Self-supervised learning for domains where labels are expensive to obtain. Crystal structures are arbitrary graphs—infinite point clouds with no canonical ordering—making reconstruction-based approaches intractable. Deep InfoMax provides an alternative by maximizing mutual information between representations without reconstruction.
@article{Moran2025DeepInfoMax,author={Moran, M. and Gaultois, M. W. and Gusev, V. V. and Antypov, D. and Rosseinsky, M. J.},title={Establishing Deep InfoMax as an effective self-supervised learning methodology in materials informatics},journal={Digital Discovery},year={2025},volume={4},pages={790},doi={10.1039/D4DD00202D},url={https://doi.org/10.1039/D4DD00202D},}
2024
PhD Thesis
An information oriented approach to materials informatics
This thesis develops novel transformer architectures and representation learning frameworks for materials science. The work addresses fundamental challenges in applying deep learning to crystal structures: their infinite periodic symmetry, heterogeneous graph topology, and the need to capture long-range interactions while respecting symmetry constraints. Contributions include Site-Net (global self-attention over supercells) and establishing Deep InfoMax as an effective self-supervised learning methodology for domains where reconstruction is intractable.
@phdthesis{Moran2024Thesis,author={Moran, Marin},title={An information oriented approach to materials informatics},school={University of Liverpool},year={2024},url={https://livrepository.liverpool.ac.uk/3193136/},}
Novel architecture combining transformers with crystallographic domain knowledge. Crystal structures have infinite periodic symmetry—standard graph methods miss long-range interactions. Site-Net uses real-space supercells with global self-attention to capture these interactions while respecting symmetry constraints.
@article{Moran2023SiteNet,author={Moran, M. and Gaultois, M. W. and Gusev, V. V. and Antypov, D. and Rosseinsky, M. J.},title={Site-Net: using global self-attention and real-space supercells to capture long-range interactions in crystal structures},journal={Digital Discovery},year={2023},volume={2},pages={1297--1310},doi={10.1039/D3DD00005B},url={https://doi.org/10.1039/D3DD00005B},}
Database of experimental lithium solid electrolyte conductivities combined with machine learning analysis. Comprehensive collection of measured conductivity data used to train ML models for predicting ionic conductivity in solid-state battery materials.
@article{Hargreaves2023Database,author={Hargreaves, C. J. and Gaultois, M. W. and Daniels, L. M. and Watts, E. J. and Kurlin, V. A. and Moran, M. and Dang, Y. and Morris, R. and Morscher, A. and Thompson, K. and Wright, M. A. and Prasad, B. E. and Blanc, F. and Collins, C. M. and Crawford, C. A. and Duff, B. B. and Evans, J. and Gamon, J. and Han, G. and Leube, B. T. and Niu, H. and Perez, A. J. and Robinson, A. and Rogan, O. and Sharp, P. M. and Shoko, E. and Sonni, M. and Thomas, W. J. and Vasylenko, A. and Wang, L. and Rosseinsky, M. J. and Dyer, M. S.},title={A database of experimentally measured lithium solid electrolyte conductivities evaluated with machine learning},journal={npj Computational Materials},year={2023},volume={9},number={1},pages={9},doi={10.1038/s41524-022-00951-z},url={https://doi.org/10.1038/s41524-022-00951-z},}
Cloud-based platform democratizing access to materials informatics tools. Combines supervised learning models, crystal structure analysis, and property prediction in a user-friendly web interface.
@article{Durdy2023Liverpool,author={Durdy, S. and Hargreaves, C. J. and Dennison, M. and Wagg, B. and Moran, M. and Newnham, J. A. and Gaultois, M. W. and Rosseinsky, M. J. and Dyer, M. S.},title={The Liverpool materials discovery server: a suite of computational tools for the collaborative discovery of materials},journal={Digital Discovery},year={2023},volume={2},number={5},pages={1601--1611},doi={10.1039/D3DD00093A},url={https://doi.org/10.1039/D3DD00093A},}
Experimental validation of ML-guided materials discovery. Combined crystal structure prediction with machine learning to identify novel low thermal conductivity oxides for thermoelectric applications. Synthesized and characterized predicted compound, confirming computational predictions.
@article{Collins2021LowThermal,author={Collins, C. M. and Daniels, L. M. and Gibson, Q. and Gaultois, M. W. and Moran, M. and Feetham, R. and Pitcher, M. J. and Dyer, M. S. and Delacotte, C. and Zanella, M. and Murray, C. A. and Glodan, G. and P{\'e}rez, O. and Pelloquin, D. and Manning, T. D. and Alaria, J. and Darling, G. R. and Claridge, J. B. and Rosseinsky, M. J.},title={Discovery of a Low Thermal Conductivity Oxide Guided by Probe Structure Prediction and Machine Learning},journal={Angewandte Chemie International Edition},year={2021},volume={60},number={30},pages={16457--16465},doi={10.1002/anie.202102073},url={https://doi.org/10.1002/anie.202102073},}
Note: Some earlier publications may appear under the name Michael J. Moran.