Learning Self-Supervised Representations of Powder-Diffraction Patterns

Crystals, Journal Year: 2025, Volume and Issue: 15(5), P. 393 - 393

Published: April 23, 2025

The potential of machine learning (ML) models for predicting crystallographic symmetry information from single-phase powder X-ray diffraction (XRD) patterns is investigated. Given the scarcity large, labeled experimental datasets, we train our using simulated XRD generated databases. A key challenge in developing reliable diffraction-based structure-solution tools lies limited availability training data and presence natural adversarial examples, which hinder model generalization. To address these issues, explore multiple pipelines testing strategies, including evaluations on data. We introduce a contrastive representation approach that significantly outperforms previous supervised terms robustness generalizability, demonstrating improved invariance to effects. These results highlight self-supervised advancing ML-driven analysis.

Language: Английский

Ab initio structure solutions from nanocrystalline powder diffraction data via diffusion models

Nature Materials, Journal Year: 2025, Volume and Issue: unknown

Published: April 28, 2025

Language: Английский