Hands-On Graph Neural Networks: From Theory to Practice

Graph Neural Networks (GNNs) have emerged as a powerful paradigm for deep learning on relational, non-Euclidean data. This two-hour hybrid session offers a practical introduction to GNNs, seamlessly bridging theoretical foundations with hands-on implementation. The first part covers the core principles of graph-structured learning, including message-passing mechanics, neighborhood aggregation, graph convolutions, and pooling strategies. The session then transitions into an interactive, guided Google Colab workshop using PyTorch Geometric. Working with structured starter code, participants will implement GNN architectures step-by-step, train models for both node-level and graph-level learning tasks, and visualize embeddings in real time.

Designed for developers, data scientists, and researchers with basic Python and PyTorch knowledge, this session provides attendees with both the theoretical clarity and actionable coding skills required to apply GNNs to real-world relational problems.