Scalable Methods for Uncertainty Quantification

A key challenge when performing uncertainty quantification (UQ) with modern computational models is achieving acceptable accuracy in statistical quantities of interest when the scale of the UQ problem moves beyond a handful of random dimensions.  When the dimension of the problem is significant (tens to hundreds or random variables), typical approaches based on global approximation (e.g., polynomial chaos, sparse interpolation, Gaussian processes) suffer from the curse of dimensionality, requiring both algorithmic techniques that can effectively discover lower dimensional features within a higher dimensional problem and computational techniques that can effectively exploit emerging computer hardware.

This minisymposium will focus on the latest research and development in the area of scalable approaches to UQ, including algorithms for adaptive refinement, l1-regularized regression, and general dimension reducing techniques, as well as computational techniques that exploit hybrid computing architectures.