Graph Canonization

Graph canonization is the process of transforming graphs into a unique, canonical representation, regardless of their initial labeling or structure. Current research focuses on leveraging canonization to improve the expressivity and stability of graph neural networks (GNNs) for graph representation learning, as well as enhancing the reliability of explainable AI (XAI) methods for deep neural networks by addressing implementation-dependent issues. This involves developing novel algorithms and frameworks for efficient canonization, particularly for complex architectures like DenseNets and Relation Networks, and analyzing the trade-offs between expressivity and stability in canonization-enhanced GNNs. The ultimate goal is to improve the accuracy, robustness, and interpretability of machine learning models applied to graph-structured data and other complex network architectures.