# An Even Gentler Introduction to Graph Neural Networks

Graph neural networks (GNN), like all neural networks, sound intimidating simply because of their name. However, if we take the name at face value we may reason that somewhere there seems to be a graph and it may be associated with a neural network. OK, so far, so good.

Based upon your mathematics education you may know that a graph is simply a set of objects with some relation to each other that form a structure. Further, we may also know from mathematics that if you map a function you’ve essentially created a graph.

We may also already know that a neural network is a model of how the human brain works. It has a series of nodes that are placed in layers. It is a key tool of artificial intelligence and it uses algorithms to cluster, classify or perform a regression.

However, what is meant by using these two now well defined terms together and how is applied?

A GNN processes data represented in a graph structure, such as those in a mathematical function. The neural network can interpret graphs in such a way that is not always intuitive to humans. They were first found to be quite handy in supervised learning problems related to the properties of various molecules.

Variants of this advanced form of graph analysis add to their mystique, but not necessarily to the practical knowledge required by working data scientists. A common sense approach would be to review possible applications, sort through what applies to your primary field of study, and forget the rest until you have time on a beach in Mexico to sit and read about them for fun.

If you really think about it, data structured as a graph is everywhere and thus GNNs can be applied in several different situations. The applications can be classified into these categories:

1. Node Classification: Perhaps you must determine the labeling of samples (represented as nodes) by looking at the labels of those points closest to them. This is a semi-supervised method with only part of the graph being labelled. An example could be that a social network wants to map out the relationships between users. If you understand the relationship between users you can better market products to them. It is not unreasonable to think that part of it could be…