Alienpenguin

Published

- 1 min read

Graph Neural Networks

deeplearning gnn
img of Graph Neural Networks

Graph Neural Networks

Previously we have seen convolution neural networks (CNN) that is used for process arrays of data, i.e. images and then transformers that specialises in processing sequences of variable length data, i.e. text. Similarly, we have graph neural networks to process graphs.

Graphs

Graph consists of set of nodes or vertices, where pairs of nodes are connected by edges or links. We can categorise graph based on:

  • Does the direction of edge matter?
  • Are the nodes same entities (e.g. people, countries, companies)
  • Number of edges and types of them between nodes

Here are some categories of graphs and example of where we might see them:

  1. Directed graph
  2. Undirected graph
  3. Heterogeneous graph
  4. Homogeneous graphs
  5. Multigraphs
  6. Geometric graphs
  7. Hierarchical graphs

All type of graphs can be processed using deep learning. However, we will focus on undirected graph in this article.

We might want to store following information regarding a graph:

  1. Graph structure
  2. Information about the node
  3. Information about the edge