My Activity
Recently Viewed
Machine Learning in Materials Science
Author(s):
- Keith T. ButlerSenior ScientistRutherford Appleton LaboratoryEmail: [email protected]More by Keith T. Butler
- Felipe OviedoApplied Scientist and Research AffiliateMicrosoft AI For Good and Massachusetts Institute of TechnologyEmail: [email protected]More by Felipe Oviedo
- Pieremanuele CanepaAssistant ProfessorNational University of SingaporeEmail: [email protected]More by Pieremanuele Canepa
Publication Date:
June 16, 2022
Copyright ©
2022 American Chemical Society
eISBN:
9780841299467
DOI:
10.1021/acsinfocus.7e5033
Read Time:
six to seven hours
Collection:
1
Publisher:
American Chemical Society
Machine Learning for Materials Science provides the fundamentals and useful insight into where Machine Learning (ML) will have the greatest impact for the materials science researcher. This digital primer provides example methods for ML applied to experiments and simulations, including the early stages of building an ML solution for a materials science problem, concentrating on where and how to get data and some of the considerations when choosing an approach.
The authors demonstrate how to build more robust models, how to make sure that your colleagues trust the results, and how to use ML to accelerate or augment simulations, by introducing methods in which ML can be applied to analyze and process experimental data. They also cover how to build integrated closed-loop experiments where ML is used to plan the course of a materials optimization experiment and how ML can be utilized in the discovery of materials on computers.
This book is coming soon. In the meantime, please contact your library or ACS sales representative for purchase options.
Detailed Table of Contents
About the Series
Preface
Chapter 1
Applying Machine Learning (ML) to Materials Science
1.1
Chapter Overview
1.2
Data and Databases
1.2.1
The Five Vs of Data Science
1.2.2
Databases
1.2.3
Interacting with Material Databases and Application Programming Interfaces
1.2.4
Common Materials Databases
1.2.5
Understanding Data Distributions and Relations
1.3
Representations and Representation Learning
1.3.1
Distance Metrics
1.4
Evaluation
1.4.1
Evaluation Metrics
1.4.2
Protecting against Overfitting
1.5
Insider Q&A: Aron Walsh
1.6
That’s a Wrap
1.7
Read These Next
Chapter 2
Building Trust in Machine Learning
2.1
Chapter Overview
2.2
Convolutional Neural Networks (CNNs)
2.3
Insider Q&A: Aron Walsh
2.4
Advanced Validation
2.4.1
Leave-One-Cluster-Out Cross Validation
2.4.2
Dimensionality Reduction
2.4.2.1
Principal Component Analysis
2.4.2.2
Manifold Learning
2.4.2.3
Autoencoders
2.4.3
Clustering Data
2.4.3.1
The k-Means Clustering Method
2.4.3.2
Gaussian Mixture Models
2.5
Uncertainty Quantification
2.5.1
Sources of Uncertainty
2.5.2
Bayesian Methods and Gaussian Processes
2.5.2.1
Bayesian Neural Networks
2.5.2.2
Gaussian Processes
2.5.3
Ensemble Methods
2.6
Interpretability
2.6.1
Interpreting Classical Models
2.6.2
Interpreting Deep Models
2.7
Insider Q&A: Maxim Ziatdinov, Ph.D.
2.8
A Day in the Life
2.9
That’s a Wrap
2.10
Read These Next
Chapter 3
Machine Learning for Materials Simulations
3.1
Chapter Overview
3.2
Representing Materials for ML
3.2.1
Atomic Representations
3.2.2
From Atoms to Compounds
3.2.3
From Composition to Structure
3.2.3.1
Matrix Descriptors
3.2.3.2
Local Environment Descriptors
3.2.3.3
Graph Neural Networks
3.3
Insider Q&A: Nongnuch Artrith
3.4
Which Representation Is Best?
