Machine learning in materials research - Developments over the last decade and challenges for the future

Aug 4, 2026

Upload by Riley

The review analyzes how machine learning (ML) in materials science expanded rapidly over the past decade, with study volume increasing at about 1.67× per year. It surveys commonly used tools, databases, materials science methods, and ML approaches, showing classical ML remains dominant despite deep learning growth. It also tracks matbench performance gains, highlighting an error reduction by about 7× from feature-based models to graph neural networks using density functional theory data. The article concludes with future challenges and opportunities around data size and complexity, extrapolation, interpretation, and access.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

AccessFree to download

Introduction

Field growth and enabling factors

Cross-fertilization of methods and data

Tools, benchmarks, and future challenges

Lawrence Berkeley National Laboratory LBL Publications Title Machine learning in materials research: Developments over the last decade and challenges for the future Permalink [https://escholarship.org/uc/item/6pf002vg](https://escholarship.org/uc/item/6pf002vg) Author Jain, Anubhav Publication Date 2024-12-01 DOI 10.1016/j.cossms.2024.101189 Copyright Information This work is made available under the terms of a Creative Commons Attribution License, available at [https://creativecommons.org/licenses/by/4.0/](https://creativecommons.org/licenses/by/4.0/) Peer reviewed [eScholarship.org](eScholarship.org) Powered by the California Digital Library University of California Current Opinion in Solid State and Materials Science 33 (2024) 101189 Contents lists available at ScienceDirect Current Opinion in Solid State & Materials Science journal [homepage: www.elsevier.com/locate/cossms](homepage: www.elsevier.com/locate/cossms) | Machine learning in materials research: Developments over the last decade and challenges for the future
Anubhav Jain
Energy Technologies Area, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, United States | | | | --- | --- | --- | | A R T I C L E I N F O | | A B S T R A C T | | Keyword:
Machine learning in materials science | | The number of studies that apply machine learning (ML) to materials science has been growing at a rate of approximately 1.67 times per year over the past decade. In this review, I examine this growth in various contexts. First, I present an analysis of the most commonly used tools (software, databases, materials science methods, and ML methods) used within papers that apply ML to materials science. The analysis demonstrates that despite the growth of deep learning techniques, the use of classical machine learning is still dominant as a whole. It also demonstrates how new research can effectively build upon past research, particular in the domain of ML models trained on density functional theory calculation data. Next, I present the progression of best scores as a function of time on the matbench materials science benchmark for formation enthalpy prediction. In particular, a dramatic improvement of 7 times reduction in error is obtained when progressing from feature-based methods that use conventional ML (random forest, support vector regression, etc.) to the use of graph neural network techniques. Finally, I provide views on future challenges and opportunities, focusing on data size and complexity, extrapolation, interpretation, access, and relevance. | 1. Introduction The use of machine learning techniques in materials research has grown in the last decade from a small niche topic to an entire subfield within materials science & engineering. Indeed, there have been over 2000 papers on the topic of materials machine learning in the year 2023 alone, and over the past decade there has been a 1.67 times yearly growth in the number of papers (Fig. 1). A 2020 review by Morgan and Jacobs [1] found that not only were the number of papers on the topic exponentially increasing, but that the number of review papers per year on the topic had already reached nearly 40 by 2019. Indeed, there already exist many excellent reviews on various aspects of materials machine learning, including its applications in simulation and modeling [2–4], synthesis and characterization [5–7], manufacturing [8,9], and literature mining [10]. Reviews also exist for specific topics such as structural materials [11] or best practices for research reporting [12]. This review both looks back and looks ahead. Looking back, it examines what has enabled the field of machine learning to advance so rapidly. Indeed, about five years ago it was unclear whether the field would enter a “trough of disillusionment” or an “AI winter” [13]. However, the development of the Transformer architecture [14] in the computer science domain and the crystal graph neural network

FAQs

How has the volume of machine learning studies in materials science changed over the last decade?

Studies applying ML to materials science have grown at approximately 1.67 times per year over the past decade.

Which ML approach remains dominant despite advances in deep learning?

Overall, classical machine learning is still dominant, even though deep learning techniques have increased.

What leads to large improvements on the matbench formation enthalpy prediction benchmark?

Progressing from feature-based classical ML methods to graph neural network techniques yields a dramatic improvement, reducing error by about 7× over time.

Aug 4, 2026

Upload by Riley