Robust recognition and exploratory analysis of crystal structures via Bayesian deep learning

Abstract

Due to their ability to recognize complex patterns, neural networks can drive a paradigm shift in the analysis of materials science data. Here, we introduce ARISE, a crystal-structure identification method based on Bayesian deep learning. As a major step forward, ARISE is robust to structural noise and can treat more than 100 crystal structures, a number that can be extended on demand. While being trained on ideal structures only, ARISE correctly characterizes strongly perturbed single- and polycrystalline systems, from both synthetic and experimental resources. The probabilistic nature of the Bayesian-deep-learning model allows to obtain principled uncertainty estimates, which are found to be correlated with crystalline order of metallic nanoparticles in electron tomography experiments. Applying unsupervised learning to the internal neural-network representations reveals grain boundaries and (unapparent) structural regions sharing easily interpretable geometrical properties. This work enables the hitherto hindered analysis of noisy atomic structural data from computations or experiments.

arXiv commentPublished in Nature Communications. This document contains main text (4 main figures, 2 main tables) and supplementary information (supplementary methods and notes, as well as 13 supplementary figures and 8 supplementary tables)

References in corpus (22)

• UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction
• 2D materials and van der Waals heterostructures
• Improving neural networks by preventing co-adaptation of feature detectors
• Gaussian Approximation Potentials: the accuracy of quantum mechanics, without the electrons
• Point-Set Registration: Coherent Point Drift
• On representing chemical environments
• Two-dimensional materials from high-throughput computational exfoliation of experimentally known compounds
• Moment Tensor Potentials: a class of systematically improvable interatomic potentials
• Structure identification methods for atomistic simulations of crystalline materials
• The Computational 2D Materials Database: High-Throughput Modeling and Discovery of Atomically Thin Crystals
• Robust Structural Identification via Polyhedral Template Matching
• Comparing molecules and solids across structural and alchemical space
• Bayesian Active Learning for Classification and Preference Learning
• Accelerated Hierarchical Density Clustering
• Deep Learning of Atomically Resolved Scanning Transmission Electron Microscopy Images: Chemical Identification and Tracking Local Transformations
• Deciphering chemical order/disorder and material properties at the single-atom level
• Insightful classification of crystal structures using deep learning
• Materials Cartography: Representing and Mining Material Space Using Structural and Electronic Fingerprints
• Three-dimensional coordinates of individual atoms in materials revealed by electron tomography
• Strained tetragonal states and Bain paths in metals
• Discovering the Building Blocks of Atomic Systems using Machine Learning
• Tuning of crystal structure and magnetic properties by exceptionally large epitaxial strains