Reproduce “Unlearning Noise in PINNs: A Selective Pruning Framework for PDE Inverse Problems” 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
Yongsheng Chen, Yong Chen, Wei Guo, Xinghui Zhong
P-PINN modifies a pretrained physics-informed neural network after training on noisy observations. It scores samples using prediction mismatch and PDE residuals, separates suspect from retained data, prunes neurons with differing activations, and fine-tunes under physical constraints. On a synthetic Heat data-assimilation benchmark, mean relative L2 error fell from 0.963 ± 0.219 to 0.0326 ± 0.000156, a 96.6% reduction over three runs, with synthetic Gaussian noise; this result does not establish the same improvement for arbitrary real measurements.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.