methods
PRDNet
What it does
PRDNet (ICLR 2026) combines graph learning with pseudo-particle diffraction to build symmetry-invariant representations for crystal property prediction.
Approach & capabilities
Combines graph embeddings with a learned pseudo-particle diffraction module.
Builds synthetic diffraction representations invariant to crystallographic symmetries.
Evaluates crystal-property prediction on Materials Project, JARVIS-DFT, and MatBench benchmarks.
Read more
We propose PRDNet, a novel architecture that integrates graph embeddings with a learned pseudoparticle diffraction module. It generates synthetic diffraction patterns that are invariant to crystallographic symmetries.
We extensively evaluate PRDNet on multiple large-scale benchmarks, including Materials Project, JARVIS-DFT, and MatBench. Our model achieves state-of-the-art performance across a wide range of crystal property prediction tasks, demonstrating its effectiveness.