Augment to Interpret: Unsupervised and Inherently Interpretable Graph Embeddings
In this paper, we study graph representation learning and show that data augmentation that preserves semantics can be learned and used to produce interpretations. Our framework, which we named INGENIOUS,