Mastering PyTorch - Second Edition

6

Graph Neural Networks

In the previous chapters, we have discussed various kinds of neural architectures, ranging from convolutional to recurrent, from attention-based transformers to auto-generated neural networks (NNs). While these architectures cover a wide range of deep learning problems, they work best with data that exists in a continuous space, typically represented as vectors, or coordinates in a Euclidean space such as text (1D), images (2D), and videos (3D). However, a huge portion of real-world datasets exists in the form of graphs or networks, such as social networks, protein-interaction networks, literature citation networks, and the World Wide Web, to name a few. In this chapter, we’ll learn about graph neural networks (GNNs