Explore graph neural networks (GNNs) in depth to unlock new potential in data analysis and modeling.
Overview
Syllabus
Introduction
- Overview of graph neural networks
- Prerequisites
- Message passing in GNNs
- Aggregation and transformation math
- Aggregation and transformation math in matrix form
- Introducing graph attention
- Computing the attention coefficient
- Including attention in GNN layers
- Getting set up with Colab and the PyTorch Geometric library
- Exploring the Cora dataset
- Setting up the graph convolutional network
- Training a graph convolutional network
- Node classification using a graph attention network
- Using the GATv2Conv layer for attention
- Understanding graph classification
- Exploring the PROTEINS Dataset for graph classification
- Minibatching graph data
- Setting up a graph classification model
- Training a GNN for graph classification
- Eliminating neighborhood normalization and skip connections
- A quick overview of autoencoders
- Introducing graph autoencoders
- Splitting link prediction data
- Understanding link splits
- Designing an autoencoder for link prediction
- Training the autoencoder
- Summary and next steps
Taught by
Janani Ravi