Author: Rostyslav Demush

Colossal hype ensues whenever there's progress or even the slightest shift in the world of machine learning (ML). When it comes to AI, the craze always ends up growing out of proportion and insane claims invariably follow. You've probably noticed this recently as more people are beginning to use Midjourney, ChatGPT, Copilot, and so on.

The truth about these systems, however, is always more prosaic. Machine learning is mathematical, and the real implications of the ML models are not nearly as profound as some bloggers would have you believe. They could be beneficial and, in some cases, transform large portions of workflows in specific areas, but only if the user, whether an organization or an individual, has a sufficient understanding of their inner workings, limitations, capabilities, and potential.

After CNNs exploded in 2012, showing unprecedented levels of prediction accuracy on image classification tasks, a group of researchers from Yann LeCun's team decided to extend their success to other, more exotic domains. Specifically, they started working on generalizing convnets to graphs. Their efforts were described in this influential paper. 

Since then, Graph Neural Networks have become a hot area of research within the ML community and beyond. Numerous papers have been published explaining how different kinds and flavors of GNNs can be applied to complicated, irregular, high-dimensional, non-grid-like structures (graphs, manifolds, meshes, etc.).