Recent advances in machine learning and its applications in solid-state material science

 Recent advances and applications to machine learning in solid state materials science

Abstract

Machine learning has been one of the most important tools in material science over the past few years. This set of statistical methods is already capable of significantly speeding up both applied and fundamental research. Presently, there is a flurry of work that applies machine learning to solid state systems. We present a comprehensive overview of the latest research in this area. We start with machine learning principles and algorithms, as well as descriptors and databases in materials sciences. Next, we discuss different machine-learning approaches that are used for the prediction and discovery of stable materials. We then discuss research in a variety of quantitative structure-property relationship and the various methods for replacing first-principle techniques with machine learning. We will discuss the use of surrogate-based and active learning to enhance rational design, and provide examples. There are two main questions: how machine learning models can be understood and what it can do for the human body. We examine the importance of interpretability in materials science and discuss their different aspects. Finally, we offer future research pathways and solutions to various problems in computational materials science.

Introduction

The availability of large datasets, combined with improvements in algorithms and an exponential increase in computing power has led to a phenomenal rise in interest in machine learning. Machine learning algorithms are used for clustering, classification, regression, and dimensionality reduction of large-dimensional input data. Machine learning algorithms power large swathes of our daily lives. These include image and speech recognition5,6 web searches,7 fraud detection8,8 email/spam filtering and 9 credit scores,10.

Although data-driven research, specifically machine learning, has a long history within biology11 and chemistry12, they have only recently gained prominence in the field of solid state materials science.

In the past, experiments played an important role in discovering and characterizing new materials. Due to the high demands on resources and equipment, experimental research should be done over a long period of time for a limited number of materials. Due to these limitations, most important discoveries in materials science were made through intuition or serendipity.14 The advent of computational methods such as density functional theory (DFT),15-16 Monte Carlo simulations and molecular dynamic allowed researchers to better understand the composition and phase space. Combining computer simulations with experiments has helped to significantly reduce the time and cost involved in materials design.17-18,19,20 The continuous increase in computing power, as well as the development of more efficient code, allowed for computational high throughput studies21 of large groups of material to help in the selection of the most promising experimental candidates. Combining large-scale simulations and calculations with experimental high throughput studies22-24,25, makes it possible to use machine learning techniques in materials science. An Easy Introduction to a Diamond Mine Card Stud - Free Bull Run Card Stud .