Preface

      Deep learning has achieved promising progress in many fields such as computer vision and natural language processing. The data in these tasks are usually represented in the Euclidean domain. However, many learning tasks require dealing with non-Euclidean graph data that contains rich relational information between elements, such as modeling physical systems, learning molecular fingerprints, predicting protein interface, etc. Graph neural networks (GNNs) are deep learning-based methods that operate on graph domains. Due to its convincing performance and high interpretability, GNN has recently been a widely applied graph analysis method.

      The book provides a comprehensive introduction to the basic concepts, models, and applications of graph neural networks. It starts with the basics of mathematics and neural networks. In the first chapters, it gives an introduction to the basic concepts of GNNs, which aims to provide a general overview for readers. Then it introduces different variants of GNNs: graph convolutional networks, graph recurrent networks, graph attention networks, graph residual networks, and several general frameworks. These variants tend to generalize different deep learning techniques into graphs, such as convolutional neural network, recurrent neural network, attention mechanism, and skip connections. Further, the book introduces different applications of GNNs in structural scenarios (physics, chemistry, knowledge graph), non-structural scenarios (image, text) and other scenarios (generative models, combinatorial optimization). Finally, the book lists relevant datasets, open source platforms, and implementations of GNNs.

      This book is organized as follows. After an overview in Chapter 1, we introduce some basic knowledge of math and graph theory in Chapter 2. We show the basics of neural networks in Chapter 3 and then give a brief introduction to the vanilla GNN in Chapter 4. Four types of models are introduced in Chapters 5, 6, 7, and 8, respectively. Other variants for different graph types and advanced training methods are introduced in Chapters 9 and 10. Then we propose several general GNN frameworks in Chapter 11. Applications of GNN in structural scenarios, nonstructural scenarios, and other scenarios are presented in Chapters 12, 13, and 14. And finally, we provide some open resources in Chapter 15 and conclude the book in Chapter 16.

      Zhiyuan Liu and Jie Zhou

      March 2020