Overview
Dive deep into Graph Attention Networks (GAT) in this comprehensive 38-minute video explanation. Learn about basic graph theory, the intricacies of GAT, and its similarities with transformers. Explore the detailed method explanation, multi-head GAT versions, visualizations, spatial pooling, and GNN depth. Understand GAT properties, receptive fields of spatial GNNs, and the differences between transductive and inductive learning. Examine results on various benchmarks and visualize representations using t-SNE. Gain valuable insights into geometric deep learning and expand your knowledge of graph neural networks.
Syllabus
- A note on geometric deep learning
- Graph theory basics
- Intro to GATs related work
- A detailed explanation of the method
- A multi-head version of the GAT
- Visualizations, spatial pooling, GNN depth
- A recap of GAT properties
- Receptive field of spatial GNNs
- Datasets, transductive vs inductive learning
- Results on transductive/inductive benchmarks
- Representations visualization t-SNE
Taught by
Aleksa Gordić - The AI Epiphany