This course introduces the theoretical foundations, architectures, and applications of Graph Neural Networks. It covers message-passing frameworks, graph convolutional models, and advanced architectures for heterogeneous and dynamic graphs. Students will learn to design, implement, and evaluate GNNs for tasks such as node classification, link prediction, and graph-level representation learning.
The course provides a comprehensive understanding of graph-based deep learning. It begins with fundamental graph theory and network representation concepts, followed by detailed study of spectral and spatial GNN formulations. Key models, including GCN, GAT, GraphSAGE, and heterogeneous GNNs, are examined. Practical sessions involve applying GNNs using PyTorch Geometric and DGL libraries on real-world datasets. The course concludes with discussion of current research trends, scalability challenges, and interpretability in graph learning.
At the end of the course, the learner will be able to:
-explain the mathematical principles underlying graph representation learning;
-implement and train standard GNN architectures;
-evaluate model performance on graph-structured data;
-design and adapt GNN architectures for new tasks and heterogeneous graphs;
-critically analyze research papers in the GNN field.
Solid background in linear algebra, probability, and machine learning.
Working knowledge of Python and deep learning frameworks (preferably PyTorch).
Recommended:
W. Hamilton, Graph Representation Learning, Morgan & Claypool, 2020.
Recommended:
Z. Wu et al., “A Comprehensive Survey on Graph Neural Networks,” IEEE Transactions on Neural Networks and Learning Systems, 2021.
T. Kipf and M. Welling, “Semi-Supervised Classification with Graph Convolutional Networks,” ICLR, 2017.
Y. Hu et al., “Heterogeneous Graph Neural Networks for Scalable Asymmetric Traveling Salesman Problem Optimization,” Neurocomputing, 2025.
Lectures on theory and model design.
Programming labs for implementation and experimentation.
Group project applying GNNs to a real dataset.
Paper discussions and student presentations on recent research.
Transcript of records