New method ranks neural operator models without ground truth data

Researchers have developed a novel method to efficiently select the best neural operator model for deployment without needing high-fidelity reference solutions. By analyzing the low-dimensional span of candidate differences under a squared Hilbert-space loss, a single linearized response of the governing equation can score all models simultaneously. This approach accurately recovered over 99.6% of pairwise preferences and 99.0% of optimal checkpoints across various operator libraries for fluid, reaction-diffusion, and wave dynamics. The corrected physical proxy often surpassed individual best candidates, offering a reliable and efficient way to deploy scientific surrogates. AI

IMPACT Enables more efficient deployment of scientific AI models by reducing the need for extensive ground truth data.

RANK_REASON The item is a research paper published on arXiv detailing a new method for evaluating neural operator models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• convolutional operator libraries
• fluid dynamics
• Fourier
• Hugging Face
• Neural Operator Libraries

• paper
• infra

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