Hi! Hope you doing well 
I’m working on building a graph autoencoder capable of generating embeddings for graphs of arbitrary size. Most of the literature I’ve read focuses on fixed-size node graphs, which doesn’t quite meet my requirements. The only relevant work I found is “Learning Graphon Autoencoders for Generative Graphs”, but I couldn’t find any implementations of their proposed model.
The encoding part seems relatively straightforward—you can design it to output a fixed-size embedding regardless of the graph’s size. However, the decoding part is much trickier: How would you design a decoder to handle graphs of variable sizes? Does this idea even make sense in practical terms? It seems complex, but such a model could be incredibly useful.
I’d appreciate any insights, references, or advice on this!
Thanks in advance!