This article explores modern graph neural networks, which operate on graph data like relationships between objects. GNNs have practical applications in antibacterial discovery, physics simulations, fake news detection, traffic prediction, and recommendation systems. Graphs can represent various data types, such as social networks, images, and text. Understanding the structure and attributes of graphs is crucial. The article discusses challenges in using graphs in machine learning, the types of prediction tasks on graphs, and introduces GNNs for solving these tasks. The GNN architecture preserves graph symmetries and can be used for graph prediction tasks like graph-level, node-level, and edge-level predictions.
https://distill.pub/2021/gnn-intro/