Название: Machine Learning for Materials Discovery: Numerical Recipes and Practical Applications
Автор: N. M. Anoop Krishnan, Hariprasad Kodamana, Ravinder Bhattoo
Издательство: Springer
Год: 2024
Страниц: 287
Язык: английский
Формат: pdf (true)
Размер: 11.5 MB

Focusing on the fundamentals of Machine Learning, this book covers broad areas of data-driven modeling, ranging from simple regression to advanced Machine Learning and optimization methods for applications in materials modeling and discovery. The book explains complex mathematical concepts in a lucid manner to ensure that readers from different materials domains are able to use these techniques successfully. A unique feature of this book is its hands-on aspect―each method presented herein is accompanied by a code that implements the method in open-source platforms such as Python. This book is thus aimed at graduate students, researchers, and engineers to enable the use of data-driven methods for understanding and accelerating the discovery of novel materials.

The book offers an excellent pedagogical approach towards the use of Machine Learning for materials discovery. The book is written in a lucid fashion, and accessible to audience ranging from undergraduate students to scientists. The book does not assume any prior knowledge in the domain of Machine Learning, and is self-sufficient. The second part of the book covers the basics of Machine Learning theory including supervised and unsupervised strategies with examples from the materials domain. An excellent feature of the book is that theory on Machine Learning is followed by codes that allows instructors, students, and practitioners to try the approaches in a hands-on fashion. The third section discusses a wide range of applications giving an overview of different avenues where Machine Learning can be used for materials discovery.

In this chapter, we discuss advanced Deep Learning algorithms focusing on their application and impact in the field of Machine Learning. We discuss the transformative impact of Deep Learning compared to classical approaches, which heavily rely on handcrafted features and hyperparameter tuning. To this extent, the chapter explores a range of advanced Deep Learning models, including Convolutional Neural Networks (CNNs) for materials image analysis, Long Short-Term Memory networks (LSTMs) for sequential materials data, Generative Adversarial Networks (GANs) for generating new material structures, Graph Neural Networks (GNNs) for analyzing materials graphs, Variational Autoencoders (VAEs) for materials representation learning, and Reinforcement Learning (RL) which has been widely used in materials domain. Each model is presented with a detailed explanation of its underlying principles, architectures, and training methodologies. By exploring these advanced Deep Learning techniques, researchers and practitioners in the field of materials can gain valuable insights into leveraging Deep Learning models to accelerate the exploration of novel materials and optimize material properties.

Contents: