Collaborative Research: Accelerating the Discovery of Electronic Materials through Human-Computer Active Search

协作研究：通过人机主动搜索加速电子材料的发现

• 批准号: 1940224
• 负责人: Roman Garnett
• 金额: $ 30.59万
• 依托单位: Washington University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-10-01 至 2023-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1940224&HistoricalAwards=false
• 关键词: Collaborative; Research; Accelerating; Discovery; Electronic

The overarching goal of this project is to accelerate the discovery of materials with tailored electronic properties through human-computer active search. These efforts will lay the groundwork for accelerating materials discovery, and advance the capability to control electronic properties in materials with the potential for profound societal impact. The thermoelectric and photocatalytic materials predicted, synthesized, and characterized in this research can realize societal advances in the space of energy and solar fuels. High-efficiency thermoelectric materials can revolutionize how heat sources are transformed into electrical power by eliminating the traditional intermediate mechanical energy conversions. Earth-abundant light-responsive catalysts are emerging as an alternative to costly, rare metal catalysts to store solar energy as portable liquid fuels, like ethanol. These green reactions are enabling low-cost, carbon-neutral fuels. The team brings together expertise in materials science, chemistry, machine learning, visualization, metadata, and knowledge frameworks to develop multi-fidelity, expert-guided active search strategies within materials science and chemistry. Resonances among the team's existing outreach programs will broaden inclusion of students from underrepresented groups and be moderated via the Alliance for Diversity in Science and Engineering. The work will provide cross-disciplinary training to graduate students and postdocs in all aspects of material informatics, including participating in and leading team efforts, co-mentorship of Ph.D. and postdoctoral researchers, inclusive symposia at national conferences, and a summer workshop focused on the intersection of visualization, machine learning, ontological engineering and materials science. Through enabling the acceleration of the discovery of new materials, this project supports the goals of the Materials Genome Initiative. An interdisciplinary team will create a search framework for scientific discovery that leverages recent advances in material databases, machine learning, visualization, human-machine interaction, and knowledge structures. To broadly assess the efficacy of this approach, the search effort will span the electronic behavior of both molecules and crystalline materials: (i) new organic photocatalysts for solar fuels production and (ii) new thermoelectric materials for electricity generation. Central to this effort is the engagement of domain experts and associated feedback in a human-in-the-loop active search process. Dynamic visualizations will enable the user to (i) understand the underlying reasons why the materials are being suggested and (ii) provide a user steering capability to identify and annotate specific aspects of the explored search space. Domain-expert annotations and feedback will be parsed against a suite of ontologies, further aiding the search process by providing relational insight between features. New molecules and materials will be explored through a combination of first principles calculations and high-throughput, automated experimentation; these results will be incorporated into a continually growing open-access database. Efficiently integrating and directing evolving data-streams from experiment, computation, and human steering during the search will be achieved with a multi-fidelity active search policy. Through enabling the acceleration of the discovery of new materials, this project supports the goals of the Materials Genome Initiative. This project is part of the National Science Foundation's Harnessing the Data Revolution (HDR) Big Idea activity, and is jointly supported by HDR and the Division of Materials Research within the NSF Directorate of Mathematical and Physical Sciences.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

