SCOPE OF MACHINE LEARNING IN SOLID STATE MATERIAL: A COMPREHENSIVE REVIEW

Abstract

The machine learning is one of the most exciting instrument that have joined the material science toolbox in recent years. This set of statistical methods has already proven to be able to accelerate both fundamental as well as applied research considerably. We are currently experiencing an explosion of work that generates and applies machine learning to solid-state systems. In this arena, we are providing a detailed overview as well as the review of the most recent study. We are incorporating machine learning principles, algorithms, descriptors, and databases in materials science as a starting point. We continue with the overview of various approaches to machine learning to discover stable materials and to predict their crystal structure. In various quantitative structure-property relationships and different techniques, study is then addressed to replace first-principle methods with machine learning. We review how it is possible to apply active learning and surrogate-based optimization to enhance the rational design process and related application examples. The interpretability of and the physical understanding obtained from machine learning models are still two big questions. The numerous aspects of interpretability and their meaning in the science of materials are therefore considered. Finally, in computational materials science, we suggest ideas and possible study directions for different problems.