Neural-Symbolic Logic Query Answering in Non-Euclidean Space

Researcher Lihui Liu has introduced HYQNET, a novel AI model designed to answer complex first-order logic (FOL) queries on knowledge graphs. The core innovation is its use of hyperbolic geometry—a type of non-Euclidean space—to embed both the knowledge graph and the logical query structure. This approach is mathematically better suited for representing the hierarchical and tree-like nature of logical reasoning than the flat, traditional Euclidean space used by most AI models. HYQNET operates as a neural-symbolic system, decomposing FOL queries into relation projections and logical operations over fuzzy sets for transparency, while using a hyperbolic graph neural network (GNN) to infer missing links in incomplete knowledge graphs.

This hybrid architecture aims to solve a persistent trade-off in AI reasoning: symbolic methods are interpretable but brittle with incomplete data, while neural methods are robust but opaque. By performing all reasoning in hyperbolic space, HYQNET more effectively preserves the structural dependencies within a recursive query. The paper reports that experiments on three standard benchmark datasets demonstrate HYQNET's strong performance, highlighting the tangible advantages of moving beyond Euclidean embeddings for this class of problem. This work points toward more reliable and interpretable AI systems for complex tasks like scientific discovery, advanced search, and enterprise knowledge management, where reasoning over interconnected facts is crucial.