Reproduce “Learning Mesh-Free Discrete Differential Operators with Self-Supervised Graph Neural Networks” 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
Lucas Gerken Starepravo, Georgios Fourtakas, Steven Lind, Ajay B. Harish, Tianning Tang, Jack R. C. King
NeMDO learns numerical weights from point-neighborhood geometry to estimate derivatives, rather than learning PDE solutions. A graph neural network is trained to approximately satisfy polynomial moment constraints, with reuse across resolutions and equations when stencil size is fixed. On a smooth test, its x-derivative tracks LABFM’s second-order result; weight-generation is 1.5–2 times faster, excluding training on one CPU core.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.