3.5
Neural Network Interatomic Potentials
3.6
A Day in the Life
3.7
That’s a Wrap
3.8
Read These Next
Chapter 4
Analyzing Experimental Data
4.1
Chapter Overview
4.2
Machine Learning for Electron Microscopy (EM)
4.2.1
U-Nets
4.2.2
Denoising EM
4.2.3
Semantic Segmentation
4.3
Machine Learning for Diffraction
4.3.1
Supervised Learning of X-ray Diffraction (XRD) Patterns
4.3.1.1
Data Augmentation
4.3.1.2
Model Interpretation
4.3.2
Extracting Phase Information from XRD
4.4
ML Analysis of Spectral Data
4.4.1
CNNs with Uncertainty Quantification
4.4.2
Ensemble Learning for X-ray Absorption Spectroscopy
4.5
That’s a Wrap
4.6
Read These Next
Chapter 5
Closed-Loop Optimization and Active Learning for Materials
5.1
Chapter Overview
5.2
Black-Box Optimization
5.3
Bayesian Optimization (BO)
5.3.1
Surrogate Models
5.3.2
Acquisition Functions
5.4
Including Domain Knowledge into Black-Box Optimization
5.5
Navigating High-Dimensional Spaces
5.6
BO Implementations and Other Efficient Algorithms
5.7
Insider Q&A: Shijing Sun
5.8
That’s a Wrap
5.9
Read These Next
Chapter 6
Discovering New Materials
6.1
Chapter Overview
6.2
High-Throughput Virtual Screening
6.2.1
Composition-Based Prediction
6.2.2
Predicting Structure
6.2.3
Higher-Level Predictions
6.3
Direct Optimization with Generative Models
6.3.1
Variational Autoencoders
6.3.2
Generative Adversarial Networks
6.3.3
Reinforcement Learning
6.4
Multiobjective Optimization
6.5
Insider Q&A: Olga López-Acevedo
6.6
That’s a Wrap
6.7
Read These Next
Chapter 7
Coda
Bibliography
Footnotes
Glossary
Index
Reviewer quotes
Recommended to any scientist seeking to bridge their research in materials simulation, design, or experimentation with machine learning
The authors provide an excellent pedagogical overview of standard and emerging machine learning principles used in materials design. Most experimental and theoretical material scientists have not had a formal training in machine learning or data science, while the field is collectively using these approaches to accelerate their research. This digital book provides readers with the sufficient background and tricks-of-the-trade to become literate in machine learning and directly apply these algorithms to their problems.
Incredibly useful resource to the materials engineering community
Machine Learning in Materials Sciences introduces to scientists, early or established, the possibilities of machine learning methods and the fundamental techniques needed to use them effectively. I would gladly use this book as a reference for the “Applied AI in Materials Science” course I teach.
Keith T. Butler is a senior scientist at the Rutherford Appleton Laboratory, in the Scientific Machine Learning (SciML) team, where he leads projects that apply machine learning to the discovery and characterisation of materials. Keith obtained his bachelor's from Trinity College Dublin and his PhD from University College London. Before joining RAL, Keith spent time as a post-doctoral researcher in the groups of Aron Walsh (University of Bath/Imperial College London) and John Harding (University of Sheeld), and was a visiting researcher in The University of Toronto and Tokyo Institute of Technology. Keith's research focuses on using machine learning to accelerate the characterisation of materials and also to predict new, previously undiscovered materials for renewable energy applications, such as photovoltaics and photocatalysts. Keith is an active developer of several open source materials design packages (SMACT, SuperResTomo, Macro-Density) and a strong advocate of open science.
Felipe Oviedo is an applied researcher at Microsoft AI for Good, focusing on scientific machine learning for sustainability and healthcare applications. Prior to joining Microsoft, Felipe completed a PhD at the intersection of material science and computer science at MIT under the guidance of Prof. Tonio Buonassisi (MIT Mechanical Engineering) and Dr. John Fisher (MIT CSAIL). His dissertation was focused on accelerated development of photovoltaics by physics-informed machine learning. Felipe developed and deployed machine learning algorithms to accelerate the experimental screening and optimization of renewable energy materials and technologies. Before MIT, Felipe briefly worked in the energy industry and CERN.
Pieremanuele Canepa is an Assistant Professor in the Department of Materials Science and Engineering at the National University of Singapore (NUS). He received his bachelor's and master's degrees in Chemistry from the University of Torino (Italy) and a PhD from the University of Kent (UK). Prior to NUS, he was a Postdoctoral fellow at the Lawrence Berkeley National Laboratory and the Massachusetts Institute of Technology under the guidance of Prof. Gerbrand Ceder. His research contributes to the rational design of materials for clean energy technologies, including electrode materials for batteries, and electrolytes for sustainable energy storage devices. In 2021, Pieremanuele was elected as fellow of the Royal Society of Chemistry.
