Reproduce “Generative AI for fast and accurate statistical computation of fluids” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Roberto Molinaro, Samuel Lanthaler, Bogdan Raonić, Tobias Rohner, Victor Armegioiu, Stephan Simonis, Dana Grund, Yannick Ramic and 4 more
Researchers at ETH Zurich and Google Research developed GenCFD, a conditional score-based diffusion model for the fast and accurate statistical computation of 3D turbulent fluid flows. The model generates high-fidelity flow field samples, accurately approximates statistical quantities like mean, standard deviation, and energy spectra, and achieves a speedup of up to five orders of magnitude compared to traditional computational fluid dynamics methods.
Key insights
- Generative AI, specifically score-based diffusion models, can accurately learn and generate diverse samples from complex, multi-modal probability distributions characteristic of chaotic physical systems, unlike deterministic ML models.
- The "regression to the mean" observed in deterministic ML models for chaotic systems is attributed to their Lipschitz continuity, which makes them insensitive to the small input perturbations that drive turbulent behavior.
- Diffusion models can implicitly learn to predict uncertainty and effectively recover the correct conditional probability distribution by leveraging the unstable nature of chaotic operators, even with an insensitive denoiser.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.