Author: Shadi Zabad, McGill University
This is a report on my course project for COMP766, taught by Will Hamilton.
In this paper, we propose a graph alignment approach based on representation learning with modern Graph Neural Network (GNN) architectures. Our model makes use of a Siamese architecture to learn a similarity function between the embeddings of nodes in different graphs. To make the model more robust to structural and attribute noise, we propose the idea of "anchored" node embeddings that can be combined with any GNN architecture. We show that this model has competitive performance on both synthetic and real world datasets.