Modelling magnetic material properties with uncertainty-aware neural networks
Abstract
Machine learning is increasingly used to accelerate materials discovery across large compositional and structural design spaces, but limited and heterogeneous data make reliable uncertainty estimation essential. In this work, we investigate uncertainty quantification for permanent magnet modeling across three complementary studies. First, on a public Curie temperature surrogate dataset, we benchmark Gaussian process regression, random forest bagging, and dropout-based Bayesian neural networks and show that calibration, sharpness, and confidence curve diagnostics reveal differences in uncertainty quality that are not visible from point-prediction metrics alone. Second, we transfer this framework to the prediction of intrinsic magnetic properties in Nd2 Fe14 B-based magnets. Third, we extend the same uncertainty-aware perspective to coercivity prediction from microstructural information using a graph neural network. Together, these studies show that uncertainty quantification improves the trustworthiness of magnetic material property predictions and can be transferred from composition-based surrogate models to more complex structure-sensitive learning tasks.
Versions
Related
- Finite Element Modeling of the Active Muscle: Combined Influence of Boundary Conditions and Material Properties
- Design optimization, commissioning, and uncertainty analysis of the quadrupole resonator system at Jefferson Lab for SRF material characterization
- Modern triad for magnetic material science: geometric shape, electronic bands and spin textures
- Crystal Structure Based Graph Neural Network And LLM Multi-Modal Approach for Material Magnetic Ordering Prediction
- Tunability of the structural and magnetic transition in kagome material: PrIr3B2
BibTeX
@misc{wager2026modelling,
title = {{Modelling magnetic material properties with uncertainty-aware neural networks}},
author = {Clemens Wager and Heisam Moustafa and Alexander Kovacs and Qais Ali and Harald Oezelt and Hayate Yamano and Masao Yano and Noritsugu Sakuma and Hyuga Hosoi and Akihito Kinoshita and Tetsuya Shoji and Akira Kato and Thomas Schrefl},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2606.11870},
eprint = {2606.11870},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2606.11870},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Modelling magnetic material properties with uncertainty-aware neural networks
AU - Clemens Wager
AU - Heisam Moustafa
AU - Alexander Kovacs
AU - Qais Ali
AU - Harald Oezelt
AU - Hayate Yamano
AU - Masao Yano
AU - Noritsugu Sakuma
AU - Hyuga Hosoi
AU - Akihito Kinoshita
AU - Tetsuya Shoji
AU - Akira Kato
AU - Thomas Schrefl
PY - 2026
DA - 2026/10/05
PB - arXiv
DO - 10.48550/arxiv.2606.11870
UR - https://arxiv.org/abs/2606.11870
AB - Machine learning is increasingly used to accelerate materials discovery across large compositional and structural design spaces, but limited and heterogeneous data make reliable uncertainty estimation essential. In this work, we investigate uncertainty quantification for permanent magnet modeling across three complementary studies. First, on a public Curie temperature surrogate dataset, we benchmark Gaussian process regression, random forest bagging, and dropout-based Bayesian neural networks and show that calibration, sharpness, and confidence curve diagnostics reveal differences in uncertainty quality that are not visible from point-prediction metrics alone. Second, we transfer this framework to the prediction of intrinsic magnetic properties in Nd2 Fe14 B-based magnets. Third, we extend the same uncertainty-aware perspective to coercivity prediction from microstructural information using a graph neural network. Together, these studies show that uncertainty quantification improves the trustworthiness of magnetic material property predictions and can be transferred from composition-based surrogate models to more complex structure-sensitive learning tasks.
ER -