Text-guided flow matching enables sample-efficient crystal structure generation
TFMat uses structured materials text to steer CrystalFlow toward candidate crystal structures with fewer samples.
The paper reports gains on Perov-5, Carbon-24, and MP-20 crystal structure prediction benchmarks. On MP-20, TFMat reaches a 92.04% match rate with 20 candidates. In de novo generation, it better matches element-count and density distributions while keeping coarse property consistency for composition-selected outputs. ArXiv · AI/CL/LG's note
The paper reports gains on Perov-5, Carbon-24, and MP-20 crystal structure prediction benchmarks. On MP-20, TFMat reaches a 92.04% match rate with 20 candidates. In de novo generation, it better matches element-count and density distributions while keeping coarse property consistency for composition-selected outputs. ArXiv · AI/CL/LG's note
score 5