CAE-UQ

Fits three uncertainty methods to the same GeoTransolver surrogate: a sparse variational Gaussian-process output head with a learned residual variance, concrete Monte Carlo dropout, and a five-member deep ensemble. It tests them at three scopes: whether a whole new case can be trusted, where in the field the error lies, and how uncertain a derived quantity such as drag or intrusion is.

Data

AirfRANS

Share on LinkedIn, X or Bluesky, or copy the link or the citation.