Geometry-Aware Post-Hoc Uncertainty Quantification in Operator Learning

In three linesREEF-GP, a post-hoc uncertainty quantification framework for neural operators, adapts the operator's intrinsic representations to construct geometry-aware uncertainties. Tested on 5 PDE benchmarks, it preserves predictive accuracy while providing calibrated uncertainty estimates, more efficient than deep ensembles.

Summary generated by Claude — human-verified