PRISMat: Policy-Driven, Permutation-Invariant Autoregressive Material Generation

In three linesPRISMat is a permutation-invariant autoregressive model for crystal material generation. Lighter and faster than LLMs, it reduces prediction error for cleavage energy and work function by 4× (MAE 0.188 eV/Ų and 2.79 eV). Applicable to high-throughput materials discovery.

Summary generated by Claude — human-verified