Code and data

Open-source software from the group, on GitHub.

Equivariant neural network that predicts full frequency-dependent dielectric tensors of crystals from structure.

Symmetry- and property-aware reinforcement learning for crystal generation and inverse design.

Graph Transformer for end-to-end prediction of electronic band structures.

Crystal hypergraph convolutional networks with bond, triplet, and motif hyperedges.

Multi-scale framework for coupled chemical, spin, and structural disorder in alloys (GNN + MLIP + Monte Carlo).

Graph neural networks coupled with Monte Carlo for ensemble properties of atomically disordered materials.

Attention-based GNNs with Wang–Landau Monte Carlo for configurational-disorder properties.

Persistent-homology features for graph-neural-network prediction of defect formation energies.

Density functional contrastive learning: building the density-scaling constraint into ML exchange functionals.

High-throughput OpenMX workflow with post-processing for response functions such as permittivity tensors.

Atom–motif dual graph network for motif-centric learning of crystalline materials.