该项目的首要目标是通过人机主动搜索来加速发现具有定制电子性质的材料。这些努力将为加速材料发现奠定基础，并提高控制材料电子性质的能力，可能产生深远的社会影响。本研究预测、合成和表征的热电和光催化材料可以实现能源和太阳能燃料领域的社会进步。高效热电材料可以通过消除传统的中间机械能转换来彻底改变热源转换为电能的方式。富含地球的光响应型催化剂正在出现，作为昂贵的稀有金属催化剂的替代品，将太阳能存储为便携式液体燃料，如乙醇。这些绿色反应使低成本、碳中性燃料成为可能。该团队汇集了材料科学、化学、机器学习、可视化、元数据和知识框架方面的专业知识，以开发材料科学和化学中的多保真、专家指导的主动搜索策略。该团队现有外展计划的共鸣将扩大对来自代表性不足群体的学生的纳入，并通过科学与工程多样性联盟进行协调。这项工作将在材料信息学的所有方面为研究生和博士后提供跨学科培训，包括参与和领导团队努力，共同指导博士和博士后研究人员，在国家会议上举行包容性研讨会，以及夏季研讨会，重点是可视化、机器学习、本体论工程和材料科学的交叉。通过加速新材料的发现，该项目支持了材料基因组计划的目标。一个跨学科团队将创建一个科学发现搜索框架，利用材料数据库、机器学习、可视化、人机交互和知识结构方面的最新进展。为了广泛评估这种方法的有效性，搜索工作将跨越分子和晶体材料的电子行为：(I)用于生产太阳能燃料的新型有机光催化剂和(Ii)用于发电的新型热电材料。这一努力的核心是让领域专家和相关反馈参与到人在环中的主动搜索过程中。动态可视化将使用户能够(I)理解材料被建议的潜在原因，以及(Ii)向用户提供指导能力，以识别和注释所探索的搜索空间的特定方面。领域专家注释和反馈将根据一套本体进行解析，通过提供功能之间的关系洞察来进一步帮助搜索过程。新的分子和材料将通过第一原理计算和高通量、自动化实验相结合来探索；这些结果将被合并到一个不断增长的开放访问数据库中。在搜索过程中，将通过多保真的主动搜索策略来实现有效地整合和引导来自实验、计算和人工指导的不断演变的数据流。通过加速新材料的发现，该项目支持了材料基因组计划的目标。该项目是国家科学基金会利用数据革命(HDR)大创意活动的一部分，由HDR和NSF数学和物理科学局内的材料研究部联合支持。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

BINOCULARS for efficient, nonmyopic sequential experimental design

用于高效、非近视顺序实验设计的双筒望远镜

• 发表时间: 2020
• 期刊: Proceedings of the 37th International Conference on Machine Learning
• 作者: Jiang, Shali;Chai, Henry;González, Javier;Garnett, Roman
• 通讯作者: Garnett, Roman

Efficient Nonmyopic Bayesian Optimization via One-Shot Multi-Step Trees

通过一次性多步树进行高效非近视贝叶斯优化

• 发表时间: 2020
• 期刊: Advances in neural information processing systems
• 作者: Jiang, Shali;Jiang, Daniel;Balandat, Maximilian;Karrer, Brian;Gardner, Jacob;Garnett, Roman
• 通讯作者: Garnett, Roman

Simulating high-entropy alloys at finite temperatures: An uncertainty-based approach

在有限温度下模拟高熵合金：基于不确定性的方法

• DOI: 10.1103/physrevmaterials.7.063801
• 发表时间: 2023
• 期刊: Physical Review Materials
• 影响因子: 3.4
• 作者: Novick, Andrew;Nguyen, Quan;Garnett, Roman;Toberer, Eric;Stevanović, Vladan
• 通讯作者: Stevanović, Vladan

Efficient Discovery of Visible Light-Activated Azoarene Photoswitches with Long Half-Lives Using Active Search

使用主动搜索有效发现可见光激活的长半衰期偶氮芳烃光电开关

• DOI: 10.1021/acs.jcim.1c00954
• 发表时间: 2021
• 期刊: Journal of Chemical Information and Modeling
• 影响因子: 5.6
• 作者: Mukadum, Fatemah;Nguyen, Quan;Adrion, Daniel M.;Appleby, Gabriel;Chen, Rui;Dang, Haley;Chang, Remco;Garnett, Roman;Lopez, Steven A.
• 通讯作者: Lopez, Steven A.

Guided Data Discovery in Interactive Visualizations via Active Search

• DOI: 10.1109/vis54862.2022.00023
• 发表时间: 2020-10
• 期刊: 2022 IEEE Visualization and Visual Analytics (VIS)
• 作者: S. Monadjemi;Sunwoo Ha;Quan Nguyen;Henry Chai;R. Garnett;Alvitta Ottley