CMI researchers from Lawrence Livermore National Laboratory conducted the research for this highlight.
Achievement
Identification of crystal phases in atomic structures with 100% accuracy.
Significance and impact
- Identification of structures at high-temperature, e.g., during solidification, is prerequisite of all analysis and long-standing challenge.
- Traditional methods require in-depth physical domain knowledge; Existing machine-learning methods require lots of training labels.
- New denoising method developed by LLNL team achieved perfect accuracy in ordered phases classification.
Details and next steps
- Score-based probabilistic model
- Simple (no data curation), generally applicable
- Paper under review (Nat. Comm.)
- arxiv: 2212.02421
- Algorithm is implemented in popular Avito software
- Application: characterization of molecular dynamics simulation of alloys & magnets.