A perspective on Bayesian methods applied to materials discovery and design
A perspective on Bayesian methods applied to materials discovery and design
复制标题
DOI：
10.1557/s43579-022-00288-0
复制
发表时间：
2022-10-26
影响因子：
1.9
通讯作者：
Allaire, Douglas
中科院分区：
文献类型：
--
作者：
Arroyave, Raymundo;Khatamsaz, Danial;Allaire, Douglas
二十多年来，人们对开发加速发现和设计新型材料的框架越来越感兴趣，这些材料可以实现有前景的变革性技术。集成计算材料工程（ICME）计划要求集成计算工具，以沿着工艺-结构-性能-性能链建立联系。材料基因组计划呼吁将实验和计算整合到数据科学框架内，作为加速材料开发周期的策略。虽然这些框架和范例具有相当大的影响力，但传统的 ICME 或基于数据科学的方法往往存在一些局限性，主要是当查询材料空间成本高昂且可用信息很少时。贝叶斯方法由于其效率提升而更适合这种情况。为此，材料发现问题被定义为贝叶斯优化（BO）。介绍了应用 BO 解决材料发现问题的不同示例。讨论的方法/例子包括模型不确定性下的BO、多信息源BO、多目标和多约束的BO以及批量BO。贝叶斯材料发现是一个有前途的研究领域，随着人们对自主材料发现平台的更多关注，该领域的影响力可能会变得更大。因此，讨论了此类方法的潜在发展，以提高现有平台在材料发现方面的能力。最终目标是为自主材料发现铺平道路。
For more than two decades, there has been increasing interest in developing frameworks for the accelerated discovery and design of novel materials that could enable promising and transformative technologies. The Integrated Computational Materials Engineering (ICME) program called for integrating computational tools to establish linkages along process-structure-property-performance chains. The Materials Genome Initiative called for integrating experiments and computations within data science frameworks as a strategy to accelerate the materials development cycle. While these frameworks and paradigms have been quite influential, traditional ICME or data science-based approaches tend to have some limitations, mainly when querying the materials space is costly and very little information is available. Bayesian methods are more suitable in this context due to their efficiency gains. To this end, the materials discovery problem is framed as a Bayesian optimization (BO). Different examples in which BO has been applied to solve materials discovery problems are presented. The methods/examples discussed include BO under model uncertainty, multi-information source BO, multi-objective and multi-constraint BO, and batch BO. Bayesian Materials Discovery is a promising area of research that is likely to become more influential as more attention is put on autonomous materials discovery platforms. Therefore, a discussion is provided on the potential development of such methods to increase the ability of existing platforms in materials discovery. The ultimate goal is to pave the way to autonomous materials discovery.