Skip to content

References: Conclusion: The GNN Design Space and Open Problems

  1. Graph Neural Network - Wikipedia - Overview of the GNN family, historical development from spectral to spatial methods, and connections to related architectures; provides a stable reference for the field's foundational terminology.

  2. Knowledge Graph - Wikipedia - Covers the structure, history, and major applications of knowledge graphs, situating them within the broader graph ML design space discussed throughout this textbook.

  3. Transfer Learning - Wikipedia - Explains the core ideas of pre-training, fine-tuning, and domain adaptation that underpin graph foundation models — one of the central open problems in graph ML.

  4. Graph Representation Learning - William L. Hamilton - Morgan & Claypool (Synthesis Lectures on AI and ML), 2020 - A compact, self-contained treatment of node embeddings, GNNs, and knowledge graph methods that complements the material in this textbook and provides a unified theoretical lens.

  5. Deep Learning on Graphs - Yao Ma & Jiliang Tang - Cambridge University Press, 2021 - Broad coverage of GNN architectures, training strategies, and applications; particularly strong on heterogeneous and dynamic graphs, filling gaps beyond the core GCN/GAT curriculum.

  6. A Comprehensive Study on Large-Scale Graph Neural Networks (arXiv:2005.09979) - arXiv - Benchmarks GNN scalability methods across architectures and datasets, directly supporting the open problem of scaling graph learning to billion-node graphs.

  7. Towards Foundation Models for Knowledge Graph Reasoning (arXiv:2310.04562) - arXiv - Proposes a framework for graph foundation models that generalise across knowledge graphs without retraining, illustrating one of the most active open research directions in graph ML.

  8. Open Graph Benchmark: Datasets for Machine Learning on Graphs (arXiv:2005.00687) - arXiv - Introduces the OGB benchmark suite used throughout this textbook, establishing standardised evaluation protocols that are essential for tracking progress on open problems.

  9. PyTorch Geometric Documentation — Creating Message Passing Networks - PyTorch Geometric Docs - Step-by-step guide to implementing custom GNN layers via the message-passing API, directly supporting the "how to build new architectures" dimension of the GNN design space.

  10. Papers With Code — Graph Neural Networks - Papers With Code - Continuously updated leaderboard and method catalog for GNN research; an invaluable resource for tracking state-of-the-art results on the benchmark tasks discussed in the open-problems sections.