Happy to announce a new paper "Discrete Graph Auto-Encoder"

Happy  to announce a new paper "Discrete Graph Auto-Encoder" published in Transactions on Machine Learning Research.

The paper introduces a novel generative model for graphs, leveraging a graph-to-set discrete auto-encoder invariant to permutations. It develops a new 2D-Transformer specifically designed to model sequences across two dimensions, and designs a new technique of p-path features for graph feature augmentation.
The paper is a result of successful collaboration of Magda Gregorova with the DMML group in Geneva, Swizterland.

Congrats to Yoann Boget for the great work as the first author and thanks to Alexansros Kalousis for enabling the collaboration.