The power of Graph Neural Networks

Kåre von Geijer published on
15 min, 2944 words

There is something about graph algorithms - they are just always fun to think about! With that said, Graph Neural Netorks (GNN) are no exception. These are a form of neural network, but designed in a way to learn things about graphs. They have proven to be usefull for a plethora of graph learning problems in fields such as drug discovery, recommender systems, and protein folding. In this post, I will present the difficulty of directly applying neural networks (NN) to graphs, how message passing NN solves this, and end by looking at the expressive power of such GNN.

Read More