Cross-domain pretraining boosts CFD neural surrogate performance

A new research paper explores cross-domain pretraining for neural surrogates used in computational fluid dynamics (CFD) simulations. The study demonstrates that pretraining models across diverse geometries, boundary conditions, and fidelities significantly improves their ability to generalize to new applications. Specifically, finetuning a pretrained model resulted in 2-3 times lower errors with the same amount of data and achieved the same error with 8 times fewer samples compared to training from scratch. The researchers found that simply pooling steady-state datasets was sufficient for effective pretraining, suggesting this approach can be a valuable strategy for accelerating engineering innovation by leveraging existing CFD data. AI

IMPACT Enhances the efficiency and generalization of AI models used in scientific simulations, potentially accelerating engineering innovation.

RANK_REASON The cluster contains a research paper detailing a new methodology for improving machine learning models in a scientific domain. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• Cross-Domain Pretraining for Steady-State Neural CFD Surrogates
• Hugging Face
• machine learning
• Neural CFD Surrogates

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →