Related Articles

• Feature Extraction for Graphs
• Machine Learning Tasks on Graphs
• Towards Explainable Graph Neural Networks
• Top 10 Learning Resource for Graph Neural Networks

Welcome to the world of graph neural networks where we construct deep learning models on graphs. You could think that is quite simple. After all, can’t we just reuse models that work with normal data?

Well, not really. In the graph, all datapoints (nodes) are interconnected with each other. This means that data is not independent anymore which makes most of standard machine learning models useless as their derivations strongly base on this assumption. To overcome this problem, it is possible to extract numerical data from graphs or use models that directly operate on this type of data.

Creating models that directly work on graphs is more desirable because we obtain more information about their structure and properties. In this article, we will look at one of the architectures specifically designed for this type of data, Message Passing Neural Networks (MPNNs).

#mpnn #deep-learning #cheminformatics #graph-neural-networks