GNN_Uncertainty: code for the graph neural network study of "Modelling magnetic material properties with uncertainty-aware neural networks"
Code for the graph-neural-network study (Section 5 and Appendix 9) of the article "Modelling magnetic material properties with uncertainty-aware neural networks" (Computational Materials Science 276, 115073, 2027, https://doi.org/10.1016/j.commatsci.2026.115073). A graph neural network predicts the coercivity μ₀H_c of Nd₂Fe₁₄B microstructures and provides aleatoric and epistemic uncertainty estimates (Gaussian negative log-likelihood loss and Monte Carlo dropout); confidence curves assess the quality of the uncertainty estimates. The script reproduce_fig18.py reproduces the results and Fig. 18 of the paper; it is a copy of the corresponding s
Resource information
- Format
- zip
- Original repository
- 23545085
- DOI
- 10.5281/zenodo.23545085
- Credit basis
- Creators in Zenodo record metadata
- Access
- Open files available from the original Zenodo deposit
- Last checked
- Nov 5, 2026
- License
- mit-license
- Last record update
- Nov 5, 2026
About this resource
Code for the graph-neural-network study (Section 5 and Appendix 9) of the article "Modelling magnetic material properties with uncertainty-aware neural networks" (Computational Materials Science 276, 115073, 2027, https://doi.org/10.1016/j.commatsci.2026.115073). A graph neural network predicts the coercivity μ₀H_c of Nd₂Fe₁₄B microstructures and provides aleatoric and epistemic uncertainty estimates (Gaussian negative log-likelihood loss and Monte Carlo dropout); confidence curves assess the quality of the uncertainty estimates. The script reproduce_fig18.py reproduces the results and Fig. 18 of the paper; it is a copy of the corresponding script of https://github.com/wagerc97/magnetic-uncertainty (author: C. Wager), which uses this repository as a git submodule. The training data (1054 synthetic microstructures as graphs, CC BY 4.0) are identical to the data archived with Micromagnetic_GNN (https://doi.org/10.5281/zenodo.23360688); they are included in the GitHub repository but not in this archive, and the README explains how to download them. The exact figures of the paper and the published article (open access, CC BY 4.0) are archived in a separate record: https://doi.org/10.5281/zenodo.23544171.