Learning properties of ordered and disordered materials from multi-fidelity data

Chen, Chi; Zuo, Yang; Ye, Weike; Li, Xiangguo; Ong, Shyue Ping

doi:10.1038/s43588-020-00002-x

Cited by 115 publications

(100 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Similarly, the active data mining effort of RES 3 T (https://www.hzdr.de/db/RES3T.login) contains 3172 references and includes reactions between 147 minerals and 148 ligands and a total of 7062 reaction constants that span across all known surface complexation models. ML-empowered natural language processing and automated data discovery/extraction from diverse literatureincreasingly used in chemical and material sciences 4,5,[7][8][9][10][11] can similarly transform data utilization in the Earth sciences to improve ESP.…”

Section: Narrativementioning

confidence: 99%

Process Discovery through Assimilation of Complex Biogeochemical Datasets

Zavarin

2021

View full text Add to dashboard Cite

This white paper addresses two of three focal areas identified in the white paper call: 1) Biogeochemical data acquisition and assimilation enabled by machine learning and 3) Insight gleaned from complex data using AI. We focus on AI application to complex biogeochemistry (BGC) data (e.g. laboratory experimental data, field manipulation data, literature data), which is an untapped source of information for improving Earth System Predictability (ESP). Science ChallengeLaboratory experiments and field manipulation experiments are critical to interrogate the impact of hydrological and climate perturbations on BGC processes in a controlled environment. Hydrologicallydriven BGC processes are a key aspect of ESP, particularly at dynamic interfaces (e.g. terrestrial-aquatic interfaces, hot spots/hot moments), because BGC processes govern the cycling of nutrients, metals, and organic matter. However, BGC experiments yield a complex array of data types (spatiotemporal observational and laboratory measurements, microscopy image data, spectroscopy data, etc.), which hampers assimilation and analysis, as well as the application of machine learning (ML). Ensuring that these data are findable, accessible, interoperable, and reusable (FAIR) is of paramount importance. AI/ML methods are poised to transform the way we incorporate complex BGC laboratory and field manipulation experimental data into earth and environmental systems models, quantify data uncertainty, design future experiments, and develop new models with unprecedented fidelity and resolution. 1

show abstract

Section: Narrativementioning

confidence: 99%

Process Discovery through Assimilation of Complex Biogeochemical Datasets

Zavarin

2021

View full text Add to dashboard Cite

show abstract

“…22 The CGCNN and its variants exhibit similar accuracy in predicting formation enthalpy, with mean absolute error (MAE) of 0.03-0.04 eV/atom. [27][28][29][30] For structure and stability predictions, it is imperative that the model is able to (1) predict the total energy of both GS and higher-energy structures with similar accuracy and (2) distinguish energetically favorable (low-energy) structures from those with higher energy. The CGCNN models discussed above are trained primarily on ICSD structures that are GS or near-GS structures.…”

Section: Introductionmentioning

confidence: 99%

Predicting energy and stability of known and hypothetical crystals using graph neural network

Pandey

Stevanović

et al. 2021

Patterns

View full text Add to dashboard Cite

show abstract

“…[1][2][3][4][5] The challenges of AI for materials science are extensive, making acceleration of materials discovery a formidable task that requires advancing the frontier of AI. [6][7][8][9][10][11][12][13][14] Materials discovery embodies the convergence of limited data, data dispersed over multiple domains, and multi-property prediction, motivating commensurate integration of AI methods to collectively address these challenges, as demonstrated by the Hierarchical Correlation Learning for Multi-property Prediction (H-CLMP, pronounced "H-Clamp") framework introduced herein.…”

Section: Introductionmentioning

confidence: 99%

Materials Representation and Transfer Learning for Multi-Property Prediction

Kong¹,

Guevarra²,

Gomes³

et al. 2021

Preprint

View full text Add to dashboard Cite

The adoption of machine learning in materials science has rapidly transformed materials property prediction. Hurdles limiting full capitalization of recent advancements in machine learning include the limited development of methods to learn the underlying interactions of multiple elements, as well as the relationships among multiple properties, to facilitate property prediction in new composition spaces. To address these issues, we introduce the Hierarchical Correlation Learning for Multi-property Prediction (H-CLMP) framework that seamlessly integrates (i) prediction using only a material’s composition, (ii) learning and exploitation of correlations among target properties in multitarget regression, and (iii) leveraging training data from tangential domains via generative transfer learning. The model is demonstrated for prediction of spectral optical absorption of complex metal oxides spanning 69 3-cation metal oxide composition spaces. H-CLMP accurately predicts non-linear composition-property relationships in composition spaces for which no training data is available, which broadens the purview of machine learning to the discovery of materials with exceptional properties. This achievement results from the principled integration of latent embedding learning, property correlation learning, generative transfer learning, and attention models. The best performance is obtained using H-CLMP with Transfer learning (H-CLMP(T)) wherein a generative adversarial network is trained on computational density of states data and deployed in the target domain to augment prediction of optical absorption from composition. H-CLMP(T) aggregates multiple knowledge sources with a framework that is well-suited for multi-target regression across the physical sciences.

show abstract

Learning properties of ordered and disordered materials from multi-fidelity data

Cited by 115 publications

References 46 publications

Process Discovery through Assimilation of Complex Biogeochemical Datasets

Process Discovery through Assimilation of Complex Biogeochemical Datasets

Predicting energy and stability of known and hypothetical crystals using graph neural network

Materials Representation and Transfer Learning for Multi-Property Prediction

Contact Info

Product

Resources

About