Big Data-Driven Materials Science and Its FAIR Data Infrastructure

Abstract

This chapter addresses the forth paradigm of materials research ˗ big-data driven materials science. Its concepts and state-of-the-art are described, and its challenges and chances are discussed. For furthering the field, Open Data and an all-embracing sharing, an efficient data infrastructure, and the rich ecosystem of computer codes used in the community are of critical importance. For shaping this forth paradigm and contributing to the development or discovery of improved and novel materials, data must be what is now called FAIR ˗ Findable, Accessible, Interoperable and Re-purposable/Re-usable. This sets the stage for advances of methods from artificial intelligence that operate on large data sets to find trends and patterns that cannot be obtained from individual calculations and not even directly from high-throughput studies. Recent progress is reviewed and demonstrated, and the chapter is concluded by a forward-looking perspective, addressing important not yet solved challenges.

The NOMAD laboratory: from data sharing to artificial intelligence

“…by artificial-intelligence (AI) methods. The vision of where this field should be going together with the most successful data-analytics concepts pursued in materials science and several examples thereof are described in [1].…”

“…Since going online in 2014, the number of calculations collected in the NOMAD Repository has exceeded even the most ambitious expectations. Thereby the NOMAD Center of Excellence [8] has assumed a pioneering role in data sharing and analytics, and in particular in all aspects of what is now called the FAIR handling of data [1,10] 3 . It has always been NOMAD's principle that data is Findable for anyone interested and stored in a way that make it easily Accessible; its representation follows accepted standards [12,13], with open specifications, making data Interoperable; and finally, all of this makes the data Re-purposable 4 , i.e.…”

“…The NOMAD infrastructure provides the ideal platform for data-driven science. This topic is discussed in detail in [1]. Here, we focus on the NOMAD Analytics Toolkit which provides a collection of examples and tools to demonstrate how materials data can be turned into knowledge.…”

“…Big-data driven materials science is emerging as a new research paradigm at the beginning of the 21st century. The amount of data that has been created in materials science in recent years and is being created every new day is immense, and the field is facing the 4V challenges of big data [1]. Bringing together big data with highperformance computing is a new, emerging field with a game-change potential for numerous applications.…”

Simultaneous learning of several materials properties from incomplete databases with multi-task SISSO

“…The materials-genome initiative [1] inspired the establishment of several high-throughput computational materials-science projects, leading to the creation of worldwide accessible materials databases [2][3][4][5]. In this context, the Novel Materials Discovery (NOMAD) Repository and Archive is the biggest data base for input and output files of density-functional theory (DFT) calculations for materials considering all important computer codes of the community [6][7][8]. It plays synergistically together with other important data bases, in particular AFLOW [2], Materials Project [3], and OQMD [4].…”