Machine learning is becoming a valuable tool for scientific discovery.Particularly attractive is the application of machine learning methods to the field of materials development,which enables innovations by discoveri...Machine learning is becoming a valuable tool for scientific discovery.Particularly attractive is the application of machine learning methods to the field of materials development,which enables innovations by discovering new and better functional materials.To apply machine learning to actual materials development,close collaboration between scientists and machine learning tools is necessary.However,such collaboration has been so far impeded by the black box nature of many machine learning algorithms.It is often difficult for scientists to interpret the data-driven models from the viewpoint of material science and physics.Here,we demonstrate the development of spin-driven thermoelectric materials with anomalous Nernst effect by using an interpretable machine learning method called factorized asymptotic Bayesian inference hierarchical mixture of experts(FAB/HMEs).Based on prior knowledge of material science and physics,we were able to extract from the interpretable machine learning some surprising correlations and new knowledge about spin-driven thermoelectric materials.Guided by this,we carried out an actual material synthesis that led to the identification of a novel spin-driven thermoelectric material.This material shows the largest thermopower to date.展开更多
Machine learning techniques have proven invaluable to manage the ever growing volume of materials research data produced as developments continue in high-throughput materials simulation,fabrication,and characterizatio...Machine learning techniques have proven invaluable to manage the ever growing volume of materials research data produced as developments continue in high-throughput materials simulation,fabrication,and characterization.In particular,machine learning techniques have been demonstrated for their utility in rapidly and automatically identifying potential composition-phase maps from structural data characterization of composition spread libraries,enabling rapid materials fabrication-structure-property analysis and functional materials discovery.A key issue in development of an automated phase-diagram determination method is the choice of dissimilarity measure,or kernel function.The desired measure reduces the impact of confounding structural data issues on analysis performance.The issues include peak height changes and peak shifting due to lattice constant change as a function of composition.In this work,we investigate the choice of dissimilarity measure in X-ray diffraction-based structure analysis and the choice of measure’s performance impact on automatic composition-phase map determination.Nine dissimilarity measures are investigated for their impact in analyzing X-ray diffraction patterns for a Fe-Co-Ni ternary alloy composition spread.The cosine,Pearson correlation coefficient,and Jensen-Shannon divergence measures are shown to provide the best performance in the presence of peak height change and peak shifting(due to lattice constant change)when the magnitude of peak shifting is unknown.With prior knowledge of the maximum peak shifting,dynamic time warping in a normalized constrained mode provides the best performance.This work also serves to demonstrate a strategy for rapid analysis of a large number of X-ray diffraction patterns in general beyond data from combinatorial libraries.展开更多
基金This work was supported by JST-PRESTO“Advanced Materials Informatics through Comprehensive Integration among Theoretical,Experimental,Computational and Data-Centric Sciences”(Grant No.JPMJPR17N4)JST-ERATO“Spin Quantum Rectification Project”(Grant No.JPMJER1402)I.T.is supported in part by C-SPIN,one of six centers of STARnet,a Semiconductor Research Corporation program,sponsored by MARCO and DARPA.
文摘Machine learning is becoming a valuable tool for scientific discovery.Particularly attractive is the application of machine learning methods to the field of materials development,which enables innovations by discovering new and better functional materials.To apply machine learning to actual materials development,close collaboration between scientists and machine learning tools is necessary.However,such collaboration has been so far impeded by the black box nature of many machine learning algorithms.It is often difficult for scientists to interpret the data-driven models from the viewpoint of material science and physics.Here,we demonstrate the development of spin-driven thermoelectric materials with anomalous Nernst effect by using an interpretable machine learning method called factorized asymptotic Bayesian inference hierarchical mixture of experts(FAB/HMEs).Based on prior knowledge of material science and physics,we were able to extract from the interpretable machine learning some surprising correlations and new knowledge about spin-driven thermoelectric materials.Guided by this,we carried out an actual material synthesis that led to the identification of a novel spin-driven thermoelectric material.This material shows the largest thermopower to date.
基金supported by NIST and NEC and partially supported by ONR N000141512222.
文摘Machine learning techniques have proven invaluable to manage the ever growing volume of materials research data produced as developments continue in high-throughput materials simulation,fabrication,and characterization.In particular,machine learning techniques have been demonstrated for their utility in rapidly and automatically identifying potential composition-phase maps from structural data characterization of composition spread libraries,enabling rapid materials fabrication-structure-property analysis and functional materials discovery.A key issue in development of an automated phase-diagram determination method is the choice of dissimilarity measure,or kernel function.The desired measure reduces the impact of confounding structural data issues on analysis performance.The issues include peak height changes and peak shifting due to lattice constant change as a function of composition.In this work,we investigate the choice of dissimilarity measure in X-ray diffraction-based structure analysis and the choice of measure’s performance impact on automatic composition-phase map determination.Nine dissimilarity measures are investigated for their impact in analyzing X-ray diffraction patterns for a Fe-Co-Ni ternary alloy composition spread.The cosine,Pearson correlation coefficient,and Jensen-Shannon divergence measures are shown to provide the best performance in the presence of peak height change and peak shifting(due to lattice constant change)when the magnitude of peak shifting is unknown.With prior knowledge of the maximum peak shifting,dynamic time warping in a normalized constrained mode provides the best performance.This work also serves to demonstrate a strategy for rapid analysis of a large number of X-ray diffraction patterns in general beyond data from combinatorial libraries.
