The Graph autoencoder works and even though it is not yet perfect there migth be applications where having an autoencoder for graphs is useful Especcially since they do not need many datapoints and seem to never overfit As anomaly detector it can be as good as any other code And using creative ideas, we could make it much better See grapa.readthedocs.io You can use my code for other things There are some example usecases in the documentation