![[2412.13897] Data-Efficient Inference of Neural Fluid Fields via SciML Foundation Model](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2412.13897] Data-Efficient Inference of Neural Fluid Fields via SciML Foundation Model
Summary
This article presents a novel approach to data-efficient inference of neural fluid fields using SciML foundation models, demonstrating significant improvements in fluid dynamics reconstruction with reduced data requirements.
Why It Matters
The research addresses the challenges of traditional methods in fluid dynamics that require extensive data collection. By leveraging SciML foundation models, the study offers a more efficient solution, potentially transforming applications in 3D vision and fluid dynamics, which are critical in various scientific and engineering fields.
Key Takeaways
- SciML foundation models can significantly reduce data requirements for inferring real-world fluid dynamics.
- The proposed method improves generalization and forecasting capabilities in fluid dynamics applications.
- Collaborative training strategies enhance the performance of neural fluid fields.
- Quantitative metrics show a 9-36% improvement in PSNR while reducing training frames by 25-50%.
- The findings highlight the practical applicability of SciML models in real-world scenarios.
Computer Science > Machine Learning arXiv:2412.13897 (cs) [Submitted on 18 Dec 2024 (v1), last revised 19 Feb 2026 (this version, v2)] Title:Data-Efficient Inference of Neural Fluid Fields via SciML Foundation Model Authors:Yuqiu Liu, Jingxuan Xu, Mauricio Soroco, Yunchao Wei, Wuyang Chen View a PDF of the paper titled Data-Efficient Inference of Neural Fluid Fields via SciML Foundation Model, by Yuqiu Liu and 4 other authors View PDF Abstract:Recent developments in 3D vision have enabled significant progress in inferring neural fluid fields and realistic rendering of fluid dynamics. However, these methods require dense captures of real-world flows, which demand specialized laboratory setups, making the process costly and challenging. Scientific machine learning (SciML) foundation models, pretrained on extensive simulations of partial differential equations (PDEs), encode rich multiphysics knowledge and thus provide promising sources of domain priors for fluid field inference. Nevertheless, the transferability of these foundation models to real-world vision problems remains largely underexplored. In this work, we demonstrate that SciML foundation models can significantly reduce the data requirements for inferring real-world 3D fluid dynamics while improving generalization. Our method leverages the strong forecasting capabilities and meaningful representations learned by SciML foundation models. We introduce a novel collaborative training strategy that equips neural fluid f...
