Article rants that Graph based Neural Network architectures [ most of all Machine Learning methods currently in use ] are discreet math [ meaning Matrix operations ] instead of continuous math [ like Calculus ] and therefore are computationally heavy.
Which is why we aren't getting any better results from the big money and research being poured into it for the last 5 years.
Need to look elsewhere.
.
https://www.singlelunch.com/2020/12/28/why-im-lukewarm-on-graph-neural-networks/