Recurrent Graph Neural Networks: A Journey from Start to End
Mohamed Abdel‐Basset, Nour Moustafa, Hossam Hawash, Zahir Tari · 2023
Graph attention networks are neural networks that can be learnt from graph-structured data, using masked self-attentional layers as a solution to the downsides of methods depending on graph convolutions or their approximations. This chapter explores another popular family of graph networks known as recurrent graph neural networks (GNNs). The chapter begins by discussing the deep learning for sequence modeling and shows the challenges that arise from the ignorance of time dependency. Next, the recurrent GNNs are investigated to show how they positively exploit the history of the process to afford a generic solution for generating embeddings at the node level via an information propagation scheme acknowledged as message passing. Finally, this chapter provides a detailed explanation and implementation of some state-of-the-art algorithms for recurrent GNNs.