Cross-Domain Pretraining for Steady-State Neural CFD Surrogates
Zhou et al.
What it does
Pretrains one CFD surrogate on pooled steady-state datasets that differ in geometry, boundary conditions and fidelity, then fine-tunes it on a held-out dataset.
Why it matters
Pretraining improves zero- and few-shot accuracy over training from scratch or transferring a single-domain expert, cutting error several-fold at the same sample size, and the gain grows with model size and data diversity.