Astro Working Group I: Uncertainty Quantification and Astrophysical Emulation (UQAE)

Group Leaders: Derek Bingham (Simon Fraser), Earl Lawrence (LANL)

SAMSI Webmaster: David Stenning

Weekly Meetings:  Mondays Noon-2:00 pm ET

Description: Uncertainty Quantification (UQ) and Reduced Order Modeling (ROM) are at the core of many problems in gravitation and cosmology, from direct simulations of the Einstein equations to the inverse (inference problem). The goal of this working group is to leverage expertise from areas such as UQ, ROM, generalized polynomial chaos and simulator emulation based on stochastic processes, and domain-areas such as gravitation, astrophysics, and cosmology.

Big Questions:

  1. If simulations from a complex model are so expensive that one can only afford to run a small number of the full model, how does one develop a robust framework utilizing fast emulators? The WG topics include design of computer experiments; task-driven active learning, reduced order modeling among others.
  2. How do we develop rigorous UQ for stochastic processes?
  3. Inverse problem: how does one carry out inference (either on individual signals, or on population distributions from multiple observations) with inexact models?  The WG topics include model discrepancy; model misspecification; nonparametric models.
  4. How do we address the so-called forward problem of stochastic time-dependent problems with model, boundary or initial data uncertainties?
  5. How does one establish confidence in a signal when there is a single experiment, one is looking for very rare outlier events, and the “tails” of the noise background distribution are difficult to measure?
  6. If both models and observations have uncertainties, what are statistical techniques for clustering observations in a large-dimensional parameter space?

News and Updates: Coming soon…

SAMSI Directorate Liaison: Sujit Ghosh

Questions: dstenning@samsi.info

Astronomy Opening Workshop