机器学习-计算化学Workshop

题 目：   Data-Driven Materials Discovery with the Method SISSO

报告人：   欧阳润海

单 位：   上海大学

时 间：   2019-09-05

地 点：   厦门大学化学化工学院

扫码观看  精彩  报告视频

The Materials-Genome Initiative has fostered high-throughput calculations and experiments, leading to large amount of materials data available in literature and databases. Analyzing those data and finding physical descriptors that describe and predict the target materials properties and functions is crucial for knowledge-guided low-cost and fast material discovery. In this regard, efficient data-driven approaches for descriptor identification are required, and many methods falling under the umbrella name of (big-) data analytics (e.g. data mining, machine learning, compressed sensing, etc.) have being developed and applied to the wealth of Materials Science data. In this talk, Ouyang will introduce the recent data-driven method SISSO, which is based on the theory of compressed sensing, for identifying low-dimensional descriptors (A descriptor is defined a set of features that capture the underlying mechanisms of the target materials property or function; the dimension is the number of features in the descriptor) from huge features spaces. He will review several recent applications of SISSO across Materials Science and Chemistry for materials discovery: materials map for predicting 2D and 3D topological insulators; new tolerance factor for predicting the stability of perovskite; descriptor for predicting the pressure-induced insulatorÆmetal transition of binary crystals; model for predicting the Gibbs free energy of crystalline solids; and the functional form for predicting superconducting critical temperature. In addition, he will also introduce the newly developed technique in SISSO for multi-task learning for finding a common descriptor for multiple materials properties.

Runhai Ouyang joined Shanghai University, China, in 2019 as an associate professor at the Materials Genome Institute. He obtained his PhD degree in physical chemistry in 2014 (advisor: Prof. Wei-Xue Li) from the Dalian Institute of Chemical Physics, Chinese Academy of Sciences. After one year at The University of Sydney and one year at the University of California Riverside for postdoctoral research, he joined the Theory department of Fritz Haber Institute of the Max Planck Society in Berlin and worked with Luca M Ghiringhelli and Matthias Scheffler for the NOMAD project for three years with the research topic data-driven materials science. In collaboration with Ghiringhelli, Scheffler, Curtarolo and Ahmetcik, he created the compressed-sensing method SISSO (sureindependence screening and sparsifying operator) for data-driven materials discovery.

2019年9月3日-6日，由固体表面物理化学国家重点实验室（厦门大学）、福建省理论与计算化学重点实验室和厦门大学化学化工学院主办的“机器学习-计算化学Workshop”在厦门大学化学化工学院举办。   本次Workshop邀请了相关领域的研究者报告领域前沿进展，并设置Hands-on tutorials环节帮助学员们熟悉代码的使用。   此次Workshop的举办增进了不同领域研究者的交流，促进了开源共享的观念传递，希望推动大数据技术在计算化学和材料模拟等领域的应用。