Active Learning for Surface-CFD Aerodynamic Surrogates

This example lives under the CFD examples tree, alongside the surface-CFD backbone it builds on: `physicsnemo/examples/cfd/external_aerodynamics/active_learning_aero <../../cfd/external_aerodynamics/active_learning_aero/README.md>`__.

It demonstrates end-to-end active learning on the ShiftSUV surface-CFD dataset using an uncertainty-aware GeoTransolver + Variational GP head, with three plug-and-play acquisition strategies (UQ-driven, class-balanced random, and latent-novelty). The AL loop itself is problem-agnostic — only the physics/metrology hooks are CFD-specific — so the same recipe can drive any uncertainty-quantified regression task.

See the full README for the recipe overview, configuration, results, and adapting-to-a-new-problem guide.