analysis to estimate sensitivities
5. Adjoint-Based Sensitivities

Day 2:
1. Sampling Based Uncertainty Quantification
2. Reliability Methods for rapid estimation of the probability of failure
3. Stochastic Projection and Collocation (Polynomial Chaos and its cousins)
4. Surrogating modeling and Gaussian Process Emulators
5. Predictive Models and Multi-Fidelity Simulations
6. Dealing with Epistemic Uncertainties

All attendees will receive a copy of the book Uncertainty Quantification and Predictive Computational Science. Working codes in Python will be available to students that are interested, but the course is language agnostic and computer programming knowledge is not a prerequisite.

During the one-on-one discussion periods attendees can have a consultation with Prof. McClarren about the particular scenarios of in each’s own work.

For more information and how to register please contact workshop@drryanmc.com

Upcoming workshops are planned for the DC area, Santa Fe, Portland, and Knoxville.