Reproduce “HyPINO: Multi-Physics Neural Operators via HyperPINNs and the Method of Manufactured Solutions” 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

Rafael Bischof, Michal Piovarči, Michael A. Kraus, Siddhartha Mishra, Bernd Bickel

HyPINO generates PINNs from PDE specifications using a hypernetwork, enabling zero-shot solution predictions across changing operators, sources, domains, and boundary conditions. It trains on manufactured solutions and unlabeled examples using physics-based residuals, and can add residual-driven corrections at inference. On the 1D heat benchmark, ten rounds lower MSE to 8.0e-5 from 2.3e-2 unrefined; results cover seven linear, two-dimensional benchmarks.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.