Research map / Label-efficient and robust learning / Recognition and long-tail
InformedXRD: Reproducible Benchmarks and Physics-Informed Evaluation for Powder Diffraction Symmetry Classification
Edward G Friedman, Elizabeth J Baggett, Abhishek Shetty, Derrick Chan-Sew, Vanellsa Acha, Harshitha Dwaracherla, P. A Kienzle, William Ratcliff
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a standardized, reproducible benchmark and physics-informed evaluation protocol for machine-learning powder X-ray diffraction symmetry classification on the RRUFF database.
Problem
Machine learning for PXRD symmetry classification lacks a common evaluation standard: papers use different RRUFF subsets, target taxonomies, preprocessing, and summary metrics, and models are rarely released for rescoring.
Approach
A unified benchmark and physics-informed evaluation for PXRD symmetry classification intended to make results comparable and rescorable across papers (details unavailable beyond the abstract).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| score a 230-way space-group model in 99-way extinction-group space (ablation) | space-group classifier · synthetic 2.0M-pattern dataset | Top-1 accuracy (%): 15.17 -> 18.46 | Improved |
| train directly on extinction groups instead of space groups (ablation) | classifier at full scale · synthetic 2.0M-pattern dataset | Top-1 accuracy (%): +0.86 % (full scale), +2.77 % (one-eighth scale) | Improved |
| evaluate by Top-1 versus mean DAG distance | ViT checkpoints · RRUFF-325 | model ranking: Top-1: 13.54 vs 10.46; mean dist.: 2.51 vs 2.46 | Mixed |
| compare preferred-orientation-only ViT with final calibrated ViT at Top-5 instead of Top-1 | ViT checkpoints · RRUFF-325 | Top-k accuracy (%): Top-5: 43.69 vs 40.92 | Mixed |
| compare CNN10, CNN11 and PatchTST by paired tests | XRDBench architectures · RRUFF-473 and RRUFF-325 | Top-1/3/5 accuracy: smallest raw p=0.124 | No clear effect |
| train the 1D CNN 40 more epochs selecting on synthetic validation loss (ablation) | 1D residual CNN · RRUFF-473 and RRUFF-325 | Top-k accuracy | Hurt |
| compare final calibrated ViT against label-shuffled candidate sets (permutation test) | ViT: final mixed-data calibrated · RRUFF-473 | Top-5 accuracy (%): 48.52 -> 50.74 | Improved |
| compare input-blind oracle prior to final calibrated ViT | ViT: final mixed-data calibrated · RRUFF-473 | Top-1/Top-5 accuracy (%): oracle 12.9/51.4 vs ViT 9.94/50.74 | Hurt |
| reweight strata to common label distribution (entropy reweighting) (ablation) | ViT: final mixed-data calibrated · RRUFF-473 catastrophic stratum | Top-1 accuracy (%): 6.76 -> 10.80 | Improved |
| hierarchy-aware analysis of generic Transformer errors | Transformer (XRDBench) · RRUFF-473 | within-two-hops locality (%): 16.71 vs null 32.54 | Hurt |
Experimental setup
Datasets: RRUFF
Benchmarks: RRUFF
Closest papers by meaning
- All that structure matches does not glitter (NeurIPS 2025 · similarity 0.87)
- WyckoffDiff -- A Generative Diffusion Model for Crystal Symmetry (ICML 2025 · similarity 0.86)
- A Denoising Pre-training Framework for Accelerating Novel Material Discovery (AAAI 2025 · similarity 0.85)
- Space Group Equivariant Crystal Diffusion (NeurIPS 2025 · similarity 0.84)
- Local-Global Associative Frames for Symmetry-Preserving Crystal Structure Modeling (NeurIPS 2025 · similarity 0.84)
- SymmCD: Symmetry-Preserving Crystal Generation with Diffusion Models (ICLR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
