科学研究的第五范式——以智能驱动的材料设计为例  

Fifth Paradigm in Science:A Case Study of an Intelligence-Driven Material Design

作　　者：Can Leng Zhuo Tang Yi-Ge Zhou Zean Tian Wei-Qing Huang Jie Liu Keqin Li Kenli Li 

机构地区：[1]Science and Technology on Parallel and Distributed Processing Laboratory,National University of Defense Technology,Changsha 410073,China [2]Laboratory of Software Engineering for Complex Systems,National University of Defense Technology,Changsha 410073,China [3]National Supercomputing Center in Changsha,Changsha 410082,China [4]College of Computer Science and Electronic Engineering,Hunan University,Changsha 410082,China [5]Institute of Chemical Biology and Nanomedicine,State Key Laboratory of Chemo/Biosensing and Chemometrics,College of Chemistry and Chemical Engineering,Hunan University,Changsha 410082,China [6]Department of Applied Physics,School of Physics and Electronics,Hunan University,Changsha 410082,China [7]Department of Computer Science,State University of New York,New Paltz,NY 12561,USA

出　　处：《Engineering》2023年第5期126-137,I0003,I0004,共14页工程（英文）

基　　金：supported by the National Key Research and Development Program of China(2021ZD40303);the National Natural Science Foundation of China(62225205 and 92055213);Natural Science Foundation of Hunan Province of China(2021JJ10023);Shenzhen Basic Research Project(Natural Science Foundation)(JCYJ20210324140002006)。

摘　　要：科学正在进入一个新时代——第五范式——它被认为是知识整合到不同领域的主要特征,是基于无所不在的机器学习系统的计算社区中智能驱动的工作。在此,我们通过在天河一号超级计算机系统上构建的催化材料专门设计的典型平台案例,生动地阐明了第五范式的本质,旨在促进第五范式在其他领域的培养。第五范式平台主要包括模型自动构建(原始数据提取)、指纹自动构建(神经网络特征选择)以及跨学科知识串联的重复迭代(“火山图”)。与分解一起进行的是对迭代中实现的体系结构的性能评估。通过讨论,第五范式的智能驱动平台可以极大地简化和改进研究中极其繁琐和具有挑战性的工作,并通过补偿机器学习中缺少样本和替代一些由于计算资源不足而导致的数值计算来实现数值计算与机器学习之间的相互反馈,从而加速探索过程。在数据驱动的学科中,跨学科专家的协同作用和对动态数据需求的急剧增长仍然是一个挑战。我们相信,对第五范式平台的一瞥可以为其在其他领域的应用铺平道路。Science is entering a new era—the fifth paradigm—that is being heralded as the main character of knowledge integrating into different fields to intelligence-driven work in the computational community based on the omnipresence of machine learning systems.Here,we vividly illuminate the nature of the fifth paradigm by a typical platform case specifically designed for catalytic materials constructed on the Tianhe-1 supercomputer system,aiming to promote the cultivation of the fifth paradigm in other fields.This fifth paradigm platform mainly encompasses automatic model construction(raw data extraction),automatic fingerprint construction(neural network feature selection),and repeated iterations concatenated by the interdisciplinary knowledge(“volcano plot”).Along with the dissection is the performance evaluation of the architecture implemented in iterations.Through the discussion,the intelligence-driven platform of the fifth paradigm can greatly simplify and improve the extremely cumbersome and challenging work in the research,and realize the mutual feedback between numerical calculations and machine learning by compensating for the lack of samples in machine learning and replacing some numerical calculations caused by insufficient computing resources to accelerate the exploration process.It remains a challenging of the synergy of interdisciplinary experts and the dramatic rise in demand for on-the-fly data in data-driven disciplines.We believe that a glimpse of the fifth paradigm platform can pave the way for its application in other fields.

关 键 词：机器学习 自动构建 天河一号 数据驱动 神经网络 动态数据 知识整合 科学研究 

分 类 号：TB30[一般工业技术—材料科学与工程]