Neuro-Symbolic Reasoning with Graph Grammars
Year: 2026
Type: preprint
Abstract: This paper proposes a novel approach to neuro-symbolic reasoning by leveraging graph grammars. The core challenge in combining neural networks and symbolic reasoning lies in the absence of a shared representation. This work addresses this issue by employing graph grammars to formally represent logical rules and constraints, providing a structured and unambiguous framework. A neural network is then trained to predict the next step in a graph grammar derivation, effectively enabling the network to learn and execute symbolic reasoning processes. The key innovation resides in the graph-based representation, which serves as a bridge between the continuous nature of neural networks and the discrete logic of symbolic systems. This framework allows for the efficient and interpretable representation of complex reasoning tasks, potentially leading to significant advancements in areas such as knowledge representation, automated theorem proving, and intelligent agent design. The system is designed to learn from examples, iteratively refining its ability to generate valid derivations based on the defined graph grammar rules. Ultimately, we aim to achieve a robust and adaptable reasoning system that benefits from both the learning capabilities of neural networks and the deductive power of symbolic logic.
Author: Zhang, Jincheng
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7210279273
