Graph Neural Networks (GNNs)

preview_player
Показать описание
Slides and Code can be found at:

0:00 What can be modelled with Graph Neural Network (GNN)?
6:15 GNN introductory video
12:17 How many hops needed for n nodes?
15:37 GNN for finding out maximum node value
19:27 Overview of GNN
29:27 Viewing a CNN as a graph
32:51 Graph Convolutional Network (GCN)
37:38 Graph Attentional Network (GAT)
49:51 How to train GNN end-to-end
54:27 Permutation Invariance in GNNs
1:00:57 How Transformers is actually a GAT with complete edge set
1:02:36 Cora dataset
1:04:32 Jupyter Notebook on GCN/GAT on node classification on Cora dataset
Рекомендации по теме
Комментарии
Автор

Correction:
In 48:30, I mentioned that the W will include maximum number of nodes. That is actually incorrect. It will just do a mapping of the concatenated hidden space representation of two nodes into a single scalar value. The node connections itself is in the edge matrix which just gives information of who are the neighbours of each node in order to do message passing.

johntanchongmin