Learning Graph Grammars from Observational Data
Year: 2026
Type: preprint
Abstract: This paper proposes a novel approach to graph grammar learning, leveraging recurrent neural networks (RNNs) to automatically generate graph structures from observational data. Traditional graph grammars are often manually defined, limiting their applicability to complex systems with poorly understood rules. Our methodology, termed "Observational Graph Grammar Learning" (OGGL), aims to overcome this limitation by learning the grammar directly from data. We hypothesize that a sufficiently complex observational dataset can encode the underlying rules governing the formation of graph structures. The core of OGGL involves training an RNN to predict the next step in a graph grammar based on the current state of the graph, effectively learning the grammar's production rules. This approach offers a flexible and adaptive method for modeling systems where explicit grammatical definitions are unavailable or impractical. We demonstrate the feasibility and potential of this technique, highlighting its capacity to generate graphs representing intricate relationships observed in various domains. The proposed method addresses the key challenges in graph grammar learning by providing a data-driven solution that can learn complex rules without human intervention.
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: W7208818857
