Closed-loop superconducting materials discovery 

Elizabeth A. Pogue, Alexander New, Kyle McElroy, Nam Q. Le, Michael J. Pekala, Ian McCue, Eddie Gienger, Janna Domenico, Elizabeth Hedrick, Tyrel M. McQueen, Brandon Wilfong, Christine D. Piatko, Christopher R. Ratto, Andrew Lennon, Christine Chung, Timothy Montalbano, Gregory Bassen & Christopher D. Stiles 

npj Computational Materials 9: 181 (2023)

doi.org/10.1038/s41524-023-01131-3

Published online: 05 October 2023

编辑概述

闭环搜索超导材料：机器学习+实验反馈

新材料的发现推动了工业创新，但由于“尤里卡！”时刻的罕见，材料发现的步伐往往十分缓慢。这些时刻通常是与实验工作原始目标间接相关的“偶然发现”。长期以来，统计方法被一直用于更好地理解和预测超导性，最近的例子是通过使用黑箱机器学习（ML）的方法。尽管该方法产生了众多预测，但大多研究并没有发现以前未报道的超导体系列，这不仅仅是因为外推未知系列的巨大困难，也因为预测的材料具有一些妨碍超导的化学属性—比如具有高度局域的化学键（如包含多原子阴离子的材料），或极端的亚稳性阻碍了其合成的可能。此外，前期的研究将材料和材料属性数据库视为固定的而非不断发展的系统，这也限制了ML模型在稀疏数据上学习的能力。在本文中，来自美国约翰霍普金斯大学应用物理实验室的Elizabeth A. Pogue等人，结合ML技术与材料科学和物理专业知识，提出了一种“闭环”的机器学习方法，用来加速材料的主动发现。作者展示了如何使ML模型在不同的材料空间中泛化，以识别与训练语料库中不同的超导体。通过在ML属性预测和实验验证之间交替进行，该方法能够系统地提高在现有的稀疏表示材料数据库中ML属性预测准确性。至关重要的是，这种方法既添加了负面数据（错误预测为超导体的材料），也添加了正面数据（正确预测的材料）到ML训练中，使得ML模型对材料空间整体表示的迭代细化成为可能。通过对ML生成的超导性预测结果进行实验验证，并将这些数据反馈到ML模型中进行精炼，作者证明了超导体发现的成功率可以翻一番以上。该工作证明了实验反馈在ML驱动发现中的关键作用，并为如何加速材料进步提供了一个蓝图。

Editorial Summary

Closed-loop superconductors discovery: Machine learning + experimental feedback

The discovery of novel materials drives industrial innovation, although the pace of discovery tends to be slow due to the infrequency of “Eureka!” moments. These moments are typically tangential to the original target of the experimental work: “accidental discoveries”. Statistical approaches have long aimed to better understand and predict superconductivity, most recently through the use of black-box ML methods. Although resulting in numerous predictions, these studies have not yielded previously unreported families of superconductors, likely not only because of difficulties in extrapolating beyond known families, but also because the predicted materials have chemical attributes that make them unlikely to be superconducting—whether it is highly localized chemical bonding, e.g., those containing polyatomic anions, or an extreme metastability that precludes synthesizability. Further, existing works have treated materials and databases of material properties as fixed snapshots rather than evolving systems, which limits the ability of ML models to learn over sparse data. In this work, Elizabeth A. Pogue et al. from the Research and Exploratory Development Department, Johns Hopkins University Applied Physics Laboratory, combined ML techniques with materials science and physics expertise to “close the loop” of materials discovery, accelerating the intentional discovery of superconducting compounds. The authors demonstrated how to make ML models generalize across diverse materials spaces, to identify superconductors that are dissimilar to ones in the training corpus. By alternating between ML property prediction and experimental verification, this method can systematically improve the fidelity of ML property prediction in regimes sparsely represented by existing materials databases. Crucially, this adds both negative data (materials incorrectly predicted to be superconductors) and positive data (materials correctly predicted) to ML training, enabling the ML model’s overall representation of the space of materials to be iteratively refined. By experimentally validating the results of the ML-generated superconductivity predictions and feeding those data back into the ML model to refine, the authors demonstrated that the success rates for superconductor discovery can be more than doubled. This work demonstrates the critical role experimental feedback provides in ML-driven discovery and provides a blueprint for how to accelerate materials progress.