This repository includes the material of the GNNs course offered by Deep Learning Italia Academy.
| Module | Description | Notebook |
|---|---|---|
| Graph Representation | Represent different types of graphs in NumPy and NetworkX |  |
| Traditional ML methods on Graphs | Extract meaningful features from both nodes and edges in a graph with NetworkX | |
| Basic Training Loop from NumPy to PyTorch | Perform a general training with NumPy and PyTorch to understand the key principles of learning | |
| Shallow embedding methods | Apply shallow embedding methods, such as Node2Vec, for graph classification using PyG | |
| GCN Layer | Explore using pure NumPy the key principles of Graph Convolutional Networks (GCNs) | |
| GAT Layer | Understand the inner working of the GAT Layer in NumPy and compare it with the GCN layer | |
| GraphSage Layer | Understand the inner working of the GraphSage Layer in NumPy and compare it with the GCN layer | |
| GraphGPS | Test with PyG the effectiveness of Graph Transformer architectures for node property prediction | |
| Graph Isomorphism | Identify using NumPy the key aspects related to the structural similarity between graphs | |
| Permutation invariance and equivariance | Test permutation equivariance and invariance in Graph Neural Networks with NumPy | |
| Weisfeiler-Lehman Isomorphism Test | Measure the expressiveness of GNNs with the Weisfeiler-Lehman algorithm implemented in NumpY | |
| GCN vs GIN | Compare the expressive power of GCNs and GINs for graph classification using PyG | |
| Node classification | Apply node classification comparing MLP (multilayer perceptron) and GCN with PyG | |
| Graph Classification | Predict the categories of graphs based on the structural graph properties leveraging PyG | |
| Scaling GNNs | Scale GNNs with PyG by adopting the Cluster-GCN algorithm | |
| GNNs explainablility | Explain GNNs results using PyG and Captum | |
| Link Prediction | Forecast missing connections in graphs with PyG | |
| Link Regression | Predict continuous-valued edge attributes in graph-structured data with PyG | |
| R-GCN Layer | Extend the definition of GCN for processing heterogenous graph (aka Knowledge Graphs) | |
| KG Embeddings | Implement basic KG embedding algorithms with NumPy and Pykeen | |