Symmetry-Based Parameter Efficient Low-Rank Fine-Tuning for High-Throughput Calculation of Lattice Thermal Conductivity
The study introduces Symmetry-Based Few-Shot Fine-Tuning (SBFS-FT) of MatterSim 1M to accurately predict lattice thermal conductivity (LTC) for large datasets of materials sharing a common crystal symmetry. This method fine-tunes the universal machine learning interatomic potential on small subsets, achieving accuracy comparable to individual fine-tuning while drastically reducing computational time by a factor of 5–7.