Advanced Graph Neural Networks

Course details

Explore graph neural networks (GNNs) in depth. Instructor Janani Ravi begins by delving into the workings of GNNs, covering message passing, aggregation, transformation, transformation math, and attention mechanisms like GATv2Conv. Janani explores practical applications such as node classification, graph classification, and link prediction using datasets like Cora and PROTEINS. Hands-on exercises on Colab with PyTorch Geometric provide experience in setting up and training GNN models. Learn about mini-batching and neighborhood normalization to tackle graph data challenges. This course is ideal for researchers, data scientists, and anyone interested in deep learning or graph theory. Tune in to unlock new potentials in data analysis and modeling with GNNs.

Concepts

0. Introduction

01 - Overview of graph neural networks
02 - Prerequisites

1. Overview of Graph Neural Networks

03 - Message passing in GNNs
04 - Aggregation and transformation math
05 - Aggregation and transformation math in matrix form

2. Node Classification with Graph Attention Networks

06 - Introducing graph attention
07 - Computing the attention coefficient
08 - Including attention in GNN layers
09 - Getting set up with Colab and the PyTorch Geometric library
10 - Exploring the Cora dataset
11 - Setting up the graph convolutional network
12 - Training a graph convolutional network
13 - Node classification using a graph attention network
14 - Using the GATv2Conv layer for attention

3. Graph Classification Using Graph Convolution

15 - Understanding graph classification
16 - Exploring the PROTEINS Dataset for graph classification
17 - Minibatching graph data
18 - Setting up a graph classification model
19 - Training a GNN for graph classification
20 - Eliminating neighborhood normalization and skip connections

4. Link Prediction Using Graph Autoencoders

21 - A quick overview of autoencoders
22 - Introducing graph autoencoders
23 - Splitting link prediction data
24 - Understanding link splits
25 - Designing an autoencoder for link prediction
26 - Training the autoencoder

Conclusion

27 - Summary and next steps