Graph-Based Methods

Graph-based machine learning has experienced a resurgence in recent years, as methods such as graph neural networks have proven to be useful in large-scale, real-life scenarios, including traffic optimization and drug interaction prediction. While recent theoretical foundations on geometric deep learning have highlighted the link between graph neural networks and other types, we consider graph-related learning separately, as applications may often differ. In this chapter, we provide an overview of the field of graph-based machine learning. We begin by discussing the fundamental concepts and definitions related to graphs, including their representation and properties. We then explore the key challenges and opportunities in graph-based machine learning, including scalability, modelling complex dependencies, and interpretability. Next, we turn our attention to two selected tasks that fall under the umbrella of canonical graph-based machine learning: node classification and link prediction. Node classification is the task of assigning a label to each node in a graph based on its attributes and the structure of the graph. Link prediction, on the other hand, involves predicting the existence or absence of edges between pairs of nodes in a graph. For each of these tasks, we examine the state-of-the-art methods and their applications in real-world scenarios. We also discuss the limitations of these methods and identify areas for future research. Throughout the chapter, we provide examples and illustrations to help readers understand the concepts and methods discussed. In summary, graph-based machine learning is an exciting and rapidly evolving field that has shown great promise in various applications. By providing an overview of the field and examining two canonical tasks, this book provides a useful resource for researchers and practitioners interested in this area.