Efficient Techniques for Quantifying Uncertainty

Jasmine Foo, Brown University

Photo of Jasmine Foo

Quantifying uncertainty in systems with noisy or unknown parameters is useful in model prediction and design. Stochastic spectral methods are a class of numerical methods designed as an alternative to Monte Carlo for more efficiently quantifying uncertainty via approximation in the random space. Here we introduce the Multi-Element Probabilistic Collocation Method and an extension using an ANOVA-type decomposition, which provide improvement in efficiency and functionality over existing stochastic spectral and Monte Carlo methods. We provide analysis of h/p-type convergence rates as well as numerical examples from applications in physics and biology.

Abstract Author(s): Jasmine Foo, Xiaoliang Wan, George Karniadakis