Problem
Research questions and friction points this paper is trying to address.
distribution shift
pretraining
neural PDE surrogates
CFD data
How Does Distribution Shift Shape Pretraining Gains in Neural PDE Surrogates?
๐ค AI Summary
็ ็ฉถ้่ฟ้ข่ฎญ็ป็ฅ็ปPDEไปฃ็ๆจกๅๅจไธๅๅๅธๅ็งปไธๅๅฐๆฐCFDๆฐๆฎ้ๆฑ็ๆๆ๏ผๅ็ฐๅ
ถไปทๅผๅ็ฎๆ ๆฐๆฎ้ใ่ฆ็่ๅดๅ็ฉ็ๆจกๅๅทฎๅผๅฝฑๅใ
๐ Abstract
Pretraining a neural PDE surrogate can reduce the amount of new CFD data needed when geometry or modeled physics changes. However, it remains unclear how different components of distribution shift affect this benefit. We pretrain a surrogate on 254,909 RANS solutions from one airfoil family and fine-tune it on a new family under two target settings with matched freestream ranges: the same Spalart-Allmaras (SA) modeling and SA with added $e^N$ transition modeling. At $N=1000$, the pretrained model matches the accuracy of a model trained from scratch on $3.25\times$ as many samples for the same-SA target, but $2.58\times$ as many for the transition-modeled target. By $N=5000$, this ordering reverses ($1.56\times$ versus $1.86\times$). At $N=1000$, sampling more distinct airfoils lowers error on both targets, but only for the same-SA target is the gain increase larger than the observed draw-to-draw variation ($3.3\times$ to $4.0\times$). These results show that pretraining value depends jointly on target-data budget, target-data coverage, and whether source and target differ in modeled physics.
Innovation
Methods, ideas, or system contributions that make the work stand out.
pretraining
distribution shift
neural PDE surrogates
CFD data
physics modeling
๐ Similar Papers
No similar papers found.
