AI RESEARCH
Tensor Channel Equivariant Graph Neural Networks for Molecular Polarizability Prediction
arXiv CS.LG
•
ArXi:2605.16891v1 Announce Type: new On optimized QM7-X geometries, the proposed model achieves lower full-tensor and anisotropic error than both a PaiNN-style readout baseline and a dielectric MACE baseline under matched