Reproduce “Stabilizing Neural Emulation of Spatiotemporal Physical Dynamics through Denoising Koopman Latent Dynamics” 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
Chenxu Wu, Sixing Chen, Junming Hou, Jiahan Huang, Jianqiao Zeng, Peiming Luo, Liang Yan, Jian Wei You
KoopFlow forecasts PDE-governed fields by evolving encoded states with linear latent dynamics, then applying generative correction toward target latent states. The correction is trained on degradation across forecasts from different historical starting points. In temporal extrapolation tests across four task families, it reports the lowest normalized mean squared error on each; for Navier–Stokes steps 31–50, error is 0.05572 versus FNO's 0.1154.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.