Computing Graph Neural Networks: A Survey from Algorithms to Accelerators

Graph Neural Networks (GNNs) have exploded onto the machine learning scene in recent years owing to their capability model and learn from graph-structured data. Such an ability has strong implications a wide variety of fields whose data are inherently relational, for which conventional neural networks do not perform well. Indeed, as reviews can attest, research area GNNs grown rapidly lead development GNN algorithm variants well exploration ground-breaking applications chemistry, neurology, electronics, or communication networks, among others. At current stage research, however, efficient processing is still open challenge several reasons. Besides novelty, hard compute due dependence on input graph, combination dense very sparse operations, need scale huge graphs some applications. In this context, article aims make two main contributions. On one hand, review field presented perspective computing. This includes brief tutorial fundamentals, overview evolution last decade, summary operations carried out multiple phases different variants. other in-depth analysis software hardware acceleration schemes provided, hardware-software, graph-aware, communication-centric vision accelerators distilled.