Deep Homomorphism Networks Enhance Graph Learning Over Relational Databases

A recent paper published on arXiv proposes a novel framework known as Deep Homomorphism Networks (DHNs) to enhance the interpretation of relational database data. The study highlights the strong correlations between DHNs and key SQL elements, particularly in the context of conjunctive queries. Researchers analyze the expressive capabilities of DHNs, linking them to various natural fragments and logical extensions within first-order logic. Additionally, they establish connections between DHNs employing aggregations like max, sum, and mean, and specific fragments of unary negation and quantifiers, thereby deepening insights into the relationship between DHNs and SQL query structures.