Self-Supervised Learning of Graph Grammars from Unlabeled Data
Year: 2026
Type: preprint
Abstract: Graph grammars provide a formal and expressive means to describe and generate graph structures, finding applications in diverse domains such as computer graphics, digital modeling, and data visualization. However, their conventional development relies heavily on manual rule crafting, a labor-intensive and often domain-specific process. This research addresses this limitation by proposing a novel self-supervised learning framework for automatically discovering graph grammar rules from unlabeled graph data. The core of the approach involves learning embeddings of graphs that capture their structural properties. These embeddings are then leveraged to train a generative model that learns to infer rules for constructing valid graph structures. The algorithm is evaluated on synthetic graph datasets, demonstrating its ability to identify significant structural patterns and generate rules that accurately reflect these patterns. The key contribution lies in the automation of graph grammar creation, opening up new possibilities for applying graph grammars to complex, unlabeled graph datasets where manual rule definition is impractical. The system's capacity to learn from raw data without human intervention represents a significant step towards democratizing the use of graph grammars.
Author: Zhang, Jincheng
Language: English
Topic: Graph Theory and Algorithms
Subfield: Computational Theory and Mathematics
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7206183889
