label-free-fe-surrogates: trained networks

Trained networks of the paper "The norm, not the labels: training and evaluating finite-element surrogates in the energy norm" by Ruifeng Cao and Xidan Song. Of the networks trained in the runs that the paper reports, the archive holds the 37 whose SHA-256 hashes the records of the code's version statics-paper-1.0 and its provenance note give, as PyTorch state dicts. All were trained on 1,024 instances; they come from the paper's two- and three-dimensional comparisons, its three-dimensional energy-norm run, its latent-regularisation and bottleneck runs, and its July run. The paper's other networks, among them some trained on 1,024 instances, are not archived. README_trained-networks.md maps each file to its run and configuration and shows how to load it; networks_sha256.txt lists the SHA-256 hash of each state. The code, configurations and records are archived at https://doi.org/10.5281/zenodo.23815187. Licensed under CC BY-NC 4.0; licences for commercial use are available from Ruifeng Cao ([e-mail